PDF hosted at the Radboud Repository of the Radboud University Nijmegen

Transcription

1 PDF hosed a he Radboud Reposiory of he Radboud Universiy Nijmegen The following full ex is a publisher's version. For addiional informaion abou his publicaion click his link. hp://hdl.handle.ne/2066/ Please be advised ha his informaion was generaed on and may be subjec o change.

2 Towards Housing Sysem Dynamics Projecs on embedding sysem dynamics in housing policy research Marijn Eskinasi

3

4 Towards Housing Sysem Dynamics Projecs on embedding sysem dynamics in housing policy research

5

6 Towards Housing Sysem Dynamics Projecs on embedding sysem dynamics in housing policy research Proefschrif er verkrijging van de graad van docor aan de Radboud Universiei Nijmegen op gezag van de recor magnificus prof. mr. S.C.J.J. Kormann, volgens beslui van he college van decanen in he openbaar e verdedigen op donderdag 28 augusus 2014 om uur precies door Marijn Eskinasi geboren op 18 sepember 1970 e Amserdam

7 Promoor(en): Prof. dr. J.A.M. Vennix Prof. dr. J.B.S. Conijn (Universiei van Amserdam) Copromoor(en): Dr. E.A.J.A. Rouwee Manuscripcommissie: Dr. C.E. van Daalen (Technische Universiei Delf) Prof. dr. ir. M.G. Elsinga (Technische Universiei Delf) Prof. dr. E. van der Krabben Eburon Academic Publishers, Delf

8 Preface The approximae fifeenyear ime rajecory of he making of his PhD hesis can be adequaely described wih sysem dynamics. In he firs foureen years, a reinforcing loop was dominan. When my fixes o a daacongesed model failed, when I properly learn sysem dynamics a Nijmegen Universiy and acquired he ase in he Haaglanden projec, progress was presen hough no very noiceable. Laer on, wih Houdini, Middle Incomes and he Morgage model in full swing, progress was seep and visible. Bu no real world sysem conains only reinforcing feedback. In he fifeenh year I also suffered from ha balancing feedback mos PhD sudens encouner in he final sage: he sock of new ideas becomes depleed, he odo lis apparenly keeps growing and he mind definiely needs some res, bu says relenlessly occupied wih he hesis. All good hings graefully received in life are hreefold in naure: suppor, inspiraion and pracical arrangemens. Graefulness for heir loving and lasing suppor belongs especially o he mos imporan women in my life: my wife Zuzana, my moher Ineke and my daughers Charloe and Jusine. Also many friends, relaives and colleagues helped me o keep i up and wo of hem, Jörgen van de Langkruis and Keshen Mahura, are my defense assisans oday. All of you deserve my warmes love and friendship! Graefulness for inspiraion belongs o all eachers ha menored me o my curren sandpoin, especially my hesis (co)supervisors Jac Vennix, Johan Conijn and Eienne Rouwee. I am also indebed o hose ha augh and rained me in maers of personal energy, persisence and hough power. All of you deserve my sinceres respec! Graefulness belongs o all who conribued pracical arrangemens o his success, especially o Eppie Fokkema of Arivé for sparking he academic desire and providing he opporuniy o learn and o Dorien Maning of PBL Neherlands Environmenal Assessmen Agency for providing he working ime for he final long srech. All of you deserve my deepes graiude! And finally I wish ha all scienific research conribues o he wellbeing of mankind and plane earh. So be i!

9 Summary The purpose of his PhD hesis is o conribue o a sysemaic connecion beween housing research and sysem dynamics. Housing research is a vas field focusing on housing markes, residenial behavior, indusrial organizaion and governmen inervenion. Sysem dynamics is a mehod for learning abou dynamic complexiy of social sysems wih a srong emphasis on compuer simulaion. These fields share several common characerisics, bu here is no sysemaic cooperaion ye. Given his sae of affairs, he hesis mus lay some groundwork by means of exploraory research and case sudies applying sysem dynamics o housing research issues. The hesis seeks o answer six research quesions. These concern lieraure research ino 1) conemporary housing research issues suiable for applying sysem dynamics 2) causes for he lack of and recommendaions for improving sysemaic cooperaion 3) he accumulaed knowledge base of sysem dynamics on housing markes and 4) sysemaic analysis of his base for he purpose of improving cooperaion. The case sudies encompass sysem dynamics projecs in close cooperaion wih housing researchers and seek o define he added value of such projecs 5) relaed o housing conen and 6) srucural cooperaion beween disciplines. The hesis explores conemporary housing research lieraure for he presence of complexiyrelaed issues. These were found in he special characerisics of housing, he differen ime frames of housing marke processes, he need o deal wih nonequilibrium siuaions, he dynamics of household choice and demographics, he complex srucure of he housing supply marke and he presence of insiuional and policy feedback loops. I furher clarifies he sysem dynamics perspecive and mehod and illusraes i wih wo examples. The main causes of he isolaed posiion of sysem dynamics were found in he endency o specialize in mehod raher han conen and in hisorical debaes beween sysem dynamics and radiional economics. Small, comprehensible models in he language and conceps of he field of applicaion are generally conducive o cooperaion. Sysem dynamics lieraure on housing encompasses over 150 enries, ranging from groundbreaking publicaions o average conference papers. A firs group revolves around he 1969 cornersone projec Urban Dynamics and is sill producive. A second group focuses on changing governmen policies exiss in he Neherland. A hird group emerged afer he 2008 financial crisis and connecs sysem dynamics o mainsream real esae economics lieraure. Finally, several isolaed enries were caalogued. Four case sudies were carried ou for his hesis. A Group Model Building projec in he Haaglanden region helped regional policy makers o sele a policy conflic and o improve undersanding of he dynamics of he regional housing marke. The second case sudy repors he building of Houdini, a sysem dynamics model based on mainsream real esae economics wih several added insiuional feaures, like zoning, residual land pricing, fiscal morgage suppor and ren regulaion. The hird projec named Middle Incomes focused on making an impac analysis of a much conesed new regulaion on

10 Summary 7 he accessibiliy of he social renal secor. The fourh and final case sudy is concerned wih a model of he dynamics of he naional morgage deb. Nex o he main answers o he research quesions summarized above, he findings of he hesis encompass a se of building blocks or modeling ideas for furher applicaion in housing research. Conenrelaed conclusions suppor exising ideas ha housing allocaion sysems are relaively weak in simulaing housing vacancy chains, ha demographic dynamics are a predominan force, ha policy acors end o underesimae he impac of ime delays and ha generic housing marke srucures can display widely varying ime rajecories under differen (regional) parameer ses. Quesions for furher research focus on idenifying alernaive and addiional building blocks, rigorous simulaion, closer comparison of exising empirical findings and sysem dynamics simulaions and he exac demarcaion of sysem dynamics and oher simulaion mehods. As o he cooperaion beween housing researchers and sysem dynamics, i is proposed ha he accepance of sysem dynamics in social sciences is isomorphic o validaion on he projec level: i is a process of gradual confidence building. Embedding sysem dynamics in regular research projecs means o be selecive in applying sysem dynamics o he proper issues and in crossexamining model oucomes wih oher ypes of research. I requires ha sysem dynamics praciioners deeply undersand he conen issues and know where o make small, comprehensible sysem dynamics models excel. The case sudies furhermore indicae ha properly framing and communicaing he scope, purpose and limiaions of models conribue o successful projecs. The hesis is concluded wih a dynamic hypohesis how embedded sysem dynamics may conribue o close cooperaion and inegraion of he mehod ino regular social science.

11 Samenvaing Naar Housing Sysem Dynamics 1 Projecen rond de inbedding van sysem dynamics in woningmarkbeleidsonderzoek Doel van di proefschrif is bij e dragen aan een sysemaische verbinding van woningmarkonderzoek en sysem dynamics. Woningmarkonderzoek is een breed gebied me onder andere hema s als woningmarken, woonvoorkeuren en verhuisgedrag, de woningbouwsecor, overheidsinervenie. Sysem dynamics is een op compuersimulaie gebaseerde mehodiek om he inzich in dynamisch gedrag van complexe sociale sysemen e vergroing. De disciplines kennen de nodige overeenkomsen, maar sysemaische samenwerking is er eigenlijk nog nie. Daarom dien di proefschrif een eerse basis e leggen via verkennend onderzoek en case sudies. Er saan zes vragen cenraal over 1) onderzoeksvragen geschik voor sysem dynamics, 2) oorzaken van he gebrek aan samenwerking en besaande aanbevelingen voor samenwerking, 3) de o nu oe opgebouwde sysem dynamicskennis over woningmarken 4) een sysemaische analyse van deze kennis me beere samen werking en doel. De case sudies zijn oepassingen van sysem dynamics in samenwerking me woningmarkonderzoekers me als doel de oegevoegde waarde in beeld e brengen bereffende 5) de inhoud en 6) de samenwerking. He proefschrif verken recene woningmarklierauur op de aanwezigheid van onderzoeksvragen waar complexiei een rol speel. Deze zijn e vinden in de specifieke eigenschappen van woningen, verschillen in ijdshorizon van diverse woningmarkprocessen, de noodzaak om ook sysemen buien evenwich e onderzoeken, dynamiek van woonvoorkeuren, keuzeprocessen en demografie, srucuur van de woningbouwkeen en de invloed van overheidsbeleid en insiuies. Belangrijke oorzaken van de geïsoleerde posiie van sysem dynamics zijn mehodische specialisaie en hisorische discussies ussen de sysem dynamics wereld en radiionele economen. Kleine simulaiemodellen in de aal van he oepassingsgebied zijn aan de andere kan vaak zeer onderseunend aan samenwerking. De sysem dynamicslierauur over woningmarken el ruim 150 bijdragen, variërend van klassiekers o confereniepapers. Een eerse groep bouw voor op he Urban Dynamics model ui 1969 en lever nog seeds nieuwe bijdragen op. Een weede groep leg de nadruk op de beleidswijzigingen in Nederland. De derde groep is onsaan na de krediecrisis ui 2008 en maak meer gebruik van vasgoedeconomische lierauur. To slo zijn ook alle losse bijdragen vasgelegd. Vier case sudies vormen he empirische deel van di proefschrif. Een Group Model Building projec in Haaglanden heef beleidsmakers geholpen een beleidsconflic op e lossen en meer inzich in de dynamiek van de regionale woningmark e krijgen. De weede case sudy beref de onwikkeling van Houdini, een sysem dynamics model 1 Housing sysems en sysem dynamics zijn gebruikelijke begrippen in de Engelse vakaal. Housing sysem dynamics als samengeseld begrip is een nie goed in he Nederlands e veralen woordgrapje.

12 Samenvaing 9 gebaseerd op vasgoedeconomische lierauur me oevoeging van diverse insiuionele elemenen als ruimelijke ordening, residuele grondprijzen, huurregulering en hypoheekreneafrek. He derde projec Middeninkomens omva een impacanalyse van een omsreden nieuwe regeling over de oegankelijkheid van de sociale huursecor. De vierde en laase case sudy beref een model over de naionale hypoheekschuld. Naas de hierboven samengevae anwoorden op de zes onderzoeksvragen rapporeer di proefschrif bevindingen in de vorm van bouwsenen voor sysem dynamics modellen van woningmarken. Inhoudelijke conclusies bevesigen he belang van demografie voor de woningmark, da woningoewijzing nauwelijks effec heef op de doorsroming, da beleidsmakers he effec van verragingen onderschaen en da generieke woningmarksrucuren onder verschillende (regionale parameers) wijd uieenlopende ijdspaden van cenrale variabelen kunnen veronen. Vragen voor verder onderzoek bereffen onder meer he onwikkelen van alernaieve en aanvullende bouwsenen, grondige gevoeligheidsanalyses van de gepreseneerde modellen, meer vergelijking ussen empirische bevindingen en modeluikomsen en de afbakening van sysem dynamics en andere simulaiemehoden. Ten aanzien van samenwerking ussen woningmarkonderzoekers en sysem dynamiciss word geseld da bredere accepaie van sysem dynamics isomorf is aan de validaie op projecniveau: er is sprake van een geleidelijke opbouw van verrouwen in de mehodiek c.q. he model. Verdere inbedding in onderzoeksprojecen vereis een selecieve inze van sysem dynamics op de juise vraagsukken en voldoende kruisconrole van de simulaieresulaen me andere mehoden en bronnen. De berokken sysem dynamiciss dienen diep genoeg in de inhoud e zien om kleine, begrijpelijke modellen e maken op relevane onderzoeksvragen die anders nie of moeilijk e beanwoorden zijn. De case sudies onen ook de noodzaak om scope, doel en beperkingen van de modellen duidelijk e communiceren. He proefschrif word besloen me een dynamische hypohese hoe embedded sysem dynamics kan bijdragen aan nauwere samenwerking en accepaie van de mehode binnen de sociale weenschappen.

13

14 Conens Preface 5 Summary 6 Samenvaing 8 I Inroducion 13 I.1 Purpose of he hesis and inroducion o he research heme 13 I.2 Research quesions, mehodology and relevance 15 I.3 Srucure of he hesis 17 II Housing research issues and sysem dynamics 21 II.1 Conemporary research issues in housing sudies 21 II.2 The sysem dynamics perspecive and mehod 26 II.3 Sysem dynamics in isolaion 40 II.4 Conclusions 42 III Lieraure review of sysem dynamics on housing 45 III.1 Overall remarks and descripive saisics 45 III.2 Urban Dynamics Group 46 III.3 Duch Housing Policy Group 51 III.4 Real Esae Dynamics Group 52 III.5 Isolaed sudies 57 III.6 Conclusions 57 IV Haaglanden 59 IV.1 Inroducion 59 IV.2 Conex of he sysem dynamics inervenion 59 IV.3 The sysem dynamics inervenion 63 IV.4 The resuling model 65 IV.5 Validaion ess 71 IV.6 Base run and policy experimens 74 IV.7 Evaluaion of he projec 76 IV.8 Conclusions 78 V Houdini 79 V.1 Inroducion 79 V.2 Conex of he sysem dynamics modeling projec 79 V.3 The sysem dynamics modeling projec 80 V.4 The resuling model 81 V.5 Validaion ess 86 V.6 Base run and policy experimens 88 V.7 Follow up aciviies and reacions o Houdini 90

15 12 Conens V.8 Evaluaion of he projec 91 V.9 Conclusion and discussion 91 VI Middle Incomes 93 VI.1 Inroducion 93 VI.2 Conex of he sysem dynamics inervenion 93 VI.3 The sysem dynamics modeling projec 95 VI.4 The resuling model 96 VI.5 Validaion 101 VI.6 Base run and policy alernaives 102 VI.7 Followup aciviies 106 VI.8 Evaluaion of he projec 107 VI.9 Conclusions 108 VII Morgage Model 109 VII.1 Inroducion 109 VII.2 Conex of he sysem dynamics inervenion 109 VII.3 The sysem dynamics modeling projec 110 VII.4 The resuling model 111 VII.5 Validaion 113 VII.6 Base run and policy alernaives 115 VII.7 Evaluaion of he projec 116 VII.8 Conclusions 117 VIII Conclusions, discussion and quesions for furher research 119 VIII.1 Review and main research conclusions 119 VIII.2 Insigh for housing sysem dynamics modeling 122 VIII.3 Insighs on embedded sysem dynamics 126 VIII.4 Epilogue: a dynamic hypohesis for embedded sysem dynamics 130 Appendices, liss and references 133 Appendix 1 Commonly used variables in model repors 133 Appendix 2 Model and simulaion repor for he model in II Appendix 3 Model and simulaion repor for Haaglanden model 137 Appendix 4 Model and simulaion repor for Houdini model 140 Appendix 5 Model and simulaion repor for Middle Incomes model 145 Appendix 6 Model and simulaion repor for Morgage model 151 Lis of Tables 154 Lis of Figures 154 References 156

16 I Inroducion I.1 Purpose of he hesis and inroducion o he research heme Many conemporary housing research issues could fruifully benefi from he use of he sysem dynamics mehod. Sysem dynamics, however, operaes largely in isolaion of oher social sciences. The purpose of his PhD hesis is herefore o conribue o a sysemaic connecion beween housing research and sysem dynamics. Housing research 2 sudies a vas array of phenomena like residenial mobiliy, neighborhood developmen, he working of housing markes and he ineracion wih he overall economy, he relaion beween household demographics and residenial paerns, socioeconomic issues like affordabiliy, social housing managemen, povery and segregaion, housing consrucion, urban design, susainable building and more. Housing research is mulidisciplinary and draws, among ohers, from differen srands of economics, from sociology, human geography, gender and developmen sudies and from poliical science. Housing research is a mixedmehod discipline, using sais ical and economeric echniques, qualiaive mehods, large scale surveys, demographic forecasing and oher modeling echniques. Housing research is in many cases relaed o housing policy making, as he provision of housing conains boh marke mechanisms and public policy inervenions in mos Wesern counries. As housing, housing markes and housing policy are exremely mulifaceed, references o housing as a complex issue are omnipresen. Such housing marke complexiies sem mosly from he paricular properies of housing. Houses represen many characerisics, some relaed o physical properies (e.g. size, number of rooms, ameniies, qualiy), some relaed o viciniy of services, ranspor, work locaions and areas for recreaion and some relaed o social issues like neighborhood qualiy, safey and he like (Gibb, 2012). The housing supply secor is fragmened over differen ypes of acors like land developers, propery developers and conracors (Ball 2013), all reacing on marke impulses, governmen decisions and inernal risk/ feasibiliy consideraions. Behavior of households owards housing and residenial mobiliy highly depends on households characerisics such as age, household composiion, income, educaion and culure. Housing has also been subjec o governmen inervenion ever since medieval aldermen sared o inervene for prevening caasrophic ciy fires. Regulaion regards consrucion, land use planning, affordabiliy issues and ohers. These governmen measures inerac wih he oher processes on he housing marke and add o he complexiy of is behavior. Sysem dynamics is a mehod o enhance learning abou dynamic behavior of such complex sysems and developing more effecive policies for influencing hem. I helps undersanding and managing complex sysems by using compuer simulaion models 2 Throughou his hesis, he erms housing research and housing sudies will be used a synonyms.

17 14 chaper I as managemen fligh simulaors, jus as aviaion uses simulaion for raining pilos and air raffic conrollers (Serman, 2000, pp. 45). Sysem dynamics is solidly grounded in heories of feedback and nonlinear dynamics iniially developed in mahemaics and engineering. I applies hese ideas o social sysems using insighs from human sciences like psychology, economics, housing research, ecology, medical science ec. Mos noably, since is incepion, sysem dynamics has also been sudying housing and urban developmen. Cenral o sysem dynamics is he feedback perspecive. Sysem dynamics emerged in he 1950 s from operaions research, which aimed a supporing managemen decision making by means of mahemaical and saisical analysis. Operaions research, however, proved o be ineffecive for solving broad, sraegic quesions, because of is openloop approach where no feedback exiss beween he sysem o be influenced and he decision o be made. Founding faher Jay Forreser proposed a closedloop approach as an alernaive: decisions are made on basis of informaion on he sae of he sysem o be influenced. Changes in he sysem, brough abou by hese decisions, hen influence fuure decisions. In oher words: decision making for influencing a social sysem is inrinsically a par of he sysem. Causes and effecs are no linear bu circular: here is muual feedback beween sysem and decision (Vennix, 1996, p. 43). Sysem dynamics is a mehod o enhance learning abou behavior of and improving policies and decisions wihin complex sysems. I does so by building compuer simulaions of he complex sysem involved, simulaing proposed and alernaive decisions, confroning decision makers wih he oucomes and helping hem undersand why inended and uninended consequences emerge from he sysem srucure. Sysem dynamics relies on compuer modeling as is main mehodology, bu perceives compuer models only as imperfec mahemaical represenaions of imperfec human menal models of realworld sysems. Therefore, models are mere ools for incremenal improvemen in undersanding a paricular dynamic problem. All models are subjec o limiaions in scope of use, deail, boundaries, conex ec. Pu aphorisically: all models are wrong, bu some models are useful. Sysem dynamics models are useful when hey help acors o beer undersand he sysem hey are dealing wih. Judging from he above, housing research and sysem dynamics share many aspecs in order o make close cooperaion muually beneficial. Housing markes or housing sysems consis of many paries ineracing wih one anoher on he basis of informaion sreams. They are exacly he complex social feedback sysems sudied by sysem dynamics. Some housing processes involve long ime delays and moreover, socks and flows are common concepual elemens in boh disciplines. Finally, sysem dynamics is srongly focused on devising beer policies hrough beer undersanding of feedback processes. Bu surprisingly, here is no sysemaic cooperaion beween boh disciplines ye. Housing research ges by wih oher mehodologies, even if some conemporary research issues could benefi, a leas poenially, from he sysem dynamics approach. Sysem dynamics ges by in relaive isolaion from oher social sciences, bu hriving in managerial applicaions and ecology and wih a scaered bu no unsubsanial knowledge base in he field of housing and real esae, largely unnoiced by he housing research and policy communiy. Hence he purpose of his hesis.

18 Inroducion 15 I.2 Research quesions, mehodology and relevance Bu if here is only an inuiively sensed poenial for applying sysem dynamics in housing research sysemaically, we mus conclude ha he errain of sysemaic connecion beween he wo disciplines is virually erra incognia. This conclusion defines he saring poin for he research quesions. Research quesions Graned he above conclusion, research ino a sysemaic connecion mus sar a he bare basics. Firs, we mus idenify clearly for wha issues in housing research sysem dynamics offers he mos fruiful perspecive. We should also be aware ha boh disciplines have coexised in virual isolaion of one anoher for nearly half a cenury. We mus herefore undersand he causes of his counerinuiive siuaion. The firs wo research quesions revolve around hese issues: 1. Which conemporary research issues in housing sudies are paricularly fi for ackling wih sysem dynamics? 2. Wha facors have conribued o he lack of sysemaic cooperaion beween housing research and sysem dynamics up o he presen? Wha pracices and recommendaions are presen in exising lieraure for improving cooperaion? As menioned above, sysem dynamics does have a cerain rack record in our field of ineres. We mus explore he exising sysem dynamics lieraure base on housing, real esae and urban developmen, which unforunaely is available only in a fragmened way. This lieraure base needs iniial caaloguing of books, journal aricles, conference presenaions and oher sources, in order o provide oversigh for fuure research. However basic, his is a fundamenal firs sep. Furhermore, we mus endeavor ino an iniial aemp o inegrae and sysemaize he insighs from his lieraure in such a fashion ha hey become useful for he purpose of his hesis, i.e. by aking ino accoun he relevan housing research issues and he lessons and recommendaions for improving sysemaic cooperaion. This main ask is worded in he following research quesions: 3. Wha is he accumulaed knowledge of sysem dynamics on housing relaed issues up o now? 4. How can i be sysemaized and inegraed ino a form ha is supporive of he research purpose of his sudy? Anoher se of research quesions connecs o he projecs menioned in he subile of his hesis. If he sysem dynamics projec on housing have been conduced mosly in isolaion from mainsream housing research, i is necessary o explore he use of sysem dynamics in a housing policy research conex. This will help generae model conen closely linked o mainsream housing research. I will also add experience in cooperaion beween boh disciplines. 5. Wha sysem dynamics models can be buil in close connecion o mainsream housing research? Wha is heir added value o he exising knowledge base of boh sysem dynamics and housing research regarding conen? 6. Wha lessons can be learn from he model building experiences in research quesion 5 abou fruiful cooperaion beween sysem dynamiciss and housing researchers?

19 16 chaper I Research mehodology Terrae incogniae require discovery in he firs place. Exploraory research is primarily concerned wih such discovery and wih generaing insighs and/or building heories (Davies, 2006, p. 110). Confirmaory research on he oher hand is focused on heory verificaion hrough horough, rigorous hypoheses esing on basis of solid welldefined (saisical) procedures. However, i necessarily assumes he apriori exisence of heories. Exploraory research precedes he sage of heory esing and is involved in he acual developmen of heory on he basis of unrelaed and scaered daa or observaions of he real world. Exploraory research is broad and horough in is own paricular sense and requires flexibiliy and pragmaism 3 raher han solid (saisical or deducive) rigor. Exploraory and confirmaory represen differen bu indissolubly conneced phases of scienific endeavor, like yin and yang in Chinese philosophy. The limiaions of exploraory research mean ha no definiive answers will be provided o he issues above. The mos suiable mehodology for research quesions 1 o 4 is lieraure sudy. I should concenrae on finding hose conemporary challenges in housing research ha bes mach he niche of sysem dynamics. Nex o lessons and insighs on cooperaion beween social science in general and sysem dynamics, a main secion of his work consiss of lieraure research on he exising sysem dynamics knowledge base on housing relaed issues. Where necessary, lieraure research will be complemened wih addiional echniques such as causal loop diagramming and sysem archeype analysis. Research quesion 5 and 6 requires he use of several mehodologies. Firs, all mehodologies for building sysem dynamics models in cooperaion wih housing researchers are needed. Sysem dynamics modeling and relaed echniques will presumably play an imporan role, bu should no in ligh of he purpose of conribuing o sysemaic connecions a priori be aken as he dominan or only mehodology o he exclusion of ohers. The modeling projecs mus fulfill some basic requiremens: 1. Obviously, hey cover housing relaed hemes and use conceps found in mainsream housing sudies. 2. They should adhere o sandards and guidelines of properly conduced sysem dynamics projecs. 3. Housing expers and/or researchers paricipae in hese projecs. I is also necessary o have some reference or sandard for measuring he success or impac of he sysem dynamics projecs, as he basic requiremens only es wheher hey were conduced properly. High sandards for qualiy were se by Forreser (2007b). The founding faher is criical of he sae of affairs and claims ha sysem dynamics is a a raher aimless plaeau (p. 350), ha i lacks proper impac on governmen due o is inabiliy o find new high leverage policies for addressing he big issues in sociey. 3 Many anecdoes circulae abou he exploraory research mehods of hisory s greaes scieniss. Archimedes used bahing echniques, Newon slep under an apple ree, Einsein conemplaed acceleraing elevaors in space when bored wih his dayime job. This led science philosopher Feyerabend o provocaively sugges ha wellesablished mehods can even obsruc scienific progress. Oher mehodologiss like Kuhn and Lakaos propagae more moderae sances where scienific heories and relaed mehods evolve in schools of hough and are being replaced in innovaive burss of scienific revoluions.

20 Inroducion 17 He se nine crieria ha help unfold he full poenial of sysem dynamics. He claims ha mos works fall shor of hese sandards because mos praciioners have no opporuniy of receiving sysem dynamics raining beyond a basic level. He opposes endencies o simplify sysem dynamics as i will dilue is powerful poenial. The crieria include idenificaion of he problems in he real world sysem, a compac dynamic hypohesis (or model) wih srong generic and endogenous properies leading o new, differen defendable policy opions, including a discourse on expeced resisance and how o overcome i. These sandards will be used for assessing he qualiy of work presened here. In he overall framework of his hesis, he use of hese modeling projecs is a form of case sudy research. Case sudy research allows he invesigaors o reain he holisic and meaningful characerisics of reallife evens (Yin, 2003, p. 2), which is very appropriae given he open characer of research quesions 5 and 6. Case sudy research is generally seen as suiable for exploraory research in siuaions where conrol over behavioral evens is limied or absen. As opposed o conrolled experimens wih many subjecs, modeling projecs require a large endeavor and are no easily replicaed. Relevance This research hesis is relevan for science because of is perspecive o connec wo previously unrelaed subjecs. The lieraure review in he firs main ask will make sysem dynamics insighs on housing, urban developmen and real esae accessible o a wider audience. The pilo projecs will in any case conribue o he knowledge base of sysem dynamics, bu will also generae relevan insighs for he researchers and oher sakeholders involved in hem and conribue o a posiive image of sysem dynamics among housing researchers. In he ideal case, hey will be an iniial sepping sone for fuure breakhrough research in housing sudies, bu i is beyond he scope of his hesis o judge wheher hese high hopes are realisic. The relevance for sociey lies in he imporan role of housing in he overall economy and he recen economic collapse. In rerospec, he grea financial crisis of 2008 is he dynamic behavior of a complex feedback sysem encompassing he financial marke, he housing marke, governmen budges and he overall economy. This complex sysem displayed a rapid shif in loop dominance from growh hrough overshoo ino collapse. This sudy focuses on connecing such a paramoun aspec of human life and wellbeing i.e. housing wih a mehod poenially capable of improving insigh ino he dynamics of complex socioeconomic sysems as he housing marke. The sysem dynamics mehod helps human acors o improve heir undersanding and policies owards such sysems. Therefore, even if his sudy conribues only small specks of improved undersanding and beer or less derimenal policies, i holds relevance for sociey. I.3 Srucure of he hesis Chapers II and III cover he lieraure research necessary for fulfilling he purpose of he hesis. Chaper II firs relaes o research quesion 1. I idenifies hose housing research issues suiable for sysem dynamics modeling and provides a more elaborae inroducion o he sysem dynamics mehod, however, wihou being a full uorial. Several excellen

21 18 chaper I exbooks are available for learning sysem dynamics, e.g. Fisher (2004), Serman (2000) and Vennix (1996). Chaper II also covers research quesion 2: i explains hisorical and concepual causes of he isolaed posiion of sysem dynamics and summarizes exising recommendaions on cooperaion wih social scieniss. Chaper III answers research quesions 3 and 4. I sars wih overall merics of he sysem dynamics lieraure base on housing, real esae, urban developmen and relaed hemes. I discerns hree main schools or groups of sysem dynamics projecs. Firs, i covers he rich lieraure surrounding Urban Dynamics, he conroversial cornersone projec ha sill influences he relaionship of sysem dynamics and oher social sciences, mainly economics. The second school is locally based in he Neherlands, a counry wih a srong hisory of housing policy and a focal poin of sysem dynamics research. The hird school relaes o he pos2008 oupu of sysem dynamics on housing, real esae and he grea financial crisis. Chaper III also caalogues remaining isolaed sudies. Chapers IV o VII cover he projecs menioned in he ile of his hesis. These pilo projecs are supporive of he second bach of work. The pilo projecs represen a decade of professional involvemen wih housing, sysem dynamics modeling and applied policy research. In hindsigh, heir concepual bases evolved owards increased use of academic housing marke concepualizaions, even if he descripions of he respecive modeling conexs are mosly narraive and commonsense based. The projecs were published earlier as applied policy research repors of wo insiues, as conribuions o housing and sysem dynamics conferences and in an academic journal. They provide he groundwork for answering research quesions 5 and 6. Chaper IV presens he Haaglanden projec, carried ou in he region around The Hague in he Neherlands around Cenral o he modeling problem were he effecs of urban ransformaion and greenfield consrucion on he chance of households finding a new renal dwelling. The paricipaing sakeholders gained new insighs in housing marke dynamics and succeeded in reconciling a policy conflic. The projec is relevan as i demonsraes proper applicaion of sysem dynamics and models realisic housing marke processes such as waiing liss, vacancy chains and redlining. I was published as Eskinasi, Rouwee, and Vennix (2009). Conenwise, he Haaglanden projec is largely based on he menal models of regional housing policy makers and consulans, raher han on exising academic concepualizaions. As i conribued o organizaional learning, i is a relaively successful sysem dynamics case. I conribues several iniial modeling building blocks and insighs on applicaion of he sysem dynamics mehod o he purpose of his hesis. The second case sudy (see chaper V) focuses on he model developmen of Houdini. Houdini connecs o he naional discussion on housing policy effeciveness. In disance o he Haaglanden model, Houdini is solidly founded on a wellknown housing economics model and added insiuional aspecs like land use planning, ren regulaion, fiscal morgage suppor and residual land pricing policy. Furhermore, i adds slow changes on he demand side moving from growh o populaion shrinkage and ells of debaes wih main sream economiss and how his conribued o model improvemens. Houdini iself is documened in several publicaions, i.e. Eskinasi, Rouwee, and Vennix (2011) and Eskinasi (2011b). Insiuional and/or policy modeling componens of Houdini were

22 Inroducion 19 also used in he Middle Incomes and Morgage model. Bare essenials of Houdini are conained in he second illusraion in secion II.2. The hird modeling projec named Middle Incomes sprang from he debae on new regulaions for sae suppor o housing associaions which affeced housing availabiliy for middle income households (see chaper VI). Model consrucion was embedded in a mixed mehodology research projec wih poliical exposure. The model is a descendan of Houdini, adding furher refinemen of demographic and housing choice processes, housing allocaion sysems and behavior of differen ypes of supply side acors on he basis of academic housing lieraure. The model gained sufficien confidence of leading academics and high ranking policy officials o be used in debaes wih Parliamen. Some of he new insighs sill reverberae among policy makers. The full projec repor was published as Eskinasi, De Groo, Van Middelkoop, Verwes, and Conijn (2012). A shorer repor on he model is available in Eskinasi (2013). Some imporan insighs were inegraed in De Groo and Eskinasi (2013); De Groo, Van Dam, and Daalhuizen (2013). The final projec repored in chaper VII focuses on he dynamics of he morgage debs of Duch households and he impossibiliy of significan reducions. The model was developed in close cooperaion wih housing economiss and is finding were circulaed wih policy officials. Research repors are available in Duch in Schilder, Conijn, and Eskinasi (2012) and Schilder and Conijn (2012a). The model adds new elemens of morgage debs o he knowledge base. The hesis concludes wih preliminary findings on successful applicaion of sysem dynamics in housing research, open discussions and quesions for furher research. The appendices conain full model specificaions and experimenal seups.

23

24 II Housing research issues and sysem dynamics The purpose of his chaper is o find answers o he firs and second research quesions. I describes wha conemporary housing research issues could possibly benefi from a sysem dynamics approach in secion II.1. This firs secion herefore focuses on exploring he presence of complexiyrelaed research issues in a wide range of housing sudies, raher han on criically crossexamining he varying and someimes opposing sances wihin he housing lieraure, he laer being ouside he scope of his hesis. Secion II.2 explains and illusraes he naure of sysem dynamics modeling in more deail. This par covers research quesion 1 and also provides handson illusraions how sysem dynamics can be applied in housing research. Secion II.3 hen conemplaes he alleged isolaed posiion of sysem dynamics among social sciences and ries o draw lessons for fruiful cooperaion wih housing researchers, hus providing a leas parial answers o research quesion 2. II.1 Conemporary research issues in housing sudies A common concepualizaion of he housing marke A common economic concepualizaion of housing and real esae markes is he four quadran model (furher: 4QM) by Di Pasquale and Wheaon (1996). This model discerns hree imporan and closely ineracing submarkes (see figure 1). I is useful in he ligh of he purpose of his hesis, as i is sock and flow based and includes a basic feedback srucure. Figure 1 The four quadran model Source: Di Pasquale and Wheaon (1996)

25 22 chaper II The upper righ quadran represens he marke for housing services. Here, consumers bid periodical paymens or ren o acquire consumpion of housing services. The demand curve is negaively sloped and parameerized by he oal housing sock and demand fundamenals like number of households, household incomes ec. The upper lef quadran represens he housing asse marke, where hese periodical rens are being capialized ino real esae asse prices. The angle of he posiively sloped curve represens he capializaion facor used. The lower lef quadran is he housing consrucion marke. Here, housing prices, consrucion and developmen coss and characerisics of he building indusry deermine he level of new consrucion. Finally, he lower righ quadran adjuss he oal housing sock on basis of new consrucion and depreciaion or demoliion. The overall srucure of he four quadran model is equilibrium seeking or a balancing feedback loop, which is in line wih neoclassical microeconomic heory. Di Pasquale and Wheaon (1996, pp. 1218) demonsrae he effec of differen exogenous shocks o he model (i.e. a demand shif, changing capializaion rae, differen consrucion coss), which bring he model ino new equilibriums. The shocks have differen resuls on he ses of he four axes of he model (sock, ren, price and consrucion). Modeling of real housing marke processes Maclennan (2012), however, sresses ha no only modeling based on neoclassical microeconomics is insrumenal o improving undersanding of housing marke dynamics. He proposes a complemenary modeling approach wih sronger emphasis on modeling he acual processes on housing markes. He poins ou several argumens why such an approach is valuable nex o mainsream neoclassical microeconomic analysis assuming perfecly informed and compeiive markes. Housing has many innae complexiies due o produc variey, is fixaion in space and is longeviy. I herefore differs from oher consumpion goods. These characerisics make housing markes more complex han sylized markes. Maclennan (2012) suggess ha, imperfec and delayed marke informaion makes expecaions of consumers and expers (like brokers) maer for he overall dynamics on he shor run and ha hese facors are herefore relevan for analysis. He suppors his argumens (2012, p. 6) by poining a unseling gaps beween common academic concepualizaions and he noions of serious marke paries on he working of he housing marke. He also argues ha for acual housing policy making, he common microeconomic perspecive of longrun equilibrium may no be saisfacory. Policy issues may afer all arise from he fac ha housing markes are no in equilibrium, or ha insiuional characerisics obsruc equilibrium seeking behavior (Maclennan, 2012). Ye anoher issue is he poenial difference beween he oucome of efficien marke processes and poliically defined desirable oucomes. Paramoun o he dynamics of he housing marke is he small size of supply hrough new consrucion in relaion o he exising housing sock (Ball, Meen, & Nygaard, 2010). Also vacan exising housing plays an imporan role in maching househuners and houses. Vacan housing and sale ime were demonsraed o have a srong impac on housing prices (Di Pasquale, 1999; Di Pasquale & Wheaon, 1996). I is plausible o claim ha he housing marke has muliple clearing processes: hrough new consrucion

26 Housing research issues and sysem dynamics 23 and hrough supply of exising housing, semming from migraion, demographic change and ohers. Residenial mobiliy of households riggers vacancy chains which link mobiliy wih (socioeconomic) change in urban areas (Clark, 2012). On such a local level, reinforcing processes may lead o nonlinear, complex and chaoic behavior, e.g. when neighborhoods undergo rapid processes of filering up or down (Galser, 2012). Furhermore, he spaial fixiy, durabiliy and capial inensive naure of boh he consrucion process and real esae ownership make owners (Galser, 2012; Maclennan, 2012) and conracors (Ball, 2012) suscepible o risks and adjus heir behavior o hese risks. Home owners end o display loss aversion (Van Dijk, 2013b) and value housing equiy differenly from oher forms of equiy when deciding on housing consumpion (Van Dijk, 2013a). Complexiies and whie spos on he supply side Di Pasquale (1999) reviewed real esae economics lieraure and wondered why we don know more abou housing supply. Her review yielded several solid conclusions, bu also some difficul puzzles. Even hough more maerial is available on he supply of singlefamily houses han of mulifamily renal dwellings, she claims ha overall empirical evidence on he working of he supply side is far less convincing han on he demand side. From he viewpoin of mainsream microeconomic heory, he explanaory power of he mos obvious independen variables is insufficien. Neiher house prices nor consrucion coss maer o he exen he neoclassical model predics. On he oher hand, he impac of sale ime and of inflaion is larger han expeced (Di Pasquale, 1999). Consrucion apparenly responds more o changes in house prices raher han o he price level (Ball e al., 2010). Home improvemens were found o have higher income elasiciy han repair expendiures. Considering governmen inervenion in he housing marke, Di Pasquale (1999) found ha subsidies for renal housing for middleincome families end o displace privae invesmens. Providing public or social housing for lowincome groups, on he oher hand, generally increases he housing sock and does no exhibi a displacemen effec. Tax reamens for renal housing significanly affec he level of consrucion. The common noion is ha housing supply is slow and sluggish due o a) produc characerisics b) he lenghy, complex and risky naure of he developmen process c) he dependence on land availabiliy and d) he presence of land use or planning sysems. Housing lieraure shows lile agreemen on he proper way of measuring price elasiciy of supply and consequenly, esimaes vary widely from zero o infiniy. Bu even wih comparable mehodologies, significan variaion remains when comparing naions wih differen spaial and insiuional characerisics, when comparing local siuaions wih diverse land use regulaions and spaial condiions and even beween differenly sized consrucion firms (Ball e al., 2010). Ball (2012) claims ha much less research effor has been concenraed on he acual house building indusry han on he impac of land availabiliy, local land monopolies and planning resricions. Maclennan (2012, p. 13) expecs ha supply side sluggishness

27 24 chaper II canno be aribued o planning resricions alone:.. he challenge for applied analysis is o idenify he balance of marke failure versus planning resricions. Bu mos of all, he mainsream lieraure on housing supply is lacking in horough undersanding of he complex decision making processes of developers and suppliers in he marke (Di Pasquale, 1999, p. 21). More precisely, developers and ohers funcion (and make decisions on basis of marke informaion) only wihin he framework of he real esae marke and is economic and insiuional conex, so ha feedback processes emerge from he ineracion beween paries (Trevillion, 2002). Housebuilding involves a chain of specialized, inerrelaed firms raher han heoreical monolihic suppliers. These linkages beween hese enerprises are crucial for undersanding he naure of housing supply (Ball, 2012). One apparen obsrucion for such research is he lack of saisical daa on he company level, which is unforunaely ime consuming and expensive (Di Pasquale, 1999). Dynamics of housing demand and behavior There is a vas body of lieraure surveying he relaionship beween age, life evens and housing behavior of households or individuals. Life evens include demographic evens like leaving he parenal home, parnership and household formaion, childbirh and separaion hrough divorces or deah. Life evens and household decisions also relae o he educaional and labor career. Housing behavior includes decisions on e.g. residenial mobiliy, enure and neighborhood choice and housing expendiure (Van Ham, 2012). Under he curren dynamic lifecourse approach (Clark, 2012), facors from boh he macro conex and from he individual level deermine (revealed or saed) housing preferences and acual housing decisions or behavior. Household resources and resricions (like income, healh, family size, social neworks, job locaion) and facors from he macroconex like housing availabiliy and affordabiliy deermine and someimes significanly limi he realisic se of opions of a household. The dynamic lifecourse approach sems from he older lifecycle approach wih a raher fixed, linear progression of life and housing sages. The newer approach allows for muliple pahs hroughou life (e.g. he increasing number of singles, childless couples, divorce and remarriage ec.). Moreover, individual life and housing evens (labeled micro ime ) are embedded in he macro conex (or macro ime ) or hisory of he economic, social, poliical, insiuional and spaial developmen on he sociey level (Van Ham, 2012). The role of micro and macro ime in housing careers is explicily named as one of he areas of fuure research for housing sudies (Van Ham, 2012, p. 59). Differen birh cohors experience hisorical or macro ime evens a differen sages in heir housing lifecourse, for insance a housing boom or bus. Moreover, he simulaneous concurrence of lifecourse evens of one cohor may consiue a major even in macro ime for oher cohors. As an example, when he large cohor of baby boomers will sar leaving he housing marke a old age, he high number of vacan dwellings may provide ample opporuniies for young households o ener ino home ownership in regions wih ensed marke, or plunge regions wih already weak demand ino a housing bus (Mankiw & Weil, 1988; Myers & Ryu, 2008). Generaional dynamics, macro and micro ime hen inerlock ino a complex process deermining he real opions for house huning families,

28 Housing research issues and sysem dynamics 25 couples and singles and possibly even influencing he macroeconomics of he housing marke. Cerain dynamics inerfere wih he process of households opimizing heir housing siuaion under a given se of preferences and resricions. Firs, he high cos of moving house (financially and oherwise) may cause significan ineria in he process of adaping he housing siuaion o he preferences. Second, evidence exiss ha households adap preferences o wha hey perceive as realisic opions (Van Ham, 2012, p. 48). Third, as neighborhoods express he social siuaion of heir inhabians and many households end o seek ou people like hemselves, such decisions become inerrelaed, allowing for reinforcing feedback and fas changes in neighborhood composiion (Gibb, 2012). Housing cycles and insiuional feedback loops House prices porray significan volailiy relaive o changes in fundamenals like ineres raes and demographic and economic growh (Glaeser, Gyourko, & Saiz, 2008). Wheaon (1999) explored he fundamenal condiions under which real esae cycles occur in he analyically solvable 4QM. He confined himself o his singlefeedback loop model based on mainsream microeconomic heory wih fully raional agens, as more complex feedback srucures are difficul o handle analyically (Wheaon, 1999, p. 212). He reconfirmed ha such models do no exhibi endogenous cycles, bu ha cycles can be produced as he model reacs o periodical exernal shocks. I should be noed ha older macroeconomic models predaing he sric applicaion of microeconomic foundaions (e.g. he 1936 Tinbergen model (Dhaene & Baren, 1989) or Keynesian models) were perfecly capable of generaing endogenous business cycles and ouofequilibrium dynamics (Boumans, 2011). Adapively or myopically acing agens (i.e. make sysemaic misakes in forecasing he resuls of shocks e.g. using curren or hisoric values for forecass) are only a precondiion for endogenous cycles. In his case, he occurrence of cycles criically depends on he imporan feaures ha characerize differen ypes of real esae (Wheaon 1999 p. 210), such as he raio of demand versus supply elasiciy, growh and depreciaion raes and supply delays. Glaeser e al. (2008), for insance, found ha emporary bubbles can occur when buyers and suppliers are overly opimisic abou fuure prices, unil he delayed supply response rebalances demand. Areas wih more elasic supply have shorer bubbles, bu face more risk of overbuilding wih negaive consequences for overall welfare. There is some evidence ha realworld behavior of housing consumers does no fully comply wih he raionaliy axiom of neoclassical models (Case & Shiller, 1989; Glaeser, 2013; Hamilon & Schwab, 1985). Finally, even wih fully raional agens, insiuional feaures and/or insiuional feedback relaionships (Wheaon, 1999, p. 210;225) may cause he 4QM o exhibi endogenous oscillaion. Such insiuional feedback may consis of governmen inervenions (Di Pasquale, 1999), bu also of feedback mechanisms wih oher markes, like he financial marke (Anundsen & Jansen, 2013), consrucion, developmen and land markes (Ball, 2012; Di Pasquale & Wheaon, 1996, pp. 3536; Trevillion, 2002)

29 26 chaper II In summary: wha conemporary housing research issues exis? In summary and in ligh of he purpose of his hesis, he following research issues in conemporary housing sudies were found, which may poenially benefi from using sysem dynamics: 1. Due o he specific characerisics of housing (spaial fixiy, durabiliy and capial inensive naure), realworld processes in he housing marke may no necessarily mach he assumpions of neoclassical microeconomics and herefore exhibi differen dynamics. 2. Realisic housing markes have processes running in differen imeframes (e.g. shorrun price dynamics, mediumrun supply responses and longrun demographics changes) in ineracion. 3. The ime horizon of research for housing policy issues may no necessarily mach he longrun horizon of microeconomic equilibrium. I may herefore be producive o also model and analyze rajecories owards equilibrium and ouofequilibrium siuaions. 4. The housing supply secor does no consis of monolihic suppliers bu is raher a supply chain consising of many specialized and ineracing eniies. Feedback beween hese eniies adds o he complexiy of dynamic behavior of he supply chain. 5. Housing supply resuls boh from new consrucion and from vacancies wihin he exising sock. These wo clearing processes can alernaely dominae he dynamics of he housing marke. 6. Due o several facors, households do no coninuously adap he housing siuaion o he preferences, bu links exis wih life sages, job decisions, age, healh ec. Social saus aspecs of housing creae reinforcing feedback on local or neighborhood level. 7. The housing marke is indissolubly conneced wih he land, consrucion, developmen and financial markes and wih he insiuional conex, which mos presumably add o he complexiy of feedback and may induce booms and buss. II.2 The sysem dynamics perspecive and mehod Graned he assumpion ha he above research issues exis in conemporary housing sudies, we mus proceed o invesigae wheher sysem dynamics has suiable characerisics for addressing hese research issues. We firs describe he general concepual naure of sysem dynamics. Concepual cornersones of sysem dynamics Sysem dynamics is he science of undersanding dynamic behavior of complex sysems by means of compuer simulaion. Is purpose is o aid policy making in social, economic, managerial and oher seings. Fundamenal o sysem dynamics is he endogenous perspecive: problemaic behavior of complex sysems sems from is inernal feedback srucure and exogenous impulses are mere riggers (Richardson, 2011). This key issue is commonly formulaed as he aphorism srucure drives behavior. Sysems wih comparable feedback srucures will exhibi comparable dynamic behavior, even if

30 Housing research issues and sysem dynamics 27 he respecive conens are wide apar (socalled isomorphism). This perspecive gave rise o he developmen of ses of socalled sysem archeypes. Senge, Ross, Smih, Robers, and Kleiner (1994) presened wellknown narraive archeypes like fixes ha fail, success o he successful and ragedy of he commons. Wolsenholme (2003) resrucured he sysem archeypes ino a more analyical core se. The synax or mahemaical specificaion of sysem dynamics models is based on several elemens: he closed boundary around he sysem; he cenral feedback loops; sock (also known as levels or accumulaions) and flow variables (or raes); goals, observed condiions, discrepancies and finally acions or decisions (Forreser, 1969; Vennix, 1996). The closed boundary around he sysem does no imply a sysem in isolaion, i is raher ha a paricular srand of dynamic behavior can be explained from he sysem srucure wihin hese boundaries. The addiion of goals, observed condiions, discrepancies and acor decision ino he models serves o embed human acors ino he complex feedback srucures of social sysems. Policies of human acors are a fundamenal par of he complex social sysem The counerinuiive behavior and policy resisance of complex social sysems sems from he inabiliy of human decision makers embedded in he sysem o properly undersand all feedback relaions wihin he sysem. The human acors in he sysem srive o pursue heir goals on basis of informaion abou he sock variables hrough influencing flow raes (Vennix, 1996, p. 45). Their policies and decision rules are herefore endogenous and a fundamenal par of he sysem. Socalled policy resisance, side effecs or adverse effecs sem purely from he imperfec percepion of causes, effecs and feedback by he acors sriving o aain cerain goals: he sysem iself jus reacs as defined by is feedback srucure and does no discern a all beween inended and uninended or adverse effecs (Serman, 2000, p. 10). There is ample empirical evidence ha human beings sysemaically misesimae he behavior of higherlevel feedback sysems (Forreser, 2007b, p. 363). Their acions and policies may be derimenal o he final oucomes. Sysem dynamics compuer simulaion help human acors undersand how feedback loop configuraions cause enacious resisance of sysems agains policies, how decisions and policies propagae and wha policy alernaives are mos effecive. Because sysem dynamics helps human acors undersand and adap social sysems, i is no deerminisic or srucuralis bu akes a middle posiion in he srucureagency coninuum (Lane, 2001, p. 113). Sysem dynamics and housing (economic) research share many conceps Even hough he sysem dynamics communiy ends o emphasize difference wih oher mehodologies (mos noably wih saisical economeric modeling), sysem dynamics shares many aci underlying assumpions wih mos oher modeling and simulaion echniques (Meadows, 1980). They are based on a logical, scienific, wesern mode of hough, in which evens and social processes have causes ha can be undersood and possibly alered. Furhermore, he worldview is managerial: problems should be acively solved, no passively endured. All mehods rely on compuers for assising he human brain and on compuer models as he bes represenaions of social sysems. Finally, hey

31 28 chaper II are based on he idea ha human behavior is o some exen predicable and can be represened by means of equaions. Furhermore, socks, flows and (balancing) feedback loops are by no means exclusively used in sysem dynamics, bu are also common in housing and economic heory, e.g. in Tinbergen s 1936 macroeconomic model (Dhaene & Baren, 1989), Poerba (1984) and ohers. Compuer simulaion over ime is also prominen in cellular auomaa and agen based simulaion (Benenson & Torrens, 2004), in cohor componen based demographic forecass (e.g. De Jong e al., 2005) and in economic dynamic modeling (e.g. Donders, Van Dijk, & Romijn, 2010). I is common parlance o alk abou he housing sock and o perceive new consrucion as an annual addiion or inflow o his sock. Equilibrium seeking is he fundamenal propery of socalled balancing feedback loops. Socalled reinforcing feedback loops exhibi exponenial growh, e.g. a savings accoun wih ineres or a wageprice spiral wih ouofconrol inflaion. Bu for relaively simple models wihou feedback or wih a single feedback loop, here is no much added value of sysem dynamics over sandard analyical soluions. Sysem dynamics excels a socalled higher order feedback problems where wo or more feedback loops inerac in nonlinear fashions. Such srucures easily surpass he possibiliies of analyical soluions and are very difficul o handle saisically. Sysem dynamics herefore resors o compuer simulaion wih dedicaed sofware packages 4. Wheaon (1999) exacly idenified his demarcaion line when he found ha, even wih models adhering o sric microeconomic foundaions, addiional insiuional feedback loops may bring an oherwise equilibrium seeking sysem ino endogenous cyclicaliy. Wih higher order feedback as is home erriory, sysem dynamics differs from oher mehodologies in he scope of answers i delivers, in informaion bases, mahemaical procedures and validaion approaches. Sysem dynamics focuses on undersanding dynamic behavior and no on poin predicion Sysem dynamics focuses on he longer erm and general undersanding of he dynamic naure of problems. I focuses on idenifying behaviordriving srucures, effecive pressure poins for and side effecs of policies. I is helpful in discerning sensiive and insensiive parameers and can help focusing saisical analysis on hose parameers ha really maer. On he oher hand, sysem dynamics is no concerned wih shorerm, raher precise predicions or forecass of economic or oher variables nor wih deailed implemenaion of policies. Furhermore, i is of limied value for problems of disribuion over classes, persons or geographical areas (Meadows, 1980). Neiher is sysem dynamics an innaely spaial simulaion mehodology as cellular auomaa and agen based simulaions (Benenson & Torrens, 2004). Many auhors, however, (e.g. BenDor & Kaza, 2012; Despoakis & Giaouzi, 1996; Hovmand, 2005; Juila, 1981; Lowry & Taylor, 2009; Sancar & Allensein, 1989; Singhasaneh, Lukens, Eiumnoh, & Demaine, 1991) have 4 Commonly used sofware packages include Vensim by Venana Sysems, IThink / Sella by ISee Sysems and Powersim by he homonymous Norwegian sofware company.

32 Housing research issues and sysem dynamics 29 worked on inegraing sysem dynamics wih spaially oriened simulaion mehodologies, GISbased analysis and visualizaion mehodologies. Saisical and economeric analysis is more suiable for shorerm precise predicion. Linear causal relaionships allow for exensive use of saisical daa and a muliude of echniques for validaing he fi of model oucomes o observed rends (Meadows, 1980). These mehods criically depend on good saisical daa sources and are somewha limied o siuaions no oo differen from hose represened by he daa. This is exacly wha Di Pasquale (1999) hined a. Sysem dynamics, on he oher hand, focuses more on feedback srucures driving behavior. I is herefore less dependen on high qualiy saisical daa for parameer esimaes and is capable of working wih boh quaniaive, qualiaive, explici and implici sources of knowledge, wih he process approach of sysem dynamics being a cornersone (Meadows, 1980). Validaion of sysem dynamics models pus srong emphasis on srucural and behavioral properies of models and no on saisical fi o observed rends alone (Forreser & Senge, 1980; Serman, 1984). Prominen echnical differences are in mos cases relaed o differences in focus and purpose of he echniques. Meadows (1980, p. 47) suggesed ha economeric analysis and sysem dynamics represen differen niches in modeling echniques wih mehodological discussions ending o degenerae in classical crossparadigm confusion. A firs illusraion: he 4QM in sysem dynamics noaion A firs illusraion will help clarify he picography of sysem dynamics 5 and he easy ranslaion of he 4QM from a mahemaical ino a sysem dynamics form. For he basic 4QM wih one feedback loop, ranslaion ino sysem dynamics form does no add much value, possibly apar from ime simulaion and he clear designaion of he feedback srucure. Conversely, we should conclude ha he 4QM wih is single balancing loop is a useable embryonic sysem dynamics model and ha added value may follow when adding more (insiuional) feedback. Ineres rae Households & Price Ren incomes B1 Consrucion sared Housing under consrucion Consrucion finished Housing sock Demoliion Consrucion coss Consrucion ime Life ime Figure 2 The 4QM in sysem dynamics noaion 5 Several excellen exbooks on sysem dynamics are readily available, e.g. Fisher (2004); Serman (2000).

33 30 chaper II Figure 2 presens he 4QM in he common noaion of sysem dynamics. Boxes represen sock variables. Double arrows wih valves depic flow variables and he small clouds on he far lef and far righ represen sysem boundaries. The socks and flows connec ino a socalled supply chain. The single arrows represen causal links beween variables. Pluses and minuses designae he polariy of hese relaionship. The polariy of he link beween incomes and ren is posiive as higher incomes lead ceeris paribus o higher rens. Increasing ineres raes lead o lower real esae prices, and herefore, his relaion has negaive polariy. Feedback loops are found where several individual links connec ino a circle. The main balancing feedback loop of he 4QM runs from sock hrough ren, price and consrucion back ino sock 6. For he sake of diagram clariy, loops are numbered and designaed wih B for balancing and R for reinforcing. Diagram clariy also requires verbose variable names. On he far righ of he figure, he flow variable demoliion is regulaed by an auxiliary variable life ime. The original 4QM has depreciaion rae insead. Even hough boh specificaions are mahemaically equivalen 7, sysem dynamics prefers ime variables. The use of ime variables allows o easily model delays in he supply chain of he 4QM, as he cauious reader already noiced. Consrucing an acual model in sysem dynamics sofware requires enering equaions, parameers and sarup values for all elemens in he causal diagram and adding opional graphs, ables and conrols for a model user inerface. This firs illusraion used he equaions in Di Pasquale and Wheaon (1996, pp. 818). As an innaely ime simulaion based mehod, sysem dynamics allows o model processes running wih differen ime frames. The common dichoomy beween longerm and shorerm (or insananeous) processes found in mainsream microeconomic models is somewha arificial from he sysem dynamics poin of view: sysem dynamics modeling auomaically enforces ime inegriy, even if shor run processes on he housing services marke work on monhly ime scales, supply reacions ake years and demographic processes reshaping demand unfold over decades. A simulaion run calculaes he ime rajecory of all variables using Euler s or RungeKua s numerical inegraion mehods wih an arbirarily small ime sep. This focus on ime simulaion allows sysem dynamics models o be in equilibrium, o be moving owards a sable or moving equilibrium or be in a sae of oscillaion or even random chaos. This is imporan as Maclennan (2012) expressed doub o he relevance of exane longrun equilibrium assumpions for applied policy research. Using ime simulaions allows sysem dynamics o ranscend hese limiaions. Modeling processes, validaion and reporing aspecs The process of consrucing a sysem dynamics model generally involves a number of seps (a.o. Vennix, 1996). The firs sep is he definiion of he problemaic issue and he 6 Smaller feedback loops exis around he socks and he flows. Demoliion, being an ouflow, decreases he housing sock. Demoliion, in urn, is proporional o and herefore dependen on he size of he housing sock. A balancing feedback loop emerges, exponenially depleing he housing sock (ceeris paribus). A similar srucure governs housing under consrucion and consrucion finished. 7 If life ime l = 100 years, hen depreciaion rae δ = 1 / l = 0,01 / year.

34 Housing research issues and sysem dynamics 31 purpose of he model, by means of describing he dynamics of cenral variables over ime, he socalled reference mode of behavior. This is no necessarily sraighforward, as Vennix (1996, p. 13) argues ha differen sakeholders may hold very differen opinions on he core problem involved. Furhermore, sysem dynamics does no necessarily equae he reference mode of behavior wih he exac numerical ime rajecory of variable. Defining a reference mode of behavior as a paricular rend or relaive movemen of variables is equally accepable. The second illusraion will provide an example of such a definiion. In he second sep, he sysem concepualizaion or dynamic hypohesis connecs all relevan variables causally ino he feedback srucure of he sysem. The hird sep akes he concepualizaion furher o a formal quaniaive model in erms of equaions and parameers. These sages are crucial in he consrucion and validaion of sysem dynamics models. Sysem dynamics does no provide he subsanive conen for is model. The conen mus be elicied from oher sources and properly ranslaed ino a valid model srucure adequaely capuring he dynamic behavior under scruiny. In validaion, proper represenaion of he sysem srucure is emphasized more han precise fis o hisorical daa. Knowledge eliciaion echniques are crucial for hese firs hree seps. In sysem dynamics parlance, knowledge eliciaion serves o make he menal models of conen expers on he social sysem under scruiny explici. Menal models include concepual models (e.g. he 4QM), policy logic, knowledge sored in descripive saisics and saisical models, opinions, guesses, esimaes and inuiions of expers and professionals in he fields. Some menal models may be explici and welldocumened in wrien sources like academic and professional lieraure, policy documens and saisical daabases (Forreser, 1992). Oher informaion and menal models may be aci. Sysem dynamics applies a muliude of mehods o elici knowledge, like srucured inerviews, lieraure desk research and cross examinaion of sources. Many auhors conribued o documening, sysemaizing and esing eliciaion mehods for effeciveness (a.o. Andersen, Richardson, & Vennix, 1997; Ford & Serman, 1997; Rouwee, 2003; Vennix, 1996). Paricipaive mehods are imporan in sysem dynamics for eliciing he more aci knowledge and menal models. Paricipaion echniques include informal echniques like brainsorming and open discussion, bu also wellsrucured mehods like Delphi, Nominal Group Technique and especially Group Model Building (Rouwee, Vennix, & Van Mullekom, 2002; Vennix, 1996). I has been demonsraed ha acors learn mos from sysem dynamics projecs when hey are deeply involved in he model consrucion process (Lane, Monefel, & Husemann, 2003; Meadows & Robinson, 1985; Rouwee, 2003). In he fourh sep, he model undergoes rigorous esing and sensiiviy analysis in he fourh sep. Assessing and acceping (or rejecing) model validiy is cenral o he fifh sep. As models are mere mahemaical represenaions of imperfec human menal models of real world problems, validiy is no absolue bu raher a process of gradual confidence building in he suiabiliy of a model for a given welldefined purpose and

35 32 chaper II he problem is seeks o address (Forreser & Senge, 1980). This is comparable o he Popperian increasing degree of corroboraion of scienific heories. Forreser and Senge (1980) provide an array of eigheen specific ess for he nonlinear sysem dynamics models, focusing on hree imporan aspecs: srucure verificaion, behavior reproducion and policy implicaion ess. Ten ou of hese eigheen are considered core ess. Proper srucure verificaion is a crucial sep in validaion. From wihin a subsanive social science, his aspec is ofen lef implici, bu for a muliconen mehodology like sysem dynamics, esing for conenwise proper model srucure is indispensable and mus be made explici. The firs es is ha he modeled sysem srucure mus no conradic exising knowledge of he real world sysem. This is where he menioned knowledge eliciaion echniques come ino play. Parameers in he model mus concepually and numerically correspond and he model should respond properly o exreme condiions. The very imporan boundary adequacy es relaes he model srucure o is dynamic behavior. The es is passed when he model includes all relevan componens o reliably reproduce he reference mode of behavior. If no, imporan sysem componens migh be missing. Finally, dimensional consisency checks agains he inclusion of arificial parameers wih socalled exoic unis, ofen revealing flawed srucural specificaion. In many modeling projecs, validaion and he modeling process seps are closely linked. Especially he seps of srucure verificaion, model concepualizaion and formal specificaion are in many cases inerwined. The second aspec of validaion refers o model behavior. Firs of all, anomalous model behavior (conradicing behavior of he real world sysem) is unaccepable. Moreover, he model should accuraely reproduce he specific sympoms of he reference mode of behavior ha moivaed model consrucion, including periodiciies of cyclical behavior and relaive phases of variables wihou he exensive use of exogenous ime series driving he model. Afer all, sysem dynamics focuses on modeling endogenous causes of dynamic behavior. Reliance on exogenous ime series would hen invalidae he model. The model should also qualiaively predic plausible fuure paern and evens. Saisical esing of model oucomes o hisorical daa wih he Theil saisics of inequaliy (Serman, 1984) is common, bu no regarded as he mos imporan aspec of validaion. Founding faher (Forreser, 2007b, pp ) even opposes srong reliance on saisical fi wih hisoric daa for validaion purposes. Policy ess of he model focus on wheher changing policies in he sysem generae plausible responses. Furhermore, i should be esed wheher uncerainy in parameer values would change he policy recommendaions based upon he model experimens. The ulimae es, however, is when policy recommendaions from he model lead o improvemen of he performance of he real world sysem. Unforunaely, his ulimae es is difficul o perform. Once he model aained a cerain level of credibiliy (on basis of he above ess and in paricipaive processes, he accepance by he problem holders), modelers carry ou policy experimens and evaluae hem in he sixh sep. Simulaion of policy experimens in he presence of sakeholders is he poin in he process where mos learning effecs occur

36 Housing research issues and sysem dynamics 33 and where sysem dynamics has mos influence on he menal models of he sakeholders (Rouwee, 2003). Finally and hopefully, involved sakeholders will use heir conclusions from he modeling projec in daily decision making or in improving he sudied sysem. Modeling projec reporing guidelines Sysem dynamics projecs are very varied and mulifaceed regarding hemes and approaches. In order o sysemaize approaches and projec repors for cumulaive knowledge building, heories on effeciveness of guidelines for reporing (group) modeling projecs were laid ou by e.g. Andersen e al. (1997), Andersen, Vennix, Richardson, and Rouwee (2007), Rouwee e al. (2002) and Rouwee and Vennix (2006). These guidelines include a wide range of aspecs, from he problem background, hrough selecion of he group of paricipans and model descripion o he assessmen of he influence of he group model building projec on he paricipans percepion of he core issue involved. Furhermore, Rahmandad and Serman (2012) developed guidelines for he reporing of he acual models, including parameer sources and policy experimen seup. Advanced modeling and analysis echniques In addiion o he sandard approach described above, several more advanced sysem dynamics echniques exis. The mos common sysem dynamics sofware packages suppor more complex sensiiviy analysis wih sochasic parameer values, Mone Carlo and Lain hypercube simulaion. Cerain sofware packages implemen objecive opimizaion by means of he Covariance Marix Adapaion Evoluion Sraegy algorihm (Hansen, 2010). Exploraory Sysem Dynamics Modeling and Analysis combines radiional sysem dynamics modeling wih he deep uncerainy concep, where no agreemen exiss on he proper concepual model o address a cerain research or policy issue (Pruy, 2010). Loop dominance analysis focuses on idenifying he feedback loop dominaing sysem behavior over a cerain ime inerval by means of boh srucural and behavioral echniques (Ford, 1998). A second illusraion: moving owards insiuional feedback This example illusraes several aspecs of he sysem modeling process, i.e. he idenificaion of a problem conex, reference mode of behavior and modeling purpose. I demonsraes how sysem dynamics incorporaes insiuional feedback and moreover, clarifies he absracion level and focus on undersanding dynamic behavior. The problem conex is found in srong percepion changes in he early 2000 s on he success of Duch housing policy afer he Second World War. Tradiionally, housing was conceived of as a meri good and many sae suppor programs exised for housing associaions, urban renewal, housing allowances, morgage ax breaks ec., no o menion he closekni relaion beween housing policy and land use planning. From 2005 onwards, however, (insiuional) economiss increasingly criicized sae inervenions. Ren regulaion and implici subsidies o enans were said o srongly increase demand for social housing and fiscal suppor o home owners was held responsible for inflaing house prices (Conijn, 2008; Donders e al., 2010). Spaial planning, building regulaion and municipal land pricing allegedly resriced new consrucion

37 34 chaper II oo much so ha prices would increase even more (Besseling, Bovenberg, Romijn, & Vermeulen, 2008; Buielaar, 2010; Eichholz & Lindenhal, 2008; Renes, Thissen, & Segeren, 2006). Financial innovaions allowed higher and more risky leveraging of households wih ineresonly morgages and saving schemes (Van Ewijk & Ter Rele, 2008). Mos auhors advocaed farreaching reforms owards marke liberalizaion. This conex is suiable for sysem dynamic analysis as he criical economiss claim policy resisance of he housing marke and as he issue involves complex insiuional feedback loops. Sysem dynamics analysis sars wih defining a reference model of behavior, a descripion of he problemaic behavior of he sysem over ime in one or several main variables. A suiable reference mode of behavior for his example is found in he dynamics of house prices, consrucion volumes and consrucion coss in figure 3. I demonsraes rapid price increases of houses, whereas rens and consrucion coss show far more moderae increases. Bu mos sriking is ha consrucion moves in he opposie direcion of house prices, which is in conradicion wih mainsream microeconomic heory. We define his paricular sympom as he reference mode of behavior. The economic auhors idenified four insiuional facors (ren regulaion, subsidies, land use planning, acive municipal land policies), ha may poenially (and parially) explain he reference mode of behavior. They do no, however, clearly demonsrae which facor is essenial for he opposie movemen of consrucion and house prices. This illusraes he poin made by Maclennan (2012, p. 13), calling for more balanced analysis of marke failure and planning or policy resricions. 300 Reference mode of behavior Index 1985= house prices consrucion coss real rens housing sock consrucion Year Figure 3 Reference mode of behavior for second illusraion Source: Besseling e al. (2008).

38 Housing research issues and sysem dynamics 35 The modeling purpose will herefore be o simulae hese facors in order o deermine he mos probable cause(s) for he reference behavior. In proper sysem dynamics pracice, his means developing srucural modificaions o he basic 4QM in figure 2 and esing wheher hese capure he essence of he reference mode of behavior, i.e. opposie movemen of consrucion and prices in an oherwise balancing loop srucure. The quinessence of sysem dynamics is in finding sock and flow srucures causing a paricular dynamic behavioral rai and no in fully maching a given reference rajecory of a variable 8. The model in figure 2 is suiable as a deparure poin for creaing a baseline simulaion and adding he four srucural modificaions 9. Is iniial values are se up o saic equilibrium and oupu variables are presened as indices wih heir iniial values a 100%. This will allow o focus properly on dynamic paerns wihou disracion from acual numerical values. In he 1990 s demand growh and decreasing ineres rae had riggered he Duch housing marke o exhibi he criicized reference mode of behavior. This illusraion will use a simple demand and ineres rae shif in simulaion years 10 o 20. The simulaion horizon is se a 100 years in order o allow he effecs of hese simuli and he srucural modificaions o play ou. Di Pasquale and Wheaon (1996) already documened he effecs of single parameer changes on he final equilibrium values. Our baseline (see figure 4) herefore only adds insighs on he ime rajecory owards new equilibrium and he balance beween he differen simuli. The ineres rae decrease appears sronger han demand growh, as rens equilibrae o a new level below 100%. Crude as i may appear, he baseline is only a saring poin for invesigaing he effecs of he srucural modificaions on he rajecories of prices and consrucion. Afer all, we are esing for srucures exhibiing opposie movemen of consrucion and prices, i.e. he fundamenal problemaic aspec of he reference mode of behavior. I is repeaedly emphasized ha his is he quinessenial purpose of sysem dynamics modeling. The firs and second srucural modificaion mimic ren regulaion and fiscal morgage suppor respecively 10. Their ime rajecories are compared o he baseline. In he former case, a ren ceiling variable is added prohibiing ren o exceed is iniial value. Ren does no reac anymore o simulaed demand so ha boosed prices and consrucion can 8 Forreser (2007b) claims ha any model wih enough parameers can be manipulaed o mach (nearly) any hisorical curve. His sance is ha fi o hisorical daa is even misleading when parameer manipulaion precedes over careful analysis of underlying sock & flow and feedback srucures. The demonsraion model here is raher crude a capuring he reference mode of behavior. I neverheless inheried is essenial insiuional sysem srucures from Houdini (see chaper 0). Houdini akes more exogenous parameers like income and household developmen and ineres rae. I maches is reference mode of behavior in a saisically significan sense. This suppors Forreser s argumen. 9 The equaions of Di Pasquale and Wheaon (1996, pp. 810) conain several elemens wih socalled exoic unis in he ren and consrucion equaions. Sysem dynamics modeling requires operaional hinking and insiss on using only variables wih a clear realworld represenaion and comprehensible unis. Adding saisically convenien variables (like elasiciy or regression parameers) is no acceped in good sysem dynamics pracice. The problemaic equaions, however, can be easily respecified. The demand funcion of a Cobb Douglas uiliy funcion under budge resricion leads o fixed proporional expendiure. Esablishing a lookup relaion beween profis (prices minus consrucion coss) and consrucion volumes correcs he consrucion equaion. These respecificaions improve boh uni consisency and compaibiliy wih microeconomic heory. 10 The model diagrams for ren regulaion and fiscal morgage suppor do no fundamenally differ from he basic model. Therefore, hey are no depiced.

39 36 chaper II be solely aribued o he decreasing ineres rae. As increased demand is no accommodaed, rens sar decreasing below he 100% mark abou 15 years laer han in he baseline simulaion. 300 Baseline simulaion 250 Index year 0 = Sock Ren Price Consrucion Simulaion ime (years) Figure 4 Baseline simulaion resuls Simulaion wih ren regulaion Index year 0 = Sock Ren Price Consrucion Simulaion ime (years) Figure 5 Simulaions resuls wih ren regulaion

40 Housing research issues and sysem dynamics 37 In he laer case, fiscal morgage suppor is modeled as a 20% subsidy on annual paymen for housing services as in Donders e al. (2010). When suddenly inroduced in year 10, fiscal morgage suppors spurs prices and consrucion even more. Is dynamic response is similar o demand simulaion, even if i differs in peak ampliudes and final equilibrium. 400 Simulaion wih fiscal morgage suppor Index year 0 = Sock Ren Price Consrucion Simulaion ime (years) Figure 6 Simulaion resuls wih fiscal morgage suppor Boh simulaions have prices and consrucion move in he same direcion (see figure 5 and figure 6), so from a sysem dynamics poin of view, we mus conclude ha hese insiuional srucures do no explain he reference mode of behavior, i.e. opposie movemen of hese wo indicaors. The wo oher insiuional facors, land use planning and municipal land price policies, involve more elaborae modificaions o he model srucure. In he Neherlands, land use planning for housing is radiionally srongly focused on accommodaing demographic growh. Eichholz and Lindenhal (2008) claim i disregards oher (economic) pressures, whereas Renes e al. (2006) sugges i is over resricive in economically srong regions. Bu will a simulaed land use planning sysem show opposie movemen of prices and consrucion? Figure 7 presens he modified model. The housing supply chain is exended wih an sock variable for zoned housing capaciy 11 and an inflow for newly zoned capaciy. We amend household growh o a fixed annual percenage and add an auxiliary variable forecasing he populaion en years ahead. This auxiliary governs he inflow of zoned housing capaciy ino he sysem. The modificaion does no add feedback o he sysem, bu imposes a resource consrain o he operaion of he main balancing loop. 11 A more elaborae model migh discern beween land area zoned in hecares and he proposed housing densiy, which may also vary. For he sake of simpliciy and in ligh of is purpose, his model direcly adds zoned housing capaciy in erms of houses.

41 38 CHAPTER II household growh demographic forecas Consrucion coss Ineres rae Profi new households Price B1 Households budge share for housing income growh Ren Income income change newly zoned zoned capaciy consrucion sared Housing under consrucion consrucion finished Housing sock demoliion Consrucion ime Life ime Figure 7 Modified 4QM wih land use planning This hird simulaion shows radically differen behavior. Figure 8 shows ha in he firs decade, his pushes up rens, prices and consrucion only slighly. In year 10 income growh and ineres decrease kick in and rens, prices and consrucion accelerae. This deplees he zoned capaciy, which abruply becomes he limiing facor for consrucion. Demoliion, being proporional o sock, sars exceeding consrucion and he housing sock slowly declines. Due o household growh, he inflow of newly zoned capaciy and consrucion will gradually increase over ime, bu do no keep up wih prices and rens, which coninue o rise significanly more han in he amended baseline. The srucural modificaion caused sharp changes in dynamic behavior. Bu i does no ye show opposie movemens of prices and consrucion. 350 Simulaion wih land use planning Index year 0 = Sock Ren Price Consrucion Zoned cap Simulaion ime (years) Figure 8 Simulaion resuls wih land use planning

42 Housing research issues and sysem dynamics 39 The fourh possible facor is municipal land pricing policy. In he 1990 s, many municipaliies inroduced acive land policies and favored residual land pricing over cos reimbursemen in order o capure planning gains and use hese for oher public services (Buielaar, 2010). Residual land values are by no means unique o he Neherlands, bu follow from he willingness o pay for locaional advanages where supply is fixed (Di Pasquale & Wheaon, 1996, pp. 3536). Maximum land prices hen equae marke prices for real esae minus consrucion coss for maerials and labor (Morley, 2002, pp. 7577). income growh developmen coss Profi R1 change of devm cos Price Ineres rae budge share for housing B1 Ren Income Households income change new households household growh consrucion sared Housing under consrucion consrucion finished Housing sock demoliion Consrucion ime Life ime Figure 9 Modified 4QM wih residual land prices In he modified sysem dynamics model in figure 9, we assume a delayed adapaion of developmen coss (maerials, labor and land prices) o house prices. Developmen coss are now endogenous and feed back ino profi as before. This connecs a new reinforcing feedback loop R1 running hrough sock, ren, price, developmen cos and consrucion. This fourh simulaion has again radically differen behavior from all previous simulaions (see figure 10). The sock is coninuously decreasing, prices, rens and developmen rise in synchrony wih he demand and ineres impulses. Consrucion follows a disinc paern. In he firs decade wihou simuli, developmen coss adjus o house prices from heir iniial level. Consequenly, profi and consrucion drops. The simuli occur in year 10 and cause a sudden upsurge of consrucion. When he simuli sop in year 20, consrucion winesses a sharp drop again and recovers only marginally in he long run. The gradual rise of prices and rens mus be aribued o he declining sock and consan demand from year 20 (demoliion is above consrucion).

43 40 chaper II Simulaion wih residual land prices Index year 0 = Sock Ren Price Consrucion Devm coss Simulaion ime (years) Figure 10 Simulaion resuls wih residual land prices In figure 10, his model does reveal occasional opposie movemens of prices and consrucion in he firs decade and around year 20. Analysis of equaions reveals ha consrucion has become dependen on price change raher han price level, in line wih he empirical findings of Glaeser e al. (2008). We may have arrived a a dynamic srucure based in housing conceps, capable of posiively confirming counerinuiive empirical findings. In oher words, we buil a enaively plausible dynamic hypohesis of real esae markes, residual land values and observed correlaions of prices and consrucion. Our experimen also enaively suggess ha he oher insiuional facors do impac house prices and consrucion, bu no in he paricular manner we chose as he reference mode of behavior. This is he auhor s percepion of he quinessence of he added value of sysem dynamics. This small illusraion reveals how sysem dynamic can conribue o e.g. he challenge [...] o idenify he balance of marke failure versus planning resricions (Maclennan, 2012, p. 13). Tha said, in a realisic research projec, he model should undergo far more esing and sensiiviy analysis, sand in much closer comparison o empirical daa, may possibly need furher adjusmen o fi a paricular naional conex and combine he four single modificaions described above. Bu in his sage, i merely serves as an illusraion where sysem dynamics is effecive, how i incorporaes insiuional feedback and a wha ype conclusions i may arrive. II.3 Sysem dynamics in isolaion Nowihsanding is suiabiliy for ackling cerain issues in housing research, sysem dynamics operaes largely in isolaion of oher social sciences (Repenning, 2003). The isolaed posiion is parially relaed o hisorical evens in he 1970 s around one of he cornersone projecs of sysem dynamics, namely Urban Dynamics (Forreser, 1969).

44 Housing research issues and sysem dynamics 41 A derailed echnical debae over validaion echniques lef he sysem dynamics communiy wih deeply ingrained noions of anagonism owards saisically based economeric modeling (Alfeld, 1995). This is covered in more deail in secion III.2. Repenning (2003) poins a he progress economics and oher social sciences have made since hen in coping wih dynamic complexiy. Bu he also concludes ha sysem dynamics is sill enangled in reinforcing processes prevening i o sep ou of isolaion and engage in cooperaion wih oher social sciences. Anoher poenial cause of isolaion can be found in he fundamenal paradigms and modus operandi of sysem dynamics. Sysem dynamiciss are in mos cases specialiss in mehod bu generaliss in model conen: Forreser covered indusrial managemen (1961), urban growh and decay (1969), economy, resource depleion and susainabiliy (1971) and more. Vennix (1996) includes housing, commercial flee and healh care opics all in one book. The generalis approach allows o ransfer learning from one conex o anoher: similar sysem srucures will behave similarly, regardless wheher he sysem is a housing marke, a fishing flee or a bacerial colony. This follows logically from he perspecive of srucure driving behavior (or isomorphism), which is a grea asse for sysemaic, generalized knowledge building on behavior of dynamic feedback srucures. Bu conversely, exacly his mehodological specializaion prevens sysem dynamics praciioners from esablishing deep rappor wih researchers in oher social sciences. Repenning (2003) acknowledges his endency as one of he poenial errors in his effors o apply sysem dynamics in oher social sciences failure o ground my work in he language and he lieraure of he field I was rying o ener (2003, p. 320). His conclusion in 2003 bears sriking resemblance o he criicisms in he 1970 s ha Urban Dynamics does no inegrae or even menion exising demographic, economic and geographic heories (Gray, 1972, p. 144; Rohenberg, 1974, pp. 1920; Sagner, 1972, p. 199). The process approach of sysem dynamics also ends o favor he involvemen of policy makers and managers over he paricipaion of oher researchers, basically because he former focus more on finding new soluions o real world problems han on mehodological debae (Repenning, 2003, p. 320). Bu again, his does no conribue o inegraing sysem dynamics wih oher social scieniss. In addiion, he mehodological specializaion may also parially explain why sysem dynamics works in a paricular field (e.g. housing, ecology) are no commonly sysemaized ino posiion papers or oherwise, wih he noable excepion of projec managemen (Lyneis & Ford, 2007). Sysem dynamics produced a significan bu fragmened knowledge base on housing, real esae and urban developmen, conaining over 150 sudies, which was largely unnoiced by he housing research communiy. Repenning (2003) draws imporan lessons from he successes and failures in his cooperaion wih managemen scieniss in he applicaion of sysem dynamics. Firs, he recommends ha modeling work should be solidly grounded in he conceps, language and lieraure of he field of ineres. Two more suggesions relae o he communicabiliy of models and modeling insighs: large and complex models will deer nonsysem dynamiciss raher han evoke heir ineres. Furhermore, proper mehods should be used for helping he audience of a sysem dynamics model o develop insigh on how he paricular srucure relaes o dynamic behavior. He illusraes how a very simple,

45 42 chaper II onesock model helped he aviaion disaser research communiy o venure in new, previously unexplored direcions. Smar inegraion of wellesablished insigh and a simple dynamic model led o his paricular success sory. His fourh and final recommendaion is o focus on solving conen problems raher han o engage in mehodological debaes in modeling echniques. II.4 Conclusions Taking ino accoun he objecive of his hesis and he housing research issues found previously, sysem dynamics is a suiable complemenary (raher han alernaive) mehodology for he following reasons. 1. I is suiable for handling nonequilibrium siuaions ranging from rajecories owards equilibrium hrough moving equilibriums and oscillaion (or marke cycles) o chaoic behavior. I is herefore able o ranscend he a priori equilibrium assumpions found overresricive for housing research. 2. Wih human acors and heir decision as fundamenal sysem componens, sysem dynamics is easily capable of handling he endogenous formaion of housing preferences o marke informaion, life sages ec. in a srucured sense. This conribues o he modeling of realisic housing marke processes. 3. The same applies for housing supply. Decisions of he many paries involved in housing developmen are also easily made endogenous, again for modeling realisic processes. Furhermore, sysem dynamics has a srong knowledge base in indusrial organizaion and supply chain managemen. Exising insighs may provide a head sar for modeling he indusrial organizaion of he housing supply secor. 4. As an innaely ime based mehod, sysem dynamics is capable of handling housing marke processes wih differen ime frames. This again helps modeling realisic housing marke processes and ranscends he somewha arificial dichoomy of shor run and long run processes in economic modeling. 5. Sysem dynamics allows for and even excels a adding addiional feedback loops o models, beyond he possibiliy of analyical soluion. This provides ample opporuniies for adding insiuional feedback like linkages o e.g. land and financial markes, governmen policies, reinforcing processes in neighborhoods and ohers. 6. Through is emphasis on menal models, knowledge eliciaion and sakeholder paricipaion, sysem dynamics can operae in environmens where daa limiaions hamper mainsream saisical analysis. Furhermore, I argue ha cooperaion beween sysem dynamics and social sciences (housing research in his paricular case) is isomorphic o modeling projecs in he sense ha validaion of sysem dynamics for social sciences is a process of gradual building of confidence. The sudied lieraure suggess he following recommendaions for successful cooperaion wih oher social scieniss: 1. Research using sysem dynamics should be horoughly grounded in he language and lieraure of he field involved, housing research in his paricular case.

46 Housing research issues and sysem dynamics The use of sysem dynamics will be more successful when inroduced unobrusively and focused on research problem solving, in order o circumven fruiless mehodological debaes. 3. Sysem dynamics will be more accepable when i firs confirms exising insighs from oher mehods and hen gradually ranscends owards is naural domain of higher order feedback.

47

48 III Lieraure review of sysem dynamics on housing III.1 Overall remarks and descripive saisics As menioned in he inroducion, a sysem dynamics knowledge base exiss on housing, real esae and urban developmen. Unforunaely, i is available only in a fragmened way and i is no conneced o mainsream housing research. The hird research quesion of his hesis is focused on mapping his knowledge base in order o provide easier access for fuure research. In order o answer his quesion, all relevan publicaions in he bibliography of he Sysem Dynamics Sociey (2009) wih iles and descripions including urban, hous*, real esae were included on he long lis. Any oher sources known o exis were added, including books, book secions, journal papers, working papers, draf documens and conference conribuions from he 2010, 2011, and 2012 sysem dynamics conference proceedings. New enries were fed back o he Sysem Dynamics Sociey for updaing he bibliography. In oal, we obained 154 enries, of which 28 journal aricles and 73 conference conribuions. 44 iems are book conribuions, of which 33 are direcly relaed o Urban Dynamics, i.e. boh volumes of Readings in Urban Dynamics (Mass (1974) and Schroeder, Sweeney, and Alfeld (1975) and Inroducion o Urban Dynamics (Alfeld & Graham, 1976). 20 ou of 28 journal aricles were published before Only of conference papers and absracs in proceedings, he period from 1990 onwards conribued a major share (54 ou of 73). Afer iniial superficial reading, hose sudies were seleced for furher analysis ha a) somehow include a housing or real esae marke i.e. aspecs of demand, supply, consrucion ec. focusing on marke or governmen mechanisms and b) presen (oucomes) of a quaniaive simulaion model. The hird research quesion focuses on making a comprehensive caalogue 12 of sysem dynamics works on housing and herefore, no addiional crieria were se for he qualiy of hese works. The chaper herefore encompasses everyhing from groundbreaking cornersone sudies o only jus accepable conference conribuions 13. The fourh research quesion, on he oher hand, is concerned wih sysemaizing and inegraing hese maerials in a form supporive of esablishing proper connecions beween sysem dynamics and housing sudies. For his purpose, we classified he maerials ino four groups on basis of modeling heme, model srucures, geography, crossreferencing and use of oher influences. 12 Alhough some oversighs are highly probable. 13 Sysem dynamics works no acceped for conferences are no included in he bibliography and herefore logically, also no presen in his overview.

49 46 chaper III The firs group is named afer he seminal book Urban Dynamics (Forreser, 1969). Works in his group eiher direcly relae o Urban Dynamics or use very similar model srucures. The descripion of his group in secion III.2 focuses especially on Urban Dynamics iself and he early 1970 s, as his period in very insighful as o he somewha isolaed posiion of sysem dynamics Repenning (2003) described. A second argumen is ha Urban Dynamics sill srongly influences sysem dynamics projecs on urban developmen. The more recen projecs in he laer Urban Dynamics Group are caalogued only briefly wihou furher analysis. The reasons for his will be explained laer. A second group of housing relaed sysem dynamics works (see secion III.3) is locally based in he Neherlands. In addiion o he naional conex, works of he Duch group share a srong focus on changing housing policies of he governmen and oher acors like housing associaions. The saring poin of his group is he housing associaion model ITS (Vennix, 1996). Many works in his group were carried ou in cooperaion wih he Nijmegen based research group. The pilo projecs in chapers IV o VII share hese defining characerisics and herefore also belong o he Duch group. The hird group, described in secion III.4 is he mos recen and focuses on he worldwide impac of he 2008 credi crisis on housing markes. Sudies in his socalled Recen Real Esae Dynamics Group focus mosly on issues like real esae marke cycles, he role of financial markes and speculaion, he effec of governmen inervenions and decision making of acors in he real esae marke. More han boh previous groups, he Recen Real Esae Dynamics Group refers o and builds is models (parly) on basis of conceps from mainsream housing and real esae research. This explains why he works in his group were subjeced o somewha deeper analysis: hey may prove mos helpful in building sysem dynamics model srucures rooed in housing heories. The pilo projecs in especially chapers V o VII are also based on mainsream housing research conceps and herefore also relae o his group. Finally, all remaining sysem dynamics sudies are caalogued in secion III.5. All sudies demonsrae some relaion o housing, urban developmen and real esae, bu do no connec o Urban Dynamics, Duch housing policy, o he 2008 credi crisis, nor o mainsream housing and real esae lieraure. III.2 Urban Dynamics Group The Urban Dynamics Group is described firs, as i is he oldes and larges group. This chaper firs presens a small summary of he book Urban Dynamics (Forreser, 1969) and proceeds o describe he criical reacions i provoked and he defenses aken. The hird par of he chaper summarizes and caalogues he more recen works based on Urban Dynamics. Forreser s Urban Dynamics Modeling conex Urban Dynamics is a cornersone of sysem dynamics modeling and addresses he causes of urban decay in he 1960 s in many U.S. major ciies. The dynamic hypohesis underlying he model is ha he mix of housing, populaion and indusry creaes endogenous

50 Lieraure review of sysem dynamics on housing 47 processes of growh, sagnaion and decline. Urban areas exis wihin a limiless environmen (1969, p. 15). This means ha ineracions beween ciy and environmen do exis, bu he urban area does no significanly aler is environmen. The sock and flow srucure of he model consiss of hree subsysems for populaion, housing and businesses. Wihin he populaion, Forreser discerns hree caegories he managerialprofessional class, working class and he underemployed. Residenial and indusrial srucures also have hree sages or socks. Processes like consrucion, demoliion, migraion, birh and deah influence hese socks. Ageing processes play an imporan par in Urban Dynamics. Premium housing may filer down o become worker or underemployed housing; companies evolve from new enerprises hrough maure businesses ino declining indusries. An inricae nework of raios (like housing densiies and jobs o populaion raes), axes and araciveness mulipliers (1969, p. 18) link ogeher he main srucure and deermine immigraion of he differen social classes, consrucion, demoliion and ageing of residenial and indusrial srucure. Urban Dynamics iself is a echnical book. Afer a general inroducion on urban decay and sysem dynamics modeling, i rapidly proceeds o describe he more han 100 equaions in he model and simulaion oucomes from he full model. Alfeld and Graham (1976) gradually build up a simplified bu comparable urban model named URBAN1 explicily for learning and eaching purposes in Inroducion o Urban Dynamics. They include he ineracion beween housing, business and populaion bu absrac from he ageing chains presen in he full model. Overall dynamic behavior of he model The overall dynamic behavior of he urban dynamics model shows wo disinc, remarkably differen phases of developmen. During a firs phase of growh, a new, sill small ciy radiaes wih opporuniies and aracs people and businesses from all over he limiless environmen. I sars expanding and grows exponenially for several decades unil i fills up is land area wih business and residenial srucures and he growh slows down raher abruply 14. Raher han going ino sabilizaion, he ciy sars o sagnae and deeriorae because he ageing processes ake precedence over he growh fosering facors. Former new enerprises develop ino declining indusry. Premium and worker housing filer down ino housing for he underemployed. The ciy will arac more underemployed people, bu wihou giving hem a perspecive of socioeconomic improvemen any more. Policy programs Forreser hen proceeds o es common urban programs for improving he urban condiions. He finds ha many wellinended inervenions fail o make a change for he beer and someimes even worsen he siuaion. Jobs and raining programs for he underemployed and sae financial aid o he ciy have a shorrun posiive effec, bu end o increase he longrun araciveness for underemployed, hus reinforcing he effec 14 Some evidence was documened of Old world ciies going hrough several subsequen phases of growh and sagnaion in Eskinasi (2012).

51 48 chaper III of rapping hese groups in a siuaion wihou educaion and employmen. The consrucion of lowcos housing deerioraes he siuaion faser and deeper han boh oher programs (1969, p. 65). Forreser simulaes several less common approaches, e.g. o consruc new houses for he working class and / or he managerial class. He finds, however, ha his does no lead o improvemen eiher, because he overall raio of housing in he oal land area becomes oo high. The consrucion of new indusrial aciviy produces small posiive changes, bu does no reverse he sagnaion process. Demolishing declining indusry has mixed meris: i gives more room for new vial businesses, decreases he ax raio bu worsens he job opporuniies for he underemployed. Slum housing demoliion has comparable effecs, bu a he expense of a higher oumigraion of underemployed groups. Forreser finally concluded ha wo main facors cause urban decay: a) an increasing share of old residenial and indusrial srucures and b) a oo high share of housing in he oal land area, leading o unfavorable raios of populaion o jobs. A ciy can only mainain is socioeconomic vialiy by coninuously implemening policies focused on encouraging new businesses and discouraging oo much housing consrucion. These policies are applicable in proper locaions wihin he ciy suffering mos from urban decay (1969, p. 105). Criicisms and defense of Urban Dynamics Urban Dynamics me srong emoional opposiion (Forreser, 2007a, p. 349) because of is message ha mos urban policies followed in he 1960 s and 1970 s were derimenal o he urban economy. Is conclusions were no acceped because of being wrong, bu mosly for being poliically unaccepable. A large effor o validae he model sranded (Alfeld, 1995, p. 100) in he ypical crossparadigm confusion beween sysem dynamics and economerics as described by Meadows (1980). On several occasions, however, model applicaion o urban problems in specific ciies and owns led o significan consensus for acion and successful implemenaion of urban policies. I should be noed ha hese policy projecs were successful because hey were argeed a very specific policy discourses and finding supporing logic by means of (adaped versions of) he model, raher han o ake any given model as a saring poin and hen commen hem for lack of specificiy and poenial daa problems. A number of criicisms o Urban Dynamics are recurren. Mass (1974) and Schroeder e al. (1975) address hese in Readings in Urban Dynamics, wo volumes providing many clarificaions of Forreser s iniial exs and many modificaions of he original model 15. The recurren criicisms include: The lack of use of exising daa (Averch & Levine, 1971, p. 158, Babcock, 1972, p. 149, carelessness abou proper calibraion Kadanoff, 1972 and specific daa of paricular ciies Gray, 1972, p. 143). Conclusions follow from Forreser s assumpions, raher han urban realiy (Babcock, 1972, p. 149, Rohenberg, 1974, pp. 1920). 15 Mos of hese modificaions do no fundamenally aler he dynamic behavior.

52 Lieraure review of sysem dynamics on housing 49 Forreser does no ground his model in exising economic, demographic and geographic heory (Gray, 1972, p. 144, Rohenberg, 1974, pp. 1920, Sagner, 1972, p. 199). Ciy cener o suburb ineracions are no presen in he model (Babcock, 1972, p. 149, Garn & Wilson, 1972, p. 154, Gray, 1972, p. 143). Forreser overesimaes he aracion of underemployed groups hough housing availabiliy (Babcock, 1972, p. 149 and underesimaes he ineracion of he ciy wih is wider environmen Garn & Wilson, 1972, p. 154). Forreser does no clearly define a healhy ciy and is herefore subjecive in evaluaing policies (Garn & Wilson, 1972, p. 155, Gray, 1972, p. 143, Jaeckel, 1972, p. 216, Sagner, 1972, p. 199). Forreser leaves ou some imporan oher policy experimens i.e. ren conrol and dynamics wihin he environmen (Gray, 1972, p. 141). Neverheless, mos criics agree ha Forreser s model and approach are cerainly of imporance for research, educaion and policy making on urban managemen. This is refleced in saemens like he following: Despie hese criicisms of Forreser s conclusions I would argue ha his model making is so brillian and beauiful ha his ideas are cerainly of examinaion and furher developmen (Babcock, 1972, p. 149). Urban dynamics is an exremely useful educaional ool for sudens of public policy boh on he managerial and on he research level (Belkin, 1972). The synhesis of all equaions (in he model) produces general behavior maching several US ciies and some credence is len o he conclusion ha some powerful forces underlying urban condiions reside wihin he ciy limis (Harlow, 1973, p. 126). The laer Urban Dynamics Group As menioned before, Urban Dynamics is sill very influenial in sysem dynamics hinking on urban issues. This secion presens work from 1980 onwards on urban dynamics in chronological order. Ineresingly, he use of Urban Dynamic conceps appears o accelerae afer 1990 and even more afer Works were included when using (a large par of) he full Urban Dynamics model (Forreser, 1969) or he simplified educaional URBAN1 model (Alfeld & Graham, 1976) or conceps closely relaed o hese models. General applicaions of he Urban Dynamics model Aracil (1980) invesigaed equilibrium condiions wihin he URBAN1 model. Braden (1994) describes he developmen of a basic, four sock Urban Dynamics syle model educaional purposes in a museum. Suksawang and Srinivas (1995) propose o use he URBAN1 model conneced o GIS for eaching graduae sudens on urban growh wih Bangkok as a case sudy. They do no documen a full operaional model, nor do hey refer o Kuroda and Mark Tsaur (1990), who also use URBAN1 for a Bangkok case sudy. Zagonel dos Sanos (1996) applies a model similar o Urban Dynamics o he governmenplanned ciy of Brasil. He finds ha he iniial sae effor in building he new ciy coninues o arac new migrans for a far longer period, only o sop when job opporuniies were in balance wih housing and populaion. The innae dynamics of he

53 50 chaper III ciy aracs more han wo imes he number of inhabians originally planned for and creaes large saellie owns of informal housing. Feng, Lu, and Wang (2001) presen preliminary oucomes of a rudimenary model on urban growh. They connec closely o Urban Dynamics by using idenical subsysems (i.e. housing, populaion and businesses) bu do no refer o Forreser (1969). Blanco (2011) applies sysem dynamics modeling o he issue of growh of informal housing in Lain America. He finds several deerminans for he growh of such selemen: increasing demand for lowcos housing, a public sysem ha canno keep up wih demand growh because of crosssubsidizaion and public policies aggravaing overcrowding and using up valuable scarce resources. Ciy cener and suburb ineracion Sanders and Sanders (2004) also develop a spaial version of Urban Dynamics o include ineracions beween differen residenial areas in a ciy and apply i o Roerdam. They find ha heir modificaions aler he overall dynamics only slighly. This finding is in line wih many oher auhors, e.g. Mass (1974) and Schroeder e al. (1975). Uchino, Furihaa, Tanaka, and Takahashi (2005) follow up on Sanders and Sanders (2004) and elaborae on relaive araciveness wihin he conex of social science in general and spaial urban dynamics specifically. Periurban dynamics and urban sprawl Dyner, Berrio, and Bolivar (1989) and Dyner, Munoz, and Quinero (1991) sudy he dynamics of he periphery of an urban area by means of a small model. They focus on land use change in he rapidly developing urban periphery in Lain America and find sysem dynamics generally useful for he problem involved yielding plausible simulaions. The auhors do no include housing marke issues in he analysis. Voyer (2004) venures ino he balance beween affordable housing and urban sprawl. He finds innovaive policy opions for aaining boh goals. Urban dynamics and ranspor ineracion Kuroda and Mark Tsaur (1990) connec an urban ransporaionplanning model for Bangkok o URBAN1 in order o find a balance beween accessibiliy and socioeconomic developmen. They conclude from heir model ha inadequae ransporaion resrics socioeconomic urban growh. Swanson (2003) exends he urban dynamics emplae o include commuing ranspor in a ciy disaggregaed ino several disrics. He ess several policy experimens for several Briish owns. Urban dynamics and policy evaluaion Jarzynska and Richardson (2006) challenge he original Urban Dynamics model by simulaing recen US housing policies for lowincome groups. They find ha hese new policies raher reinforce Forreser s iniial findings han disprove he model, bu also plead for a shorererm frame in which boh longerm urban growh and shorerm needs of lower income households can be balanced. J. Richardson and Elizabeh (2010) compare he recommendaions of urban dynamics o he policies followed by he Singapore

54 Lieraure review of sysem dynamics on housing 51 governmen. They find many correspondences even hough Singaporean planners mos likely were no aware of he Urban Dynamics book. Oher urban dynamics sudies Boman (1981) elaboraes a regional planning model for he Eindhoven region, bu does no provide any repors on model applicaion in a policy seing. Moffa (1983) consruced a simulaion model of urban growh on basis of Marxis heory. Saeed (2010) connecs he perspecives of Urban Dynamics and Schumpeer s model of creaive desrucion. He assers ha we should see he povery condiions in many developing as he sagnaion phase in Urban Dynamics raher han he saring poin of growh. Speaking figuraively, he claims ha space in such counries migh be filled up wih unfavorable social and poliical insiuions hampering growh and mainaining sagnaion. He demonsraes ha in boh models, shorer life for infrasrucure and capial sock yields beer overall performance as his akes off he edge of he ageing processes. III.3 Duch Housing Policy Group Tradiionally, he Neherlands had srongly sae influenced spaial planning and housing policies. Boman (1981) can be said o reflec he firs poswar episode of housing policy wih a srong emphasis on sae planning. From he 1990 onwards, he governmen sared o decenralize and liberalize he housing marke o a cerain exen, bu no wihou reaining a significan amoun of influence (see IV.2, V.2, VI.2 and VII.2). Insiuional sruggles e.g. beween he sae and he housing associaions over heir huge equiy, beween poliical paries, academics and economiss on he durabiliy of he morgage ineres ax reducions define he nex wo decades. The credi and housing marke crisis is mos likely o increase he pace of furher reforms. The welldocumened simulaion model ITS focused on he new siuaion for housing associaions, when many sae subsidies and loans were canceled ou around 1995, hus leaving hem wih sraegic independence from he governmen, bu also wih he full financial ownership risk. Vennix (1996) provides a deailed accoun on is consrucion for invesigaing wheher housing associaions can aain heir social goals (i.e. provide housing for lower income groups) while also reaining financial solvabiliy. Togeher wih a consulancy company for housing associaions, Vennix found he age srucure of he housing sock o be a sensiive variable in regards o his problem. Rouwee (2003) documened several ITS applicaion cases when invesigaing group model building effeciveness. Afer several iniial successes, however, he model go enangled in annoying dynamics of increasing deails and eroding clien confidence and was finally discarded as a business risk (Eskinasi & Fokkema, 2006). Yücel and Pruy (2011) simulae he dynamics of improving he energy efficiency of he Duch housing sock and find significan ineria in he sysem. Radical new policies mus be found in order o aain se goals in ime. De Groen (2011) developed an elaborae model on housing migraion chains when from 2005 onwards, housing mobiliy sared decreasing. De Groen, Pruy, and

55 52 chaper III Bouwmeeser (2012) assess he effecs of several reforms of he Duch social housing secor on he housing marke behavior of lower income household wih a preliminary, relaively disaggregaed model. Van Niselrooij (2009) models he relaion beween fiscal morgage suppor and house process in he Neherlands. We also documened several sudies using qualiaive sysems hinking for Duch housing policy problems. Qualiaive sysem hinking shares much of he dynamics and complexiy perspecive wih quaniaive sysem dynamics. I does no use quaniaive compuer modeling bu explains feedback and complexiy by means of feedback loops in a more narraive way. Fokkema, Haanemaijer, De Rooij, and Eskinasi (2005) provide an anhology of housing policy relaed hemes. Jongebreur, Blom, and Van Dieen (2009) developed a causal loop diagram focused on housing saisfacion and inenions o move house. The pilo projecs in chapers IV o VII also coun among he Duch Housing Policy Group. III.4 Real Esae Dynamics Group As menioned in he inroducion, his group of sysem dynamics lieraure connecs more o he mainsream real esae economics lieraure and uses is conceps as model building blocks. This allows o inegrae and sysemaize he conribuions of his group in he form of small concep models (see G. P. Richardson, 2012) ha add up o he modificaions of he 4QM in secion II.2. They revolve around four dynamically complex aspecs of real esae markes. Furhermore, he crossferilizaion of sysem dynamics and common housing research also brings o he fore several mehodological issues for furher consideraion. Vacan real esae The works of Barlas, Özbas, and Özgün (2007), Aefi, Minooei, and Dargahi (2010) and Mashayekhi, Ghili, and Pourhabib (2009) connec he occurrence of cyclicaliy in real esae markes wih he presence of an inermediary vacan housing sock in he producion chain of housing. Even hough vacan housing is no conained in he high level 4QM, Di Pasquale and Wheaon (1996, pp ) demonsrae a srong impac of he vacancy level (or he relaed variable sale ime ) on real esae prices. They also provide a mahemaical model for he size of his effec. The same logic is presen in several sysem dynamics sudies: Barlas e al. (2007) invesigae he relaion of real esae cycles o he abiliy of developers o properly esimae demand rends and he moves of heir compeiors. Özbas, Özgün, and Barlas (2008) make a rigorous sensiiviy analysis of his model and presen findings which variables affec lengh and ampliude of he cycles. Mashayekhi e al. (2009) demonsrae he differences in cyclicaliy of housing markes wih and wihou a vacan housing sock. Vacancy srucures influencing marke dynamics are also presen in Hu and Lo (1992) and Eskinasi e al. (2012). Serman (2000, pp ) argues ha he endency of real esae acors o underesimae maerial delays is an imporan, if no he main cause of he recurren naure of real

56 Lieraure review of sysem dynamics on housing 53 esae cycles. His posiion is similar o Wheaon (1999): boh supply lags (or maerial delays) and improper acor undersanding of he real esae sysem (a.k.a. myopic expecaions) are necessary for cycles o occur. Figure 11 presens a concep model visualizing hese ideas. Ineres rae Price B1 Households & incomes Ren Occupancy ime B2 houses vacaed Housing under consrucion Vacan houses Housing sock consrucion consrucion houses sared finished occupied Consrucion coss Consrucion ime Sale ime Figure 11 Modified 4QM wih vacan housing sock demoliion Life ime The original 4QM in figure 2 srives o balance he housing sock and real demand driven by demographic and economic facors. Real demand loop B1 is generally slow as consrucion is small in relaion o he exising sock and rends in he exogenous variables (incomes, ineres rae). Vacan housing, however, srongly affecs house prices and consrucion. Wih developers underesimaing is impac, he vacan housing loop B2 can sar o dominae sysem behavior by repressing prices and consrucion. Accumulaion of vacan real esae allows price levels below capialized rens and even below consrucion coss. A relaive conrol archeype emerges (Wolsenholme, 2003) suggesing a generic soluion in finding an absolue raher han a relaive arge or pivoal variable driving consrucion. Serman s real esae model (2000) has an auxiliary variable vacancy rae insead of a sock vacan real esae. We prefer modeling a sock because of he high impac of vacancies on prices and consrucion prices (Di Pasquale & Wheaon, 1996; Mashayekhi e al., 2009), success raios for households (Eskinasi, 2013; Eskinasi e al., 2009) and for reasons of maerial conservaion (see he secion on mehodological concerns). Reinforcing capial gain or speculaion loops Oher sysem dynamics sudies emphasize reinforcing loops based on price increases as an imporan driver of real esae marke cycles. Hu and Lo (1992) found cyclical paerns in he Taiwan housing marke and hypohesize, nex o common balancing loops, hree reinforcing loops running hrough land and propery speculaion. Boh developers and consumers demonsrae speculaive behavior. Speculaive supply exis when developers ac on price increases, raher han orders from consumers or invesors. Chen (2005) adds

57 54 chaper III exernal invesors ha feed speculaion when prices rise and add o he bus when selling off speculaively bough properies. In general, real esae price growh can increase capial invesmen demand in he propery marke (in he upper lef quadran of he 4QM, aside from he residenially driven demand for housing services in he upper righ quadran). The effec of rising invesmen demand on house prices connecs a reinforcing feedback loop R1, depiced in figure 12. invesmen demand R1 invesmen yield Households & incomes Ineres rae Price Ren B1 Housing under consrucion Housing sock consrucion sared consrucion finished demoliion Consrucion coss Consrucion ime Life ime Figure 12 Modified 4QM wih invesmen demand loop The configuraion wih a reinforcing and a balancing loop maches he ou of conrol archeype (Wolsenholme, 2003). Balancing loop B1 srives o bring he marke ino equilibrium. Bu when price increases arac addiional invesmen yield driven demand, he sysem may exaggerae is response and overproduce housing. Glaeser e al. (2008) found comparable dynamics when housing markes wih high supply elasiciy can lead o lower welfare, again because of overproducion resuling in price slumps. Real esae and he financial marke Several auhors also sudied he propery finance marke. Aefi e al. (2010) find ha in markes wih very low morgage credi availabiliy, increasing he loanovalue raio can suppor new housing producion. Mukerji and Saeed (2011), on he oher hand, model dynamics of an (overly) maure financial marke. Their causal diagram reveals a success o he successful archeype (Senge, 1990; Wolsenholme, 2003) in how households finance home ownership. House price increases make real esae a more aracive invesmen caegory han savings. Decreasing ineres raes and requiremens on morgage loans allow for higher debs. And he demand creaed by higher morgage volumes boos real esae prices. Hwang, Park, and Lee (2009) idenify comparable srucures wihin he financial marke: increasing reurns from residenial morgages arac addiional

58 Lieraure review of sysem dynamics on housing 55 funding hrough he sale of morgage backed securiies and oher derivaives. The mechanism in boh sudies is success o he successful as booming real esae and financing markes exrac funds from oher invesmen classes, wheher household savings or socks and bonds in oher economic secors. The financial markes loop R2 connecs wih he capial gain loop R1 ino he menioned archeype. A hird modificaion of he original 4QM would cerainly concern he influence of household finance and he morgage marke. An iniial aemp on he basis of he sudies menioned is presened in figure 13. invesmen demand R1 invesmen yield R2 relaive yield on oher invesmens Ineres rae Price Ren Households & incomes B1 consrucion sared Housing under consrucion consrucion finished Housing sock demoliion Consrucion coss Consrucion ime Life ime Figure 13 Modified 4QM wih financial markes loop Governmen policies & inervenions A final group of sudies relaes o he effec of governmen inervenions on he housing marke. These are relevan as Wheaon (1999) claimed ha insiuional feedback loops alone may cause marke oscillaions, nowihsanding raional expecaions of acors. Park, Lee, and Hwang (2010) evaluae a conroversial package of governmen measures aimed a increasing he axaion base, prevening speculaion and expanding he supply of land for housing. Also Hwang e al. (2009) projec proposed policy logic ono heir causal diagram. Boh works idenify several uninended side effecs of hese measures. Eskinasi e al. (2011) model he dynamic effecs of several governmen inervenions (among ohers land use planning and municipal land policies based on residual land values) in search of he mos probable cause for observed counerinuiive rends of consrucion in relaion o house prices. The modificaion wih residually based land prices (see e.g. Buielaar, 2010) feeds back house prices ino developmen coss. The resuling residual land values loop R1 obsrucs he response of consrucion o prices hrough loops B1 and provides a poenial explanaion for Di Pasquale s (1999) conclusion ha neiher house prices nor consrucion coss sufficienly explain consrucion volumes.

59 56 chaper III This final feaure is added here as residual prices are generic (Di Pasquale & Wheaon, 1996) raher han bound o a specific naional conex. income growh developmen coss Profi R1 change of devm cos Price Ineres rae budge share for housing B1 Ren Income Households income change new households household growh consrucion sared Housing under consrucion consrucion finished Housing sock demoliion Consrucion ime Life ime Figure 14 Modified 4QM wih residual land prices Noe: his figure is idenical o figure 9. Loops B1 and R1 in figure 14 form he ou of conrol (Wolsenholme, 2003) or fixes ha fail (Senge, 1990) archeype. If low consrucion is he problem, hen house price growh is he fix ha will boos consrucion. However, house price growh also increases developmen coss hrough land values and hus worsens he problem of low consrucion volumes. Mehodological concerns Examining he sysem dynamic works discussed in his paper also idenifies wo ineresing modeling issues. The firs aspecs regards he inegriy of he housing producion chain. Mos auhors (Aefi e al., 2010; Barlas e al., 2007; Eskinasi e al., 2013; Mashayekhi e al., 2009) connec all socks ogeher ino a single producion chain (i.e. zoned land, houses under consrucion, vacan houses, occupied houses ec.). Ho, Wang, and Liu (2010) and Eskinasi e al. (2009) do no connec vacan housing ino he housing producion chain. Park e al. (2010) focus mainly on price dynamics and have housing producion chain variables as auxiliaries. From he viewpoin of maerial conservaion (Serman, 2000), he firs opion is preferable. Second is he occurrence of common economic elasiciy variables in sysem dynamics models (see e.g. Mashayekhi e al., 2009) which possibly conflic wih uni consisency requiremens. Elasiciy expresses overall srengh of he correlaion beween e.g. demand and prices. As sysem dynamics focuses on relaing dynamic behavior o sysem srucure, i could be paricularly helpful in finding hose srucures ha demonsrae a given elasiciy. In oher words: building sysem dynamics models of such

60 Lieraure review of sysem dynamics on housing 57 srucural feaures and measuring he price elasiciy in he models oupu can help connecing sysem dynamics and housing & real esae research. Furhermore, microeconomic heories may also conain uni consisen building blocks for sysem dynamics models (e.g. Eskinasi e al., 2013). III.5 Isolaed sudies No all sysem dynamics sudies on housing, real esae and urban developmen build consisenly upon each oher, as refleced in e.g. Suksawang and Srinivas (1995), Boman (1981) and Feng e al. (2001). This secion covers he isolaed sudies on widely relaed hemes for caaloguing purposes only, wihou furher ambiion owards analysis or sysemaizaion. Mosekilde, Rasmussen, Joergensen, Jaller, and Jensen (1985) use a small sysem dynamics model for venuring ino he dynamics of ehnical residenial segregaion and apply mehodologies from chaos heory for analysis of he dynamic paerns. HonghMinh and Srohhecker (2002) build a model of inernal srucure of he UK privae housing consrucion focused on he causes of low performance. They discern several acors as homebuilders, building maerials merchans and manufacurers. The model emphasizes producion chain aspecs and was used for finding differen scenarios for improvemen. Sehedi (2006) models shor say housing of new immigrans, raher han he dynamics of he regular housing marke and does no display any influences from or references o oher models. Kolacek (2006) gives a hisoric accoun of ren regulaion in he Czech Republic and develops a simple model for calculaing new rens on basis of aparmen premises, which is a single sock wih a proporional growh rae. No supply, demand or oher facors are included in he model. Kolacek (2006) refers o a.o. Vennix (1996) bu does no borrow any housing relaed modeling building blocks from he lieraure. Ahn and Lee (2010) consider he need for adequae sheler for he Souh Korean populaion and deermine he driving forces of hree main acors: governmen, suppliers and enans. From heir wosock model, hey conclude ha he inflow of sufficien (governmen) capial is he mos imporan facor for supplying renal housing. They do no connec o an inernaional knowledge base on housing policy nor on sysem dynamics. Skribans (2010) describes a model developed for forecasing he demand for new aparmens in Lavia based on he capaciy of he consrucion indusry, price increases and he availabiliy of finance. III.6 Conclusions Sudying exising sysem dynamics lieraure on housing indicaes he exisence of hree main groups and a number of unrelaed sudies. The firs group depars from cornersone projec Urban Dynamics, which also played an imporan role in he anagonism beween sysem dynamics and he housing economics research communiy. In hindsigh, he making of Urban Dynamics did no mach laer lessons on engaging in fruiful coopera

61 58 chaper III ion wih oher social scieniss, in paricular in connecing heir language and conceps and in he scale of he model iself. Tha said, Urban Dynamics sill inspires sysem dynamics praciioners and moreover, as will be argued in chaper IV, qualiaively similar paerns may be observed in real world ciies. The Duch Housing Policy Group is bound o is naional conex, bu has a srong focus on he effec of policy changes in his conex. The use of mainsream real esae conceps in sysem dynamics modeling can only be raced back o 2008 wih he adven of he Grea Financial Crisis. These works can be sysemaized ino small concep models on four dynamically complex feaures of real esae and housing markes, acknowledged in mainsream lieraure.

62 IV Haaglanden 16 Absrac This paper describes a group model building projec abou new housing consrucion, urban renewal and he impac of boh processes on a regional social housing marke. A eam of seven sakeholders paricipaed in model consrucion over a oneyear period. The paper addresses he modeling process, model analysis and policy experimens. The model yielded several counerinuiive insighs, helped he sakeholders o sele a conenious issue and was used in fligh simulaor workshops wih managers and policy makers. By means of quesionnaires, we found ha mos aendans in he projec or workshops consider sysem dynamics modeling as improving communicaion, insigh, alignmen and commimen o resuls. IV.1 Inroducion In his seminal book Urban Dynamics, Forreser (1969) delves ino he inricae dynamics of urban growh and decay. His main analysis is ha afer an iniial period of growh, aging of housing and businesses in combinaion wih limis o spaial growh pull he socioeconomic srucure of a ciy ou of balance. His proposed counerinuiive policy measures clashed wih he 1970 s housing an planning paradigms (Forreser, 2007a). The seing of our projec however indicaes ha in he early 2000 s, Duch urban planners had come o a policy paradigm ha is more in line wih Forreser s iniial findings. Whereas sysem dynamics inervenions have been said o improve policy making from he 1950s, only recenly a research program on effeciveness of modeling has been oulined. Andersen e al. (1997), Andersen e al. (2007), Rouwee e al. (2002), Rouwee and Vennix (2006) elaborae heories and inervenion reporing guidelines for assessing group modeling building effeciveness. Our repor on he Haaglanden projec is srucured on he basis of hese guidelines. We firs describe he conex of he inervenion. Nex we repor on he modeling process, involvemen of he projec group and he resuling model. Finally, we presen several reflecions on he impac of he projec and is qualiy as a sysem dynamics venure. IV.2 Conex of he sysem dynamics inervenion The modeling projec repored here addresses he impac of new consrucion and ransformaion of housing on social housing marke dynamics. The geographical seing is he Haaglanden region in he densely populaed wes of he Neherlands. This region 16 This chaper was published previously as Eskinasi M, Rouwee E.A.J.A., Vennix J.A.M. (2009) Simulaing urban ransformaion in Haaglanden, he Neherlands, Sysem Dynamics Review 25(3): DOI: /sdr.423

63 60 chaper IV includes he cenral ciy The Hague and is surrounding suburbs and new owns, hisoric Delf and he horiculural area Wesland. Several organizaions play a cenral role in social housing policies in his region. Sociale Verhuurders Haaglanden (SVH) is an associaion of noforprofi housing corporaions in and around The Hague. Is main ask is sraegic consuling, represenaion and lobbying for is members in various decision making forums. SVH is a small neworking organizaion and he primary clien in his projec. Is main parner and someimes opponen is Sadsgewes Haaglanden (SGH), a governmen office uniing municipaliies in Haaglanden. The corresponding auhor, a ha ime a consulan a Arivé, was he faciliaor and modeler. I is necessary o provide clariy abou our deparure poin regarding he syle of he inervenion. In his analysis of he relaion of sysem dynamics o social heory, Lane (2001) discerns several sysem dynamics approaches in a wo dimensional marix of social heory. He describes he developmen of he iniial perspecive hrough a broad perspecive o he ineracive sysem dynamics pracice. The focus of he laer is o creae a shared inerpreaion of a problem hrough personal involvemen of sakeholders. In hindsigh, his is he sysem dynamics perspecive of his modeling inervenion, more so, because he sysem dynamics pracice of Arivé sems from Vennix (1996) work on group modeling building for housing associaions. Furhermore, i is necessary o undersand ha modeling in a consulancy role is by definiion linked o a real world clien problem and ha he driving force for such work is a reinforcing loop of successful sysem dynamics business and clien saisfacion (Eskinasi & Fokkema, 2006). Grüers (2006) poins ou ha in Arivé s unsuccessful modeling experiences, he plausibiliy of he model for he clien was los. In summary, hese hree facors may explain possible biases owards sakeholder assessmens in our accoun: he educaion of he modeling faciliaor wihin he group model building school, he necessiy of clien involvemen in a consulancy seing and finally he learning effecs of negaive experiences wih losing model plausibiliy for he clien. Hisory of he problem conex The sraegic issue in his projec is rooed in he Duch housing policy. To familiarize he reader wih he seing we ouline major developmens in his field before focusing on he problem addressed in he modeling projec. The firs social housing iniiaives dae back o he 1850 s. The 1901 Housing Ac gave a legal basis for supporing privae, nonprofi housing organizaions. Bu only afer World War II he governmen sared a largescale subsidy program for social housing. War damage o he exising sock and he poswar baby boom necessiaed an unprecedened greenfield consrucion program, (i.e. consrucion on previously unoccupied land). In he 1950 s and 1960 s, consrucion volume was he main issue and dwelling qualiy only played a minor role. The 1967 record of an annual oupu of 125,000 houses is sill unbeaen, bu o his day a srong focus on greenfield developmen remains. Managemen of he social housing sock is he responsibiliy of social housing organizaions or corporaions. Housing corporaions had lile influence on consrucion policy formulaion, nor on he housing allocaion sysem, managed by he local auhoriies. Urban renewal or ransformaion (i.e. replacing rundown houses wihin exising ciies) focused on incidenal slum removal and, especially in he 1960 s, on infrasrucure improvemens and

64 Haaglanden 61 simulaion of he cenral business disric. In he 1970 s, urban renewal policies changed under pressure of he democraizaion process. Smallscale rehabiliaion of 19h cenury residenial areas focused on consrucing new social housing for he curren inhabians. This would obviously clash wih Forreser s (1969) conclusion ha consrucion of lowincome housing is derimenal or a bes neural o urban vialiy and could have mos probably provoked commens as saed in Forreser (2007a, pp ). The 1990 s offered new challenges and perspecives, when he governmen sared o reduce public spending. Financial suppor o housing organizaions was abolished, objec subsidies (i.e. on houses) were reduced and subjec subsidies (i.e. ren subsidies o households) were increased. The 1995 Bruering Ac cancelled ou all exising loans and subsidies and made housing organizaions financially independen from he governmen. The social housing secor faced hree challenges in he 1990 s. Firs, a disurbing new fac was ha middle or highincome groups occupied many social dwellings. Lowincome groups had o be allocaed o more expensive dwellings and ren subsidies rapidly increased. The policy response was o decrease he share of social housing in new consrucion and o encourage enans o move up he housing ladder. Second, he housing allocaion sysem underwen major changes in he nineies. In order o provide more ransparency, responsibiliy and freedom of choice for house huners, a new allocaion sysem was inroduced and rapidly gained in populariy. All available dwellings were published in a newspaper or (laer on) on inerne. Regisered house huners apply for he houses hey like bes. From hese applicaions, housing corporaions selec fuure enans on he basis of crieria such as age, duraion of occupancy or duraion of regisraion. Ofen, selecion procedures disinguish sarers from people already living in he region and moving on o anoher house in he same region ( onmovers ). Sarers have no previously occupied a home, whereas onmovers have. People in urgen need, for insance for medical or social reasons or afer calamiies, ge a prioriy saus. A forced move because of urban renewal also eniles applicaions o a prioriy saus. Third, he urban renewal program also me wih new challenges. Buil o relieve he poswar housing shorage, he 1950 s largescale housing esaes had gradually developed ino unaracive and problemaic neighborhoods. The focus on greenfield developmen led o uninended side effecs such as selecive migraion ou of he ciy. Policy makers will now acknowledge ha large concenraions of social housing would concenrae povery and induce urban decay. Their policy objecive was o break up large concenraions and consruc middle class housing in order o remix he populaion and improve he economic performance of ciies. Speculaing from he fac ha policy makers now saw large scale social housing as derimenal o urban vialiy bu did no ake follow up by simulaing he growh of new businesses in place, we migh conclude ha Duch policy makers have experienced a leas some of he processes described in Urban Dynamics. Neverheless, he new urban renewal paradigm demonsraes a shif owards he findings of Urban Dynamics as large scale social housing was now seen as a par of he problem raher han he soluion. Building new marke housing was also an answer o he increasing shorage of good qualiy housing, predominanly wihin he ciies (VROM, 2000). New housing consrucion is however only possible if adequae numbers of social housing are available o provide he necessary working space for urban renewal. Exising housing marke

65 62 chaper IV prognosis models confirmed his assumpion by predicing a surplus of cheap, old social renal flas. These predicions were however based on he assumpion ha consrucion of large volumes of new housing would coninue as planned. Realiy in he meanime ook anoher course. The 1990 s economic boom increased prices of marke housing o a level where an average family was no longer capable of purchasing a house. The economic downurn in he early 2000 s in combinaion wih long spaial planning procedures decreased greenfield consrucion figures. Conrary o he expecaions of some policy makers, pressure on he (social) housing marke sared o increase, raher han o decrease. Housing policy makers were confroned wih a number of choices. Sop urban ransformaion, risking furher deerioraion of he housing sock, or proceed as agreed, ignoring declining success raios of house huners and risking even deeper sagnaion of he social housing marke. Saring poin for he sysem dynamics inervenion This was our cliens complicaed siuaion a he sar of he projec. SVH and SGH agreed on he necessiy of ransformaion for urban socioeconomic vialiy. They pared ways when i came o is desired pace in relaion o new consrucion, and on he accepabiliy of oher inervenions in he social housing marke. SGH is bound by conracs wih he naional governmen on new housing consrucion volume and is eager o improve he housing mix, especially in he cenral ciy of The Hague. SGH was no inclined o accep any delays in he ransformaion program. Their policy of annually ransforming 2,000 social dwellings for a en years period ( ) was used as he base run of our simulaion. The housing corporaions and SVH are more sensiive o he decreasing success raio of heir poenial cliens. This socalled success raio was he main problem variable. I is defined as he quoien of he annual supply of social houses and he oal number of house huners compeing for hem, housing corporaions have lile conrol over new consrucion, especially in he marke segmen and advocae he direc ineress of heir cliens, defined as a high success raio. SVH herefore favored an alernaive policy opion, i.e. 1,500 dwellings in he firs five years ( ) and 2,500 aferwards ( ). SVH also considered changing he housing allocaion sysem and o allow longer ouplacemen procedures ( sage 1, described laer on in his paper) as an opion for increasing heir cliens success rae. They hoped ha his would sop he decline of he success raio recorded in he period and would srenghen is recovery in 2002 and During he debae ha followed, boh paries increasingly emphasized heir differen viewpoins on he impac of ransformaion on he social housing marke, he exen o which decreases in success raio were accepable and how o inervene. The debae became heaed o he poin ha boh parners found hemselves in ourigh conflic over he bes policy inervenion. Table 1 summarizes he main characerisics, viewpoins and policy proposals of boh organizaions.

66 Haaglanden 63 Table 1 Characerisics, viewpoins and policy proposals of paries in Haaglanden Acor SGH SVH Represens Municipal & Regional Auhoriies Housing Corporaions Fears Delay in housing ransformaion & greenfield consrucion programs Decreasing supply of housing due o ransformaion Consequences of fears Increasing urban decay, problems Decreasing clien success raio wih sae on consrucion subsidy conracs Sensiiviy o success raio Lowmedium High Proposed ransformaion policy T=2000 T=1500 STEP (1000;5) Hope Keep ransformaion in pace, posiive effecs on success raio come laer on. Slower ransformaion now leads o higher success raio, sep up when greenfield consrucion is high. IV.3 The sysem dynamics inervenion Preprojec aciviies SVH and SGH could no resolve heir problem using radiional consuling or research aciviies. Consequenly Arivé proposed o develop a simulaion model of he conenious issue in a group model building projec. The modeling projec had wo goals: 1. o obain more insigh ino he impac of greenfield consrucion and ransformaion on he success raio and 2. o learn abou he effecs of proposed policy inervenions (Eskinasi, 2002). The op managemen and board of SVH suppored he modeling projec. SGH iniially reaced o he projec wih a mix of curiosiy and skepicism. A comparable iniial clien reacion is also presen in he work of Lane e al. (2003). The direcor of SVH aced as a gaekeeper (G. P. Richardson & Andersen, 1995). The faciliaor and he clien organizaion chose he sevenmember projec group joinly. The projec group consised of wo senior officials (he direcor of SVH and a high ranking civil servan from SGH), hree policy making officials (one from SGH, wo from differen housing corporaions) and a senior housing marke researcher from The Hague Urban Developmen Office. The researcher provided access o many daa sources and conribued o a lively debae on facs and figures. He also had srong opinions on he validiy of he resuling model. The senior officials paricipaed o a lesser exen in he debae on model srucure, figures and conen, bu dominaed he discussion abou policy runs in he final sage of he projec. The modeling eam consised of wo people: he firs auhor in he role of process manager and modeler and one recorder assising he process. Two differen persons fulfilled he role of recorder during he projec. Model building meeings The projec sared in Ocober 2002 and finished in December Ten meeings were held, varying in duraion from four hours or longer for he firs sessions o wo hours in he laer sages of he projec. Mos meeings were held in SVH s office. In all meeings

67 64 chaper IV we used a whieboard o skech model srucure and behavior and we recorded resuls wih a digial camera. The paricipans invesed abou 35 hours in he meeings, wih some of hem occasionally missing ou on a session while ohers spen addiional ime on daa collecion. The oal ime invesmen of he modeling eam is abou 220 hours for he faciliaor and 80 hours for he recorder. The recorder made agendas and meeing minues, which formed he basis for he projec repors. The resuls of he differen phases of he projec were documened in hree progress repors and a final repor (Eskinasi, 2004). We discerned four separae phases in he projec: Sar: sarup and definiion of causal relaions; Model consrucion: daa collecion, quaniaive modeling and model validaion; Simulaion: comparing he effecs of policy opions o he base run; Evaluaion and followup: inerviews, formulaion of conclusions and followup aciviies During he saring phase, we played he Beer Game as an inroducion o sysem dynamics. In he firs meeing we skeched he basic srucure of he housing marke wih he aid of causal loop diagrams. Finally, we carried ou he prees quesionnaire for he empirical evaluaion (described laer on). The repor concluding he firs phase conained he final causal loop diagram. The model consrucion phase encompassed ranslaion of he causal loop diagram ino a sock & flow model. The projec group paricipaed in all meeings by defining and verifying relaions and parameers and by analyzing preliminary model resuls. Mos of he acual modeling work was done offsie. This included crossexaminaion of daa sources, lieraure research, model consrucion, sensiiviy analysis and validaion. The modeling phase was also concluded wih a progress repor. The simulaion phase was carried ou in hree meeings wih he projec group. We esed differen policy opions, including SGH s base run and SVH s alernaive policy. Addiional policy experimens were carried ou in order o improve undersanding of housing marke dynamics and he impac of secondary inervenions. In he las meeing he projec group formulaed is main learning experiences. The progress repor of his phase focused on he differen policy experimens. The evaluaion and followup phase included compilaion of he final repor and several aciviies afer compleion of he main projec. Boh clien organizaions brough heir simulaion experiences o he aenion of oher urban regions and housing corporaions. SVH held wo workshops for policy makers of heir member housing corporaions. The firs workshop was held in March 2004 wih abou 30 aendans. The session sared wih a slide presenaion and demonsraion of he model. Paricipans could hen operae a fligh simulaor version of he model for abou 1530 minues. This group formed he conrol group for assessmen of he effecs of he group model building inervenion. SVH organized a second workshop in May 2004 focusing on he longerm sraegy wih direcors and senior managers of he member corporaions. During his session, he new housing consrucion arges proposed by he Minisry of Housing were simulaed agains he background of possible greenfield developmen locaions. Meanwhile, SGH issued is annual housing and consrucion monioring repor, conaining empirical daa and an updae of exising plans (Haaglanden, 2004b). The repor was exended wih fuure scenarios based on he model

68 Haaglanden 65 simulaions. In November 2004, SVH and SGH invied he Minisry of Housing and he Province of ZuidHolland for a simulaion workshop. They compared four differen greenfield and ransformaion scenarios and debaed he impac on he social housing marke. Finally, Arivé and AEDES, he housing corporaions branch office, organized a session in February 2005 in which housing sraegiss worked wih he fligh simulaor version of he model. IV.4 The resuling model The model consis of four secors: housing consrucion and ransformaion, migraion chains and supply, demand and auxiliary variables and policy response. Afer a descripion of each model secor we use a causal loop diagram o describe feedback relaions. We hen address model behavior and validaion ess. The final par of his secion describes he resuls of policy experimens. Model equaions are available in appendix 3. Housing consrucion and ransformaion secor The firs secor (see figure 15), focuses on consrucion, sale, ransformaion and demoliion of houses. The main srucure consiss of wo producion chains, for social renal housing and one for socalled marke housing. ransformaion program sage 1 ime Social housing under social housing consrucion commisi oned rae of social in consrucion Marke housing under consrucion marke housing commissioned consrucion oal Social housing social housing sock in sage 1 social housing social housing compleed ino sage 1 sale of social housi ng consrucion ime Marke housing sock marke housing marke housing compleed redlined ransformaion rae greenfield and land use change houses rebuil afer ransformaion social housing in sage 2 social housing ino sage 2 prio onmovers avg sage 2 ime marke housing waiing for demoliion marke housing demolished densi y facor oal demolished social housing demolished Figure 15 Haaglanden consrucion and ransformaion secor Marke houses are for insance owner occupied dwellings or renal aparmens owned by commercial invesors. The lef hand side of boh chains depics he sar of he consrucion process. New houses are commissioned, spend some ime in consrucion and are finally compleed and added o he main socks (social housing sock and marke housing sock respecively). An imporan policy parameer is he rae of social housing in new consrucion. Is iniial value of 30% means ha ou of 100 new houses 30 will be social and 70 will be marke houses. The oal volume of new consrucion originaes from hree sources. Firs, he greenfield consrucion program of abou 4,000 new houses per year. A second minor source, change of land use, consiss of abou 800 houses annually, consruced in former indusrial or oher locaions previously no used

69 66 chaper IV for housing. The hird source is houses rebuil afer ransformaion, which closes one of he main feedback loops of he model. Afer commissioning, he average consrucion ime is 1.5 years. The cenral par of his secor consiss of he social housing sock and marke housing sock. A flow from he social ino he marke housing sock depics sale of social housing. This is seen as a useful means o improve he housing mix and o provide houses for enans waning o buy heir renal home. Annually, abou 700 houses are sold. The righ hand side of he producion chain focuses on ransformaion of exising housing. Transformaion is defined as any operaion needing moving ou he enans, largescale reconsrucion works and subsequen allocaion o new enans. In mos cases his means full demoliion and new consrucion, alhough highlevel renovaion is also possible. We ignore smaller repair or renovaion works, because hese do no impac he social housing marke. Wih regard o he ransformaion of social housing, we disinguish wo sages. Houses ener sage 1 (social housing in sage 1) when a housing corporaion decides o end is exploiaion and prepare for ransformaion. Every year abou 2,000 houses, or abou 1.3%, are aken ou of he social housing sock in his way. Transformaion of social housing is one of he crucial policy levers in he model. In sage 1, no more new renal conracs wih unlimied duraion are given ou. In some cases special fixederm renal conracs are given ou o for insance sudens bu in mos cases dwellings say empy. Generally, his encourages oher enans o move ou hemselves. The average sage 1ime is a policy lever because SVH expeced a srong impac on he success raio; is saring value is se a 1.5 years even hough field observaions in specific blocks recorded sage1imes of five years and more. Housing corporaions like long sage 1imes, because his leads o less prioriy housing applicans clogging he allocaion sysem, lowering he success raio and receiving financial compensaion from he associaion. The downurn of long sage1imes, mainly perceived by SGH, is posponemen of new consrucion afer demoliion and a slowing down of he urban revializaion process. The duraion of sage 1 was herefore highly relevan o he dispue beween SGH and SVH and an imporan policy lever in he model. In sage 2, he area is graned special saus under Duch urban renewal legislaion. The remaining regular enans ge prioriy in he housing allocaion sysem, enilemen o financial compensaion and move ou. Noe ha longer sage1imes decrease he number of enans wih prioriy saus, because some of he iniial enans will find a new house by hemselves in he meanime. The average sage2ime is also se a 1.5 years. Is maximum duraion is prescribed in legislaion. Afer sage 2, he social houses are physically demolished and new consrucion can sar. As opposed o sage1ime, he projec group did no consider he sage2ime as a very relevan policy parameer. Afer all, when sage 2 ses in, he consequences for he success raio, i.e. he inflow of new prioriy cases, are ou of he hands of any of he organizaions involved. The ransformaion of marke housing follows a differen process. A proporion of marke housing is redlined and subsequenly demolished. Annually only a negligible 0.175% (ransformaion marke housing rae) of marke houses is redlined and no deemed relevan o he sraegic quesion of SVH and SGH. I is modeled as a firs order maerial

70 Haaglanden 67 delay wih a sock (marke housing waiing for demoliion) and again an average ime before demoliion of 1.5 years. The oal number of houses demolished connecs ransformaion back ino he oal of new consrucion as houses rebuil afer ransformaion oal, aking ino accoun he densiy facor. The iniial value of he densiy facor is 80%, which means ha ou of 100 demolished houses, 80 houses are rebuil. In combinaion wih a 30% rae of social housing in new consrucion, his would resul in 24 new social houses and 56 new marke houses. Supply of social housing secor The second secor in he model (figure 16) conains he supply side of social housing. Migraion or vacancy chains play an imporan par in his model secor. Vacancy chains sar by compleion of new houses. The model conains wo inflowsockouflow srucures o capure vacancy chains sared by social and marke housing respecively. The socks are labeled vacancy chains from mrk running and vacancy chains from soc hsg running. The chains run for an esimaed ime of 1.5 years. This esimae is based on an assumpion by he paricipans in he modeling sessions; no daa were found o verify his esimae. Ending vacancy chains of boh new social and marke housing causes moves wihin he social and marke secor, aking ino accoun he respecive migraion mulipliers 17. Vacancy chains produce moves by enans over he complee course of heir exisence, which is modeled using a coninuous delay wih avg vacancy chain ime as is parameer. Vacancy mulipliers show significan volailiy over ime (Eskinasi, 2004). However, no research was available on he causal relaionships deermining he mulipliers, so we relied on marix algebra (Teule, 1996) o calculae hese from he naional housing needs survey and used hem as an exogenous imeseries inpu: marke o social vacancy muliplier and social o social muliplier. The social housing becoming available from migraion ener ino he sock supply of social housing. We chose a sock insead of an auxiliary, because empy flas can accumulae in siuaions of low demand. Addiional available houses come from he housing allocaion sysem s capabiliy of increasing he share of onmovers leaving social housing (supply from onmovers). This number consiss of real onmovers among prioriy cases and regular onmovers. Oher supply sources include houses from people leaving Haaglanden s social renal secor, people moving ou of Haaglanden or he Neherlands alogeher. Afer he average rening ou ime, he social house is rened ou and flows 17 The housing marke is mainly a marke of secondary properies. The primary supply (new consrucion) invokes a migraion chain ha ounumbers he iniial supply. Suppose we build 100 new houses and 100 families move in, some of hem, le s say 80, from anoher dwelling. These vacan dwellings arac anoher round of families, also leaving homes behind. The chain ends when he final round of available dwellings are occupied by sarers or demolished. Thus our consrucion program causes socalled migraion chains, making he number of moves larger han he number of houses buil. The migraion muliplier is he raio of housing moves o new consrucion: if we assume ha consrucing 100 new houses causes 350 moves in oal, he migraion muliplier is 3.5. In he model wo separae migraion mulipliers are used, one new social and one for new marke housing. Wihin he social housing secor, consrucion of marke housing on average resuls in more migraion han consrucion of social housing. In he model we use Teule s (1996) marix algebra for calculaing migraion mulipliers from housemoving saisics (e.g. from he longiudinal Housing Needs Survey), assuming ha mulipliers are sable over ime. Recen research however shows ha vacancy mulipliers are no consan (Eskinasi, 2004). These rends undermine common policy ideas. The markeosocial migraion muliplier is declining: markehousing consrucion has less and less effec on he social marke, whereas he socialomarke muliplier increases: new renal housing increasingly aracs people from marke housing.

71 68 chaper IV ou of he supply sock. The respecive housing corporaion decides wheher houses are bes suied for sarers or for onmovers. This is refleced by he housing allocaion facor, indicaing he share of available houses deemed fi for onmovers. This facor is iniially 50%, indicaing an even division beween sarers and onmovers. Social vacancy chains saring <social housing compleed> <marke housing compleed> Marke vacancy chains saring Vacancy chains from social running average chain ime Vacancy chains from marke running Social vacancy chains ending Social housing becoming available Vacancies caused by moves in social Marke vacancy chains ending Social vacancy muliplier housing allocaion facor Available social housing Vacancies caused by moves in marke Social housing rened ou supply for sarers <avg renou ime> Marke vacancy muliplier supply for onmovers Figure 16 Haaglanden: supply of social housing secor Demand for social housing secor The hird secor (in figure 17) depics demand for social housing. The secor consiss of four inflowsockouflow srucures for four caegories of house huners: prioriy onmovers, regular onmovers, prioriy sarers and regular sarers. The curren housing allocaion is no a queuing sysem in a sric sense, as people can exer some influence on he waiing ime by being more or less selecive in choosing a new dwelling. The inflow of hree caegories of house huners is driven by exernal parameers only. Again he projec group did no feel confiden o formulae a dynamic hypohesis for hese inflows and esimaed imeseries insead 18. The inflow of prioriy onmovers is also driven by he demoliion of social houses in he firs model secor. I akes ino accoun he effec of a longer sage 1ime in he number of ransformaion prioriy cases. 18 The inflow per year for regular onmovers sars a 10,300 and afer four years has a consan value of 8,000; inflow of oher prioriy onmovers is consan a 1,500. The number of regular sarers is 12,500 a he sar of he simulaion and consan 9,000 afer four years; prioriy sarers are consan a 1,000.

72 Haaglanden 69 new prio onmovers prioriy onmovers renous o prio onmovers new prio sarers prioriy sarers renous o prio sarers <prio onmovers> <supply for onmovers> <supply for sarers> new reg onmovers avg disappm ime regular onmovers renous o reg onmovers disapp onmovers new reg sarers <avg disappm ime> regular sarers renous o reg sarers disapp sarers Figure 17 Haaglanden: demand for social housing secor The socks are depleed by renous o he four caegories of house huner flows. In he supply secor, oal supply was already spli over he sarers and onmovers channels. For boh channels he disribuion logic saes ha prioriy cases go firs, leaving remaining supply for regular cases. Furhermore, regular onmovers can exi he housing allocaion sysem afer an esimaed welveyear socalled disappoinmen ime : house huners will iniially enroll in he disribuion sysem and ry heir chances for some ime. In case hey succeed, hey exi he sock of house huners as new enans. In case hey do no succeed for a long ime (i.e. welve years), hey migh ry o buy an aparmen, ry o find renal housing in anoher urban region, resor o commercial renal dwellings or illegal subleing. In any case, hey exi he housing disribuion sysem unsuccessfully and do no use i anymore. The senior researcher esimaed he duraion and he oher projec group members acceped his as sufficienly valid. In one of he policy experimens, onmovers could also exi he house huner sock by buying a (former) social renal dwelling. The regular channel for sarers does no have urban renewal prioriy cases or a homebuyers exi. Auxiliary variables and policy response secor The fourh and final secor conains he raher sraighforward definiion of several imporan auxiliary variables, mainly used for graphs and ables. These are common knowledge for Duch housing policy expers and include: Success raio = (renous o regular onmovers renous o regular sarers) / (regular onmovers regular sarers). The success raio is he main problem variable in his projec and he reciprocal of he average waiing ime. The perceived success raio is a firs order delay of he success raio and will be used in policy experimens. Also specific success raios for sarers and onmovers exis. Muaion rae = supply of social houses / social housing sock. The muaion rae is a common auxiliary indicaor denoing he dynamics of he housing marke. A low muaion rae indicaes a sagnan marke in which oo few houses become available. Too high a muaion rae may raise fears of long erm srucural vacancies. Share of social housing in oal sock = social housing sock / (social housing sock marke housing sock). This is an auxiliary indicaor for he urban housing mix.

73 70 chaper IV Feedback loops Figure 18 provides an overview of he model feedback srucure. greenfield consrucion consrucion of marke housing B2 oal consrucion R1 consrucion of social housing Marke housing sock share of social housing Social housing sock ransformaion of marke housing B1 gap ransformaion of social housing implici goal supply of vacan social housing B3 R2 vacan housing rened ou B5 waiing lis B4 disappoin men renous o onmovers R3 success raio inflow Figure 18 Haaglanden: feedback srucure The upper middle par of he figure shows he main housing socks (marke housing sock and social housing sock) and he percenage of social housing of he oal number of houses. The cliens sraegy is o reduce he share of social housing o abou 30%. This arge mainly drives he ransformaion volume and o a lesser exen greenfield consrucion and sale of social housing. The upper par of he diagram shows he main feedback loops B1 and B2 driving ransformaion in order o aain he arge share of social housing. B1 works by demolishing social houses, B2 by building new marke housing as a resul of ransformaion. Since par of he ransformaion program resuls in new social housing, a posiive loop R1 is also operaional. The lower lef par of he causal diagram depics he supply side. Consrucion and migraion mulipliers are he main drivers of vacan houses. Available houses being rened ou o people on he waiing lis decreases boh he waiing lis and he number of vacan houses (balancing loop B3). The housing

74 Haaglanden 71 allocaion sysem, however, has a limied capabiliy of increasing he share of onmovers among new enans. Since some of he onmovers move from one social house o anoher, a delayed reinforcing loop R2 comes ino exisence. Finally, he lower righ par of figure 18 depics he demand for social housing. The waiing lis is he cenral variable. Social houses rened ou decreases he waiing lis. Transformaion of social housing increases he waiing lis by vacaing houses o be demolished, and so does he onmover and sarer inflow. A welveyear disappoinmen ime for house huners works on he waiing lis (balancing loop B4). Waiing lis and supply of houses deermine he success raio. In he policy experimens we inroduced a reinforcing loop R3 when a decreasing success raio booss he sarer inflow. IV.5 Validaion ess During is developmen, he model was pu hrough several validaion ess described by Forreser and Senge (1980). We describe he oucome of four ess on model srucure (srucure and parameer verificaion, he exreme condiions and dimensional consisency es) and hree ess on model behavior (he replicaion of he reference mode, behavior sensiiviy es and behavioral anomaly es). Srucure and parameer verificaion were incorporaed in he group model building process. The model srucure is based on angible socks and flows and he projec group s knowledge of he housing marke. Sufficien saisical daa were available on mos socks and flows. Parameers were verified using he same daa sources, excep for some cases, where he projec group made exper guesses consensually. The exreme condiions es was carried ou during he simulaion phase. Simulaion of unrealisically high or low consrucion programs, ransformaion programs and housing allocaion facor resuled in consisen paerns for muaion and he success raio. The dimensional consisency es was carried ou using he builin sofware faciliy. Comparing model oucomes o he reference mode of behavior is a behavior reproducion es. Real life daa were available from 1998 o We compared he real agains he simulaed success raio (figure 19) and scruinized is wo componens, supply of houses (figure 20) and he sock of house huners (figure 21).

75 72 chaper IV Figure 19 Haaglanden: hisorical fi for success raio Figure 20 Haaglanden: hisorical fi for supply of social housing Figure 21 Haaglanden: hisorical fi for sock of house huners

76 Haaglanden 73 The model ges he overall developmen of he success raio, albei ha he observed rend is smoohed ou and ha he simulaed level is somewha higher 19. Scruinizing he supply componen, we found ha he model capures he general movemen and figures, bu misses he observed increase in 1999 and is no very precise in iming. For he house huner componen, he model ges he developmen righ, bu does no capure he 2001 peak. The projec group, however, acceped he sympom generaion behavior as valid. Differences migh be caused by wrong esimaes, bu as menioned before, for some parameers no reliable sources were available whasoever. Moreover, hey relaed he 2001 peak in house huners o he inroducion of an inerne based allocaion sysem, aracing more house huners (We analyzed his issue in one of he boundary adequacy ess). Differences in supply could be relaed o he compleion of several larger consrucion projecs, so realiy had a more discree characer han our simulaion model. The projec group agreed ha such specific evens could no possibly be refleced in he model and were hus o be excluded, i.e. placed ouside he model boundaries. Please keep in mind ha he SGH projec group members were very skepical a he sar wheher we could produce a working model of he housing marke a all. Now ha he model apparenly was capable of producing accepable simulaions, hey were far more anxious o see he policy experimens raher han linger on furher sympom reproducion ess. In he eyes of he projec group, he model did exhibi he behavior experienced in he real sysem. Connecing o our earlier saemen on he modeling perspecive, we could only inerpre his as ha we had succeeded in reaining model plausibiliy (Grüers, 2006) and hus clien saisfacion (Eskinasi & Fokkema, 2006). A comparable experience of cliens moving from iniial skepicism o enhusiasm and model ownership is recorded by Lane e al. (2003). We do no sae ha furher more rigorous validiy ess could no have conribued o a beer model, bu only explain why we reaced as we did, again in hindsigh. The reference mode of behavior of he problem was formulaed in dynamic erms as a causal loop diagram in he saring repor (see figure 18). Bu a he sar of he simulaion projec, SGH and SVH were dispuing he impac of he base and alernaive policy on he housing marke. They would easily agree on he causal srucure of he problem in he saring phase, bu could no come o consensus as how he sysem would reac o boh policies. This indicaes ha hey were only human as i has been repeaedly been demonsraed ha he human mind is no suied for solving highorder dynamic feedback sysems (Forreser, 2007b). And herefore, hey needed a consisen quaniaive sory suppored by model oupu in order o convince hemselves and come o some kind of armisice on he necessary policy inervenions. The sensiiviy analyses performed on exernal parameers are described in he final repor. Mos noably he migraion mulipliers have a srong impac on he success raio in he social housing marke as does he densiy facor. The effec of oher parameers (e.g. sage1ime, several oher delay imes) was very low o negligible. The housing allocaion facor provokes a medium sized response, bu lower han hoped for. 19 Based on six daa poins for each of he hree reference modes, analysis of Theil inequaliy saisics (Serman, 1984; 2000, p. 875) shows ha he majoriy of error is concenraed in unequal covariaion. The bias and variaion componen are below 0.30 and 0.04 respecively, indicaing ha he model capures boh he mean and he rend in he daa well.

77 74 chaper IV Finally, hree boundary adequacy ess focus on loops B1, B5 and R3 in figure 18. All hree experimens sugges ha he addiional loops do no fundamenally aler he dynamics of our simulaion. Closing loop B1 in he model means incorporaing a 30% arge for he share of social housing in he oal sock as a driver of he ransformaion program. I is basically an auomaic pilo for one of he ime series inpus he clien waned. For he ime horizon of our main simulaion, we did no regiser any significan changes in dynamic behavior. In he second experimen, we linked he success raio o he ransformaion program via a woyear percepion delay (closing loop B5 in figure 18). One of he cliens feared ha linking ransformaion inpu o he success raio on he marke would negaively impac ransformaion progress. However, working ou his loop, we found ha is naure is balancing and no reinforcing. Embedding his policy in a feedback loop leads o an overreacion and aggravaes deviaions in he success raio in comparison o he Haaglanden daa. Increasing he policy informaion delay increases his effec. The hird and final experimen concerns he peak in he inflow of new sarer house huners observed around 2001 (see figure 21). A leas wo hypoheses may explain his exra inflow. Lengh of regisraion is he crieria for sarers applying. So if he sarers success raio decreases, more regisraion lengh is needed for geing a renal dwelling. This will encourage oher sarers o regiser for he waiing lis earlier as well. An alernaive hypohesis is ha he inroducion of inerne based housing allocaion increased he consumer base, especially among young sarers eager o use his new medium. Boh hypoheses involve an effec of success raio on inflow and creae reinforcing loop R3 (figure 18). No daa were available on he relaion beween inflow of sarers and he success raio, so we revered o esing wheher observed behavior could be explained by an assumed relaionship. The resuling model behavior maches he observed paern in he daa. We conclude ha in realiy a reinforcing loop beween sarer inflow and sarer success raio may exis. However, since he modeling eam did no feel confiden o include his relaion in he model, loop R5 is no included in he model used for simulaing policy opions. Incorporaing R5 in he model does no change he conclusions of he policy experimens. In conclusion, none of he experimens indicaes ha hese srucural changes significanly aler model behavior, poining o wellchosen model boundaries. The purpose of validaion of sysem dynamics models, afer all, is he gradual building of confidence of he modeling sakeholders, and he Haaglanden projec succeeded in doing so. When he simulaion model was run, is oupu was minuely scruinized by he projec group, mos noably he researcher. In his basic behavior anomaly es, no inconsisencies were found and he projec group acceped he model as valid for is purpose: o simulae differen greenfield and ransformaion policies o assess heir impac on he social housing marke, in paricular he success raio. IV.6 Base run and policy experimens SGH s new housing policy documen (Haaglanden, 2004a) provides he daa for he model base run. Simulaions sar in 1998 and run unil The period up o 2003 is

78 Haaglanden 75 used as a reference o hisoric realiy, while serves as he firs and as he second policy inerval. Please recall ha he cliens main difference of opinion is on he pace of he ransformaion program: SGH proposed an annual ransformaion program of 2,000 social dwellings, while SVH favored o ransform 1,500 dwellings in he firs five years ( ) and 2,500 aferwards ( ). Afer an iniial dip when ransformaion ges moving, supply sars o increase from he greenfield consrucion program. SVH s alernaive policy wih lower ransformaion in period 1 has a small posiive effec from abou 2005 o 2009, bu a significanly larger negaive effec in he second period when lower ransformaion means lower consrucion, shorer migraion chains and less supply. The effec of he 2010 increase o 2,500 maerializes only afer 2017, because of he delay in he housing producion chain. The muaion rae of SVH s alernaive lags behind because less ransformaion means a larger social housing sock. Lower ransformaion means less urban renewal prioriy house huners so ha he base run has slighly more house huners piled up around Boh effecs combined resul in a large difference in success raio beween boh policies, as is shown in figure 22. Figure 22 Haaglanden: base run & alernaive policy Transformaion has he following effecs on he success raio. Firs, i increases pressure on he housing marke by creaing more urban renewal prioriy house huners, hus lowering changes for regulars. Second, ransformaion decreases he social housing sock, lowering he supply of vacan houses by a small amoun. In he long run, afer he enire ransformaion pipeline, compleion of new houses ses migraion chains in moion and increases supply. New consrucion afer ransformaion, muliplied by he migraion muliplier, ounumbers he ransformaion volume imes he urban renewal prioriy rae. Thus lowering ransformaion has advanages in he shor run, bu leads o even larger disadvanages in he long run. The supply side works as a maerial delay: i pospones and smoohes he impac of ransformaion and greenfield consrucion. The demand side is basically abou accumulaion of house huners. So alhough posponing consrucion ulimaely resuls in he same amoun of consrucion oupu and supply,

79 76 chaper IV he accumulaion of house huners in beween ses he success raio back unrecoverably. Finally, several mulipliers influence he overall balance, he mos imporan being he volaile migraion muliplier and he ransformaion prioriy rae. Combined wih he behavior of he producion srucure, his causes he cosbeforeprofi behavior signaled by he projec group. The projec s final repor (Eskinasi, 2004) describes en simulaions made during he main projec and focuses on resolving he differen opinions on he righ pace of housing ransformaion. Seven ou of en simulaions delve ino his issue by varying greenfield consrucion and ransformaion. New greenfield consrucion came ou as an imporan lever. I has a relaively shor delay ime and a high impac. Due o he migraion muliplier, significanly lower greenfield developmen is very derimenal o he success raio. Differen ime series of ransformaion inpu (he policy sandpoins of SGH and SVH) affec he success raio as described above. The second policy experimen esed he leverage of he housing allocaion sysem on he success raio, assumed mainly by SVH. Increased allocaion o onmovers was combined wih lower greenfield developmen. Compensaing low consrucion volumes by smar housing allocaion was no possible o he expeced exen. As a resul he projec group fel ha housing allocaion was no he righ means o boos he success raio. A hird policy opion was direcing sale of social houses o house huners on he waiing lis. This enables rapid ransformaion of he housing sock wihou any inerfering delays. I also direcly decreases he accumulaion of house huners and herefore heoreically has a srong leverage on he success raio: fewer remaining house huners compee for approximaely he same amoun of supply. The simulaions confirmed his hough. However, he 1990 s rapid price increases make i virually impossible for lower income groups o buy even a former social fla and for middle income groups his is becoming increasingly difficul as well. This opion is very difficul o pu ino pracice. The fourh opion ha was esed was increasing he share of social housing in oal consrucion. Is resul is a higher success raio due o lower accumulaion of house huners. This policy may appear o go direcly agains he implici arge of decreasing he social sock. Neverheless i has a range of oher favorable oucomes, such as increasing housing qualiy, lowering concenraion of social housing wihin he cenral ciy and o increase is share in he predominanly marke housing suburbs. To sum up, he policy experimens clarified insighs on ransformaion, greenfield developmen and he share of social houses in consrucion, confirmed paricipans expecaions on sale of houses and refued inuiions on he housing allocaion sysem. IV.7 Evaluaion of he projec Opinion of he projec group Boh paricipaing organizaions acceped he model oucomes as valid and he overall conclusions as very insighful. The debae swiched from a focus on he facs concerning he impac of ransformaion on he housing marke, o he preferences of and he differen ineress of boh organizaions. The sysem dynamics inervenion has demessed

80 Haaglanden 77 he problem. In inerviews afer projec compleion, he projec group repored he following overall conclusions and learning experiences drawn from he simulaion projec: A beer undersanding of he effec of he several delays in he sysem. Before he simulaion projec, he impac of delays was underesimaed Increased insigh ino differences beween shor erm and long erm effecs of housing ransformaion. The model clearly demonsraed shor erm and long erm effecs ha differed in magniude and direcion. These effecs wen agains shared beliefs and consiued an imporan counerinuiive finding. The high leverage of new greenfield consrucion, he share of social houses in consrucion and sale was reconfirmed, albei ha he migraion muliplier had a sronger influence han expeced and varied over ime. The leverage of he housing allocaion sysem was far lower han some group members expeced or hoped for. The differences beween long erm and shor erm effecs of policy inervenions. Some empirical assessmens of projec effeciveness Furhermore, during he projec, we carried ou some empirical assessmens, repored in more deail in Eskinasi and Rouwee (2004). On he basis of he mehod developed by Rouwee (2003), we assessed wheher he modeling inervenion impaced he aiude of projec group members owards cerain policy inervenions. Comparing he projec group o he paricipans in he fligh simulaor workshops, we concluded ha only full paricipaion in he projec changes a person s aiudes owards he policy measures proposed above. Boh he projec group and he workshop paricipans, albei exposed o very differen doses of sysem dynamics modeling, agree ha working wih simulaion models creaes beer and faser alignmen of menal models han regular meeings. Opinion of he modeling eam From a conen poin of view, he model cerainly leaves room for furher improvemen. We did no model he ineracion beween he marke and social housing secor, whereas his is a very imporan issue explaining he difficulies in he Duch housing marke (Conijn, 2006). Qualiy improvemens wihin he social secor are no refleced alhough he housing marke balance is disincly differen for old and new social housing. As discussed above, in many cases he projec group did no feel confiden o formulae loops surrounding he arge share of social housing, he ransformaion program and inflow of he waiing liss. Finding usable daa on hese relaions would have been very difficul. The same applies o he causal relaions around he migraion mulipliers. In hese aspecs, here is cerainly room for furher research wih sysem dynamics modeling ino he inricacies of he Duch housing marke. And how did we do as sysem dynamics modelers? Did we, a leas o some degree, mee Forreser s sandards for qualiy of work in he sysem dynamics field (Forreser, 2007b)? Firs of all, he projec was iniiaed by wo clien organizaions in dispue over angible real world policy problems. We did succeed o some degree o provide hem wih more solid insighs ino he dynamic behavior of heir problem. We delved ino he hisory and deails of he problem a hand and developed a relaively compac model showing he causes of he observed difficuly. The model used during he simulaions was driven by

81 78 chaper IV exernal ime series, bu he boundary adequacy ess indicae ha his is no a serious flaw as regards o dynamic behavior during he policy ime frame. I may have been he case ha he modeling faciliaor was no ye skilled enough o ranslae he ime series hinking of he projec group members ino an equivalen loop srucure. We did no arrive a a high leverage policy which fundamenally alers he dynamic behavior, basically because our cliens were mainly concerned on seling heir dispue. They agreed ha he sysem dynamics modeling inervenion succeeded in his aspec and ha i provided hem wih highly relevan learning experiences abou he dynamic behavior of he Haaglanden social housing marke. IV.8 Conclusions The Haaglanden group model building projec has produced several ineresing and angible resuls. Firs of all, is conex indicaes a shif in he basic ideas of urban renewal policy in he Neherlands. Whereas in he 1970 s urban renewal was focused on echnical improvemen of he housing sock, he programs emphasize balancing of he urban populaion and housing mix in order o improve urban socioeconomic vialiy, a cenral issue in Urban Dynamics (Forreser, 1969). Second, he group model building inervenion has helped he clien organizaions in seling a highly conenious issue, i.e. he bes pace of ransformaion of social housing. In he inerviews afer he projec, paricipans menion several new insighs ino he behavior of he housing marke, such as he impac of delays, he accumulaion on he waiing lis and he effeciveness of several alernaive policy opions. Third, we found ha all paricipans see sysem dynamics modeling as a beer means of decision making han regular meeings, bu ha only wih people deeply involved, group model building has a significan impac on he aiude owards policy opions. Finally, we summarized several possible improvemens and quesions for furher research and refleced on our resuls in he perspecive of Foresers (2007b) sandards for good qualiy sysem dynamics modeling. The modeling projec helped our clien organizaions o improve undersanding of he maer a hand and o solve heir policy conflic. The projec has been useful in ha respec.

82 V Houdini 20 V.1 Inroducion This chaper describes he work on progress on Houdini, a sysem dynamics model used for explaining regional divergence and he impac of insiuional feaures on he developmen of house prices and consrucion. Houdini has is foundaions in he four quadran model of Di Pasquale and Wheaon (1996) (furher: 4QM). The modificaions of he 4QM described in chaper II.2 (ren regulaion, fiscal morgage suppor, land use planning and residual land prices) were firs applied in Houdini. Houdini is based on he exensive criical lieraure advocaing reforms of he Duch housing marke and inheried conceps from exising economic models of he housing marke (Donders e al., 2010; Romijn & Besseling, 2008; Van Ewijk, Koning, Lever, & De Mooij, 2006). Houdini was reviewed by a wellesablished exper panel, consising of represenaives of universiies, minisries and naional research and policy analysis insiues. This chaper repors Houdini s srucure, validaion, base run, policy experimens and follow up aciviies. V.2 Conex of he sysem dynamics modeling projec Hisory of he problem conex From he Reconsrucion Period afer World War II onwards, Duch housing policy was led by governmen policies raher han by marke principles. Eskinasi e al. (2009) ell he ale of differen sae housing policy approaches from pos war mass housing provision hrough he 1970s new owns and urban rehabiliaion for low income groups o he 1990s and early 2000s, when housing policies sared o pay lip service o marke principles. Housing associaions were privaized o a cerain exen, consumer demand became more imporan for new consrucion and socioeconomic revializaion became a prime objecive of urban renewal. Bu more fundamenal reforms of he housing marke were posponed: ren regulaion, morgage ineres ax reducions and spaial planning were sill in place. The 1990s winessed decreasing morgage ineres raes, growing incomes and improved availabiliy of morgage credis for households. Average house prices increased from under in 1995 o nearly in Measuring house prices in muliudes of median incomes, in 2007, inhabian Eindhoven oupriced New York and Amserdam was he 13h mos expensive worldwide (Romijn & Besseling, 2008, p. 27). Households and he sae budge were increasingly a risk from high morgages and ineres flucuaions, ren regulaion discouraged commercial invesors o build 20 This chaper is based on: Eskinasi, M., Rouwee, E., & Vennix, J. (2011). Houdini: a sysem dynamics model for housing marke reforms. Paper presened a he 29h Inernaional Sysem Dynamics Conference, Washingon DC. And: Eskinasi, M. (2011). Houdini: een syseemdynamische modellering van regionale woningmarken. Achergrondsudies (pp. 56). Den Haag: Planbureau voor de Leefomgeving.

83 80 chaper V renal housing and enans o move from cheap aparmens, hus obsrucing housing marke dynamics. I was esimaed ha he sae and he housing associaion direcly and indirecly subsidize housing wih 29 billion annually (Don, 2008, p. 3). The balancing feedback loop of he 4QM, however, should make high prices boos new consrucion. Duch consrucion saisics, however, show ever decreasing consrucion volumes from 1990 onwards. The spaial planning sysem was seen a probable culpri (Besseling e al., 2008). The Balkenende IV coaliion governmen ( ) wih Chrisian democras advocaing homeowners ineress and social democras relucan o ease ren regulaion made a compromise o once more pospone fundamenal housing marke reforms. This moraorium spurred an unprecedened sream of economic sudies, mosly very criical of he sae housing and planning policy, demonsraing many negaive effecs and calling for fundamenal reforms, (e.g. Conijn, 2006; Don, 2008; Hof, Koopmans, & Teulings, 2006) and ohers. The Minisry of Housing, he radiional sronghold of he planners inervenionis paradigm was forced ino he defense before he new conservaive Rue I governmen dismanled i in The new governmen se ou o decenralize housing and spaial planning o provinces and municipaliies and o ighen he fiscal and legal leashes for housing associaions. V.3 The sysem dynamics modeling projec The developmen of Houdini sared in 2008 as a privae projec ou of ineres for he subsanive maer and caugh he ineres of a leading academic for is prospecs of generaing insighs ino ransiion pahs owards a more sable housing marke. A his sage, only limied ime could be invesed in Houdini, bu as much lieraure was available 21, a simple prooype was buil and producing plausible firs resuls near he end of Houdini was brough o he aenion of PBL Neherlands Environmenal Assessmen Agency. Is saff was working on a large scale demographic syle housing marke model on he municipal level. The prospec of using (pars of) Houdini for his large model made PBL hire he modeler. As PBL is concerned mainly wih regional housing markes, i was decided o ranslae Houdini ino a regional model. The prooype saw many improvemens and finally evolved ino he firs fully documened and validaed version of Houdini (Eskinasi, 2011a). Three runs were made on basis of regional housing marke daa. Region A represens he naional average, region B is he densely populaed norhern par of he Randsad around Amserdam, and Urech, region C is he declining far souheas of Limburg. Furhermore, differen policy experimens were carried ou, focusing on reducing ren regulaion, morgage ax deducion and inervenions in he spaial planning sysem and comparing differences for he hree regions. Feedback beween (adjacen) regional housing markes was missing in he firs generaion of Houdini. 21 Mos noably modeling sudies of he Economic Assessmen Agency on he renal (Romijn & Besseling, 2008) and owner occupied (Van Ewijk e al., 2006) secors, a household behavioral model (Ras, Eggink, Van Gameren, & Ooms, 2006), an anhology of housing marke criiques (Don, 2008) and he 4QM (Di Pasquale & Wheaon, 1996).

84 Houdini 81 An exper panel was formed o provide guidance o he modeling projec. I consised of wo leading housing academics, housing expers from CPB, SCP 22 and he Economic Research Insiue of he Consrucion Indusry (EIB), policy makers from he (former) Minisry of Housing and exper saff from PBL. A firs plenary session was held in June 2010, when work on he firs version was in full swing. Laer on, he modeler regularly conaced members of he exper panel for advice and feedback. The shif in purpose of he model should be noed: he iniial purpose was o model he problemaic housing marke behavior on he naional level. A ha ime, CPB only jus published heir own naional housing marke model (Donders e al., 2010), based on he inegraion of wo separae sudies of he owner occupied (Van Ewijk e al., 2006) and he renal secor (Romijn & Besseling, 2008). Only wih PBL hiring he Houdini modeler, he regional aspec was added o he modeling purpose. Possibly, his saved Houdini from a compeing model issue wih he CPB housing marke model (Donders e al., 2010; Eskinasi & Fokkema, 2006), bu necessiaes fuure fundamenal rehinking of he model s focus as regional ineracions need o be added. Several small sudy models were made for his purpose, bu he sar of he middle incomes and morgages modeling projecs drew away aenion from furher developmen of spaial versions of Houdini. V.4 The resuling model Overview of he model and general aspecs Houdini is based on he 4QM and is he iniial source of he insiuional feaures added o he 4QM in illusraions 1 and 2 in secion II.2, i.e. sock and flow srucures for populaion and incomes (in order o aain uni consisency), ren regulaion, fiscal morgage suppor, land use planning and residual land pricing. Furhermore, Houdini has a double housing producion chain for owneroccupied and renal housing respecively. This srucure is somewha comparable o Haaglanden (see figure 15), bu adds he ren and price variables of he 4QM. The ren axis in he 4QM is defined here in erms of user coss: he real economic coss for using real esae, he sandard approach in he naional and inernaional housing economic lieraure (e.g. Di Pasquale & Wheaon, 1996; Poerba, 1984). User cos heory akes ino accoun hree componens: mainenance and oher coss, financing coss and housing appreciaion. Di Pasquale and Wheaon (1996, pp ) discern hree varians how households ake appreciaion ino consideraion. Wih exogenous expecaions, home owners do no base price expecaions on housing marke rends, bu on e.g. inflaion or general GDP growh. Raional price expecaions, he sandard in mainsream economics, allows households o correcly forecas fuure ime rajecories of a marke afer an exogenous shock occurred 23. Adapive or myopic expecaions, finally assume ha households ake ino accoun hisoric price increases 22 CPB Neherlands Bureau for Economic Policy Analysis deals wih economic aspecs of many policy fields, mosly on he naional level. PBL Neherlands Environmenal Assessmen Agency deals wih environmenal issues, land use, agriculure and food qualiy, waer managemen, regional developmen, regional economies and housing markes. SCP is he Neherlands Insiue for Social Research, aking mosly he household viewpoin. All hree agencies herefore work on housing marke models and sudies bu someimes have differen viewpoins and opinions. 23 They are no expeced o correcly forecas he occurrence of he acual exogenous shocks, as is suggesed in some criicisms of raional expecaions.

85 82 chaper V for curren decision making. Adapive expecaions are a precondiion for he occurrence of real esae cycles (Wheaon, 1999) and inroduce he reinforcing loop of speculaive incenive in figure 12. The exac specificaion of he appreciaion componen proved a grea opporuniy for debae wih he mainsream housing economiss from he CPB. Moreover, Houdini models he peculiar inerplay of prices beween he owneroccupied secor (wih prices simulaed by fiscal morgage suppor) and especially he social renal secor wih ren regulaion. Houdini also adds a dynamic srucure for household enure choice and fiscal feedback. Finally, houses are heerogeneous as o size, qualiy, ameniies and locaion. Housing economic lieraure defines he housing sock in erms of absrac housing services or qualiy unis. Larger or beer houses (housing srucures or housing unis) hen provide more housing services han smaller ones. Economiss criicize planners for overemphasizing housing unis and demographic prognosis and underesimaing demand for housing qualiy based on income growh (Eichholz & Lindenhal, 2008, p. 80) and he negaive welfare effecs of all governmen inervenions. Main model srucure Figure 23 shows he producion chain of he Houdini model, wih deails removed for clariy s sake. Loops B1 and B2 represen he main balancing loop of he 4QM for he owner occupied and renal secor respecively. Likewise, R1 and R2 consiue he residual land prices loops for boh secors. Furhermore, a flow variable is added represening he sale of renal housing ino owner occupaion. The inerconnecion of owner occupied and renal housing hrough housing prices generaes ineresing dynamics. The price dynamics of he owner occupied secor (influenced by exogenous demand and ineres raes) ransfer o he renal secor. When sold, renal dwellings will collec he same freemarke, peruni prices as owneroccupied housing (aking ino accoun of course differences in size, qualiy ec.). On he oher hand, he invesmen value of renal housing is based roughly on capialized regulaed rens 24. Owneroccupied prices are simulaed by fiscal morgage suppor, bu renal invesmen values conrolled by ren regulaions are far lower han he freemarke values (Conijn & Schilder, 2009). The difference beween (simulaed) marke value and renal invesmen value is he socalled value gap, which depresses renal consrucion and simulaes sale of renal housing. Balancing loop B3 will srive o equalize ou price differences hrough he sale of renal housing, bu as long as fiscal morgage suppor arificially lowers he effecive discoun facor in he owner occupied secor only, he sysem is no capable of aaining equilibrium. 24 A fixed capializaion rae was used here, based on Donders e al. (2010)

86 Houdini 83 new zoning ownocc zoned cap for onwocc profi ownocc devm coss R1 ownocc sared chng dev cos ownocc under consrucion B1 price ownocc shadowprice renal ownocc finished user cos ownocc B3 ownocc sock ownocc demoliion sale of renal housing R2 value gap invesmen value renal B2 user cos renal new zoning renal zoned cap for renal renal sared renal under consrucion renal finished renal sock renal demoliion Figure 23 Houdini: price ineracion beween renal and owner occupied secor Figure 24 shows he addiional feedback loops governing he dynamics of enure choice, subsidies and fiscal pressure on household incomes. Tenure choice is operaionalized here as he fracion of households choosing owneroccupaion. The income disribuion of owneroccupaion, however, is very skewed as fiscal morgage suppor is mos advanageous for higher income groups and renal housing allowances are limied o lower income groups. Ren regulaion applies for all income groups in renal housing and works as an addiional, implici subsidy financed hrough he lowered invesmen yields of housing associaions (Conijn, 2008; Conijn & Schilder, 2009). Bu on he overall level, boh insrumens add up o he oal fiscal pressure on incomes: housing subsidies and ax benefis influence peoples decisions on enure. Balancing loop B4 represens households consideraions on he affordabiliy of owneroccupied housing. Loop B5 is dorman, as ren regulaion prevens increased demand for renal housing o propagae ino higher user coss. Likewise, ren regulaion prevens he sae expendiure on direc renal housing o spiral ou of conrol hrough loop R3, which is acually being used as an argumen o preserve ren regulaion.. Loop B6 shows how fiscal morgage suppor (i.e. subsidy for owner occupied housing) increases income axes and in heory lowers enure choice probabiliy. Due o progressive income axes, however, higher income households benefi more from fiscal morgage suppor.

87 84 chaper V fiscal pressure subsidy for ownocc subsidy for renal user cos ownocc B6 income R3 demand for ownocc B4 enure choice user cos renal demand for renal B5 Figure 24 Houdini: fiscal and enure choice dynamics Modeling deails On he demand side, Houdini akes ino accoun he possibiliy of populaion decline. The household growh rae (see figure 25, righ hand side) is a linear funcion of ime, wih is parameers esimaed on he basis of a demographic prognosis. Is slope is negaive, so populaion growh decreases and a a given momen, even populaion decline will occur.. The hree regions in Eskinasi (2011a) were chosen in order o display large variaion in populaion growh: region C was already declining in 2010, whereas region B would coninue grow for nearly a cenury. In average region A, populaion growh decays o near zero a he end of he simulaion period, i.e. 50 years. Income growh akes hisorical figures and fuure scenarios as exogenous inpu. The dynamics of he enure choice variable in figure 24 is he subjec of ineresing debae. The firs version of Houdini used a saisical esimae (Ras e al., 2006) where enure choice depends in household incomes, a regional (consan) facor and he raio of user coss in boh secors 25. This soluion gave enure choice a srong endogenous characer, bu also caused uni consisency flaws and debaes wih he CPB economiss paricipaing in he projec. They proposed a uni consisen soluion adhering o microeconomic foundaions wih a fixed budge share for housing and a fixed enure preference, based on a nesed CobbDouglas uiliy funcion wih budge consrains. This soluion was based, however, on a model wih absrac housing services raher han concree heerogeneous housing unis. This lef Houdini wih sill unresolved ension 25 Several oher variables used in he regression were no used in Houdini, like ehniciy, educaion level ec.

88 Houdini 85 beween he planners and he economiss paradigm. Linking one house (or housing uni) o one household is a fundamenal cornersone of housing planners hinking. Vennix (1996, p ) also encounered his issue. Economiss, however, criicize exacly his poin of he planners docrine: i is no flexible enough o accommodae changes in demand for housing qualiy. Housing researchers, on he oher hand, poin a he fac ha also such common uiliy funcions are sill oo sylized o represen real housing marke processes (Maclennan, 2012) and ha hese underesimae he impac of macroconex facors resricing household choices in he prevalen dynamic lifecourse approach (Clark, 2012). Such debaes on he naure of housing choice is a proper example of a socalled messy problem, where differen sakeholders (or disciplines) hold very differen perspecives on a paricular noion (Vennix, 1996, p. 13). In any case, he firs version of Houdini had no ye reached a fully saisfacory modeling soluion on he issue of enure choice and housing qualiy when aenion moved owards he middle income projec in chaper VI. Consrucion coss Profi Consrucion sared Ineres rae Price B1 Housing under consrucion Budge share for housing Consrucion finished Ren Housing sock Income Households Demoliion income change new households income growh household growh Consrucion ime Life ime Figure 25 Houdini: household and income dynamics in he 4QM The second illusraion in secion II.2 demonsraed a simple model modificaion for inegraing land use planning (see figure 7) as an exension of he housing producion chain. Houdini feaures a comparable srucure wih zoned capaciy socks for owneroccupied and renal housing respecively (see figure 23). The inflow o hese socks is governed by household growh: land use planning agencies also use he demographic forecass on which he exogenous parameers for household growh is based. They esimae he fuure number of households a a given ime horizon (e.g. 15 years ahead), deermine he expeced housing shorage and allocae sufficien annual slices of zoned capaciy. Disribuion of capaciy over secors is based on he enure choice variable discussed above. Jus like he realworld sysem, his srucure disregards he influence of income growh and ineres raes in housing demand, as Eichholz and Lindenhal (2008) argue. Houdini also feaures he mechanism of residual land prices (see figure 23, loops R1 and R2). As argued in secion II.2, residual land prices make developmen coss adjus

89 86 chaper V dynamically o marke prices for housing. Brick and morar consrucion coss grew only moderaely (Besseling e al., 2008) from he 1970s onwards, so land prices absorb he remaining share of house price increases. Rouwendal and Vermeulen (2007) demonsraed empirically ha in he Neherlands, housing consrucion hardly reacs o prices on he shor and medium run, due o he complexiies of land use planning and residual land pricing. Buielaar (2010) explains how Duch municipaliies sared o pursue acive land policies based on residual prices. This enabled hem o recover invesmen coss made and o capure par of he very lucraive profis made in greenfield developmen. I is herefore plausible o assume ha he residual land price srucure (see figure 9) applies for he Neherlands 26. Furhermore, a high level of marke concenraion characerizes he Duch developmen and consrucion indusry (Buielaar & Pouls, 2009): a small number of large firms dominaes he marke and owns mos land o be zoned for residenial developmen, especially in he densely populaed Randsad. This socalled Courno oligopoly (Varian, 1992) allows developers o capure a higher profi rae, deermined by marke concenraion and price elasiciy of demand. In Houdini, his aspec of he Duch consrucion marke is inegraed ino he relaion beween profis and consrucion oupu (see figure 23). Figure 24 shows he feedback beween fiscal morgage suppor and renal housing allowances o he average household income level. Lowering public expendiure may harm demand for housing, bu can be (fully or parially) compensaed by decreasing axes, socalled back funneling. The firs version of Houdini also conains several oher axaion mechanisms for policy experimens, e.g. axing deregulaed rens 27 or sale of renal housing. Finally, Houdini has several oucome raios like house prices, user coss o income, housing shorage (households minus sock) and he percenage of housing subsidies o naional income. V.5 Validaion ess Several validaion ess were carried ou wih he model (Forreser & Senge, 1980). Boundary adequacy ess were no ye carried ou in his sage, bu may be based on commens received from he exper panel. Srucure and parameer verificaion were based on exising housing lieraure and saisical sources. Dimensional consisency was safeguarded wih he modeling sofware and is correc. As menioned before, he demand equaions were mos problemaic in his respec. As for behavior reproducion, he simulaion was esed agains saisical daa over wih Theil s inequaliy saisics (Serman, 1984). Figure 26 shows he reference mode of behavior and simulaion resuls for housing prices and new consrucion. The model is quie precise as o housing sock developmen a a 1% RMSPE error and has accepable saisics for price developmen 28. Housing supply is noably difficul o model 26 Iniially, he modeler also assumed ha oversupply of zoned land would pu downward pressure on he developmen cos: wih surplus supply, land would become cheaper. Land marke expers quesioned his loop because mos land is already owned by developers. 27 Such a measure was acually implemened early 2013, much o he dissen of housing associaions. 28 RSMPE 5% & U c 0,85 (Eskinasi, 2011a, p. 62).

90 Houdini 87 (Di Pasquale, 1999) and leaves room for improvemen here as well 29 : Houdini misses he upswing of consrucion from 2004 onwards, swings furher down unil 2009, when acual consrucion declined due o he credi crunch. The upswing of acual consrucion is sronges in he renal secor from 2006, when housing associaions inensified heir effors. This is no ye concepually refleced in he model. Reference mode of behavior and baseline simulaion House prices x Consrucion x houses Prices acual Prices simulaed Consrucion acual Consrucion simulaed Year Figure 26 Houdini: reference mode & hisorical fi Several parameer sensiiviy analyses were run using sofware faciliies. The sensiiviy of he model o capial marke ineres rae reflecs well documened responses of real housing markes: increasing ineres raes decrease house prices and make consrucion collapse. Also varying household income growh yields recognizable responses. The response of he model o price and especially income elasiciy of demand is difficul o inerpre. This confirms he uni consisency es in he sense ha he demand secion is poined ou as a concepually weak poin in he model. Sensiiviy analysis wih he ime offse of he planning sysem, albei farfeched a firs sigh, yields a proper sysem dynamics counerinuiive insigh: a longer offse has he planning sysem anicipae earlier on fuure populaion decline. Fewer houses are buil when demand is sill growing so shorages and prices increase. In he regulaed siuaion, his empers demand so much ha he qualiy fi of demand and supply improves. A shor ime offse causes he opposie effec in ha he planning sysem produces more in he firs years, leading o over supply when populaion declines. This resuls somewha resembles he findings of Glaeser e al. (2008) ha ample supply elasiciy can also be derimenal o overall welfare hrough overproducion of housing. 29 RSMPE 13% & U c 0,85 for owner occupied and RSMPE 25% & U c = 0,95 for renal (Eskinasi, 2011a, p. 62).

91 88 chaper V V.6 Base run and policy experimens The base runs in figure 27 and figure 28 show he long run effec of unchanged housing policies on he hree regions. Saring year 0 equals 1995 and he simulaion runs for 50 years. Region A represens he naional average, region B is he densely populaed norhern par of he Randsad around Amserdam, and Urech, region C is he declining far souheas of Limburg. B has a higher and C a lower income growh raio han average A. In Region A, populaion growh slows down and reaches 0 near year 50. Region B keeps growing hroughou he enire simulaion horizon, bu he populaion of region C declines from year 15 already. The 1990s saw significan decreases in morgage ineres raes, wih very limied regional differeniaion. For he long erm, a fixed assumpion of ineres level was made of 3% in real erms. The simulaion shows he recognizable rapid growh of prices in all hree regions. Differences in income growh, demography and saring siuaion of he housing sock (prices, raio of renal) explain differen growh curves. A bi of speculaive incenive sneaks ino households decision making and price increases are simulaed. Bu economic growh slows down (income growh decreases and ineres rae climbs) and house price growh levels off quie suddenly. Developmen coss used o lag o house price developmen, bu now rapidly cach up, decimaing profiabiliy and consrucion. When he sysem recovers from his exernal shock (ineres raes and income growh sabilize, demography slows down), i is effecively exhibiing zero real growh in house prices in region A, reurning growh in B and acceleraing decline in C. This closely maches he reference mode of behavior as described above. Consrucion recovers in A and B, bu no in C. 350 Simulaed house prices 300 House prices x Region A Region B Region C Simulaion ime (years) Figure 27 Houdini: base line simulaion of house prices for hree regions

92 Houdini 89 Simulaed consrucion 1,8% 1,6% 1,4% % of oal housing sock 1,2% 1,0% 0,8% 0,6% 0,4% Region A Region B Region C 0,2% 0,0% Simulaion ime (years) Figure 28 Houdini: base line simulaion of consrucion for hree regions Policy experimens were carried ou from year 20, focusing on he morgage ineres ax reducion and ren regulaion. Ren deregulaion is simulaed by allowing higher ren increases. Causally speaking, he hindrances o balancing loop B2 are gradually lifed. Rens and asse prices rise, bu shifs he balance in he enure choice loops owards owner occupied housing. Rens will grow unil marke rens are reached when invesors have a cerain reurn level on he asse price of renal dwellings. Because growing region B has relaively high house prices, i akes a longer ime o reach marke rens han in average region A. Likewise, declining region C wih lower prices reaches marke rens earlier. Higher rens lead o a shif in enure choice owards owner occupied housing, increasing boh price and consrucion levels in i. Decreasing he morgage ax reducion from an average 25% o 15% in year 20 leads o somewha lower prices. In causal erms, lower morgage ax reducions increase he effecive ineres rae. Consrucion responds dramaically in he shor erm and shrinks 40% relaive o he baseline simulaions. Because he growh of he housing sock sops and demand coninues o grow, he iniial price loss is compensaed o some exen in he medium erm. On he longer erm, he marke prices sabilize only jus under he level of he baseline. Region B and C respond similarly, albei wih consrucion in declining region C coming o a complee hal by year 15. Combining boh experimens shows ha he effec more or less add up. Higher rens shif demand o owner occupied housing. Reducing morgage ax reducions decreases house prices and consrucion of owner occupied housing even more. The ransiion ime of regulaed rens o marke rens, however, is shorened: lower house prices lower marke rens as well.

93 90 chaper V V.7 Follow up aciviies and reacions o Houdini The modeling repor was shared wih he members of he exper panel and several meeings were held o obain heir feedback of he model srucure and model oucomes. The firs meeing of he exper panel in June 2010 yielded suggesions as o he relevance of housing qualiy, he necessiy of regional ineracions, modeling simpliciy and he imporance of a wellchosen base line simulaion. Wihin PBL, Houdini was received posiively, bu no very argeed feedback was provided. This may be caused by he lack of a sufficienly specific purpose for Houdini. One surveyor, generally criical of large scale modeling, however, found he sysem dynamics approach in comparison more aracive as i incorporaes behavioral responses of acors and suppor whaif policy experimens. Simulaion of fuure house prices neverheless caused some nervousness. A draf aricle conaining price graphs for a professional magazine was posponed awaiing furher suppor of wellesablished academics because of poenial fuzz wih naional and local policy makers. The same aiude owards he price graphs was found wih he CPB housing economiss in he exper panel. Their model would only show he deviaion of policy experimens from he base pah. Moreover, he CPB housing economiss conribued o a srong bu consrucive debae on he underlying principles of Houdini. Firs, hey criicized he lack of economic rigor on he demand side: Houdini has no explici bookkeeping of expendiure on housing. As menioned before, he demand equaions were weak as i comes o meaningful unis. The CPB expers suggesed using a behavioral sysem of housing consumers consising of a budge consrain and maximizaion of uiliy. These suggesions provided a clear framework for modeling demand wih sraighforward equaions in comprehensible unis and will be implemened in a nex model version. The adapive price expecaions were mos conroversial. Nowihsanding some empirical suppor for adapive price expecaions in he housing marke, (e.g.case & Shiller, 1989; Glaeser, 2013; Hamilon & Schwab, 1985), raional expecaions are axiomaic in mainsream economics. And wih CPB mainsream economiss as parners in he modeling projec, his issue was a hurdle o ake in building confidence in he model (Forreser & Senge, 1980). On he oher hand, sraying from axiomaic perfec compeiion in reference o he srucure of he Duch supply side (wih planning sysem and oligopoly) provoked quesions of clarificaion raher han an axiomaic debae. Overall, CPB is supporive of Houdini, in paricular wih regard o he regional differences and ineracions and explici modeling of he planning sysem. Houdini was pu o he es in a projec on long erm spaial scenarios for he Neherlands. A land use ranspor ineracion model, TIGRIS XL (Zondag & De Jong, 2011) provides he main quaniaive framework. I is a large scale model of employmen, ranspor, housing and oher land use. I does no explicily model house prices. Boh Houdini and TIGRIS used inpus from several demographic and economic scenarios. Boh Houdini and TIGRIS simulaed new housing consrucion. A sufficien fi beween boh models in erms of Theil saisics hen allowed he house price oupu of Houdini o be acceped for he projec. Furhermore, daa colleced for Houdini conribued o he final projec repor (Hilbers e al., 2011).

94 Houdini 91 V.8 Evaluaion of he projec Opinion of he exper panel and he modeler Houdini 1.0 is only he saring poin, so no final conclusions can be formulaed a his poin. The impac of Houdini 1.0 is limied, because is purpose is no well defined: i sared as a hobby projec wih a regional dimension added only laer. Houdini does no ye saisfacorily reproduce he reference mode of behavior, especially he oal consrucion volume. This poins a possible flaws in he boundary adequacy of Houdini. I is plausible o assume ha Houdini sill lacks proper represenaion of he facors driving consrucion by housing associaions. On he oher hand, experimening wih Houdini demonsraes he agiliy of sysem dynamics in comparison wih largescale demographic modeling. Furhermore, in hindsigh, Houdini provided PBL wih a saring poin for modeling insiuional feaures of Duch housing and real esae markes and proved an imporan preliminary model for he Middle Incomes simulaion described in chaper VI. V.9 Conclusion and discussion Upon heir firs encouner in he early sevenies, urban dynamics and housing economics clashed and hereafer developed in isolaion of one anoher. Neverheless, sock, flows, feedback loops and real world policy problems are innae o boh fields. A leas one implici sysem dynamics model, he 4QM, exiss wihin housing or real esae economics. Only since 2007, references o i are found in sysem dynamics lieraure. I may be useful o explore oher implici sysem dynamics models in urban, real esae and housing economics and relaed sciences (geography, urban sociology, planning) and o model hem using formal sysem dynamics mehodology. Nowihsanding he inspiraional sparks of Urban Dynamics, a closer connecion beween sysem dynamics and he subsanive sciences may be o he benefi of boh fields. Houdini is a housing marke model based on boh sysem dynamics and housing economics. Is developmen indicaes ha a moderaely experiences sysem dynamics modeler wih a background in he subsanive field can consruc a argeed and working housing marke model in a limied amoun of ime, a leas in comparison wih oher modeling approaches. Nowihsanding a significan wish lis for a major revision, Houdini 1.0 is a funcional model wih a firs pracical applicaion in he long erm spaial scenario projec finished. Houdini has provided he PBL saff wih grea learning opporuniies abou he presence of sysem dynamics in oher approaches, on modeling insiuional feaures of he housing marke, on he ype of criicism o be expeced for his ype of modeling. Laer on, Houdini proved a suiable preliminary model for oher projecs. As o he learning aspec of sysem dynamics modeling, Houdini demonsraed more han sufficien performance.

95

96 VI Middle Incomes 30 VI.1 Inroducion Tradiionally, Duch housing associaions have provided renal housing for a large share of he populaion and accoun for 40% of he oal housing sock. Recenly, however, due o European compeiion regulaions and lobby pressures from commercial real esae invesors, he socalled sae suppor regulaion (SSR) sipulaes ha 90% of he allocaed social renal dwellings should be assigned o lower income groups. The SSR furher resriced he posiion of middle income groups on he housing marke, wih heir housing marke opporuniies already diminished by inflaed prices for owner occupied housing and an underdeveloped privae renal secor. Inroducion of he SSR also coincided wih he impac of he credi crisis on he Duch housing marke. The iming of he inroducion of he SSR complicaes proper impac analysis of he SSR. Empirical daa would always encompass boh effecs. The PBL Neherlands Environmenal Assessmen Agency herefore resored o building a sysem dynamics simulaion for assessing he isolaed impac of he SSR on housing marke success raios for differen income groups. VI.2 Conex of he sysem dynamics inervenion The modeling projec repored here addressed he impac of he SSR on he posiion of middle income groups wihin he wider framework of he enire housing marke. The model iself is embedded in a mixed mehodology research projec wih addiional policy and housing lieraure sudy, regular daa analysis and inerview wih sakeholders. The projec was sared when he Housing Secion of he Minisry of he Inerior requesed PBL Neherlands Environmenal Assessmen Agency o carry ou an impac analysis of he SSR. A projec group consising of policy officials, researchers, academics, policy advisors, PBL managemen and research saff provided guidance o he projec. The research saff of PBL carried ou he research aciviies wih regular consulaions from one academic. The projec ran from Ocober 2011 o Ocober 2012 provided inpu o oher PBL sudies on he effec of demographic change on he housing and land use (De Groo e al., 2013) and parened several aricles in professional magazines (Eskinasi & De Groo, 2013; Van Middelkoop, De Groo, Verwes, & Eskinasi, 2013). The final repor (Eskinasi e al., 2012) was discussed in Parliamen. 30 This chaper is based on wo policy repors of PBL he Neherlands Environmenal Assessmen Agency and wo papers presened a he ENHR 25h Inernaional conference, Tarragona, Spain, i.e: De Groo, C., & Eskinasi, M. (2013). Increased homeownership in an ageing sociey: he races of elderly homeowners in declining and ensed housing markes in he Neherlands. Eskinasi, M., De Groo, C., Van Middelkoop, M., Verwes, F., & Conijn, J. B. S. (2013). Simulaing success raios for middle income households wih sysem dynamics.

97 94 chaper VI Hisory of he problem conex Housing associaions play an imporan role in Duch housing, owning approximaely 2,25 million dwellings, or 31% of he oal housing sock and 70% of all renal dwellings (CBS, 2013). From he 1950 s, iniially privae housing associaions had become imporan insrumens in sae housing policies. Social housing in he Neherlands radiionally had a mass provision characer, raher han being limied o he mos disadvanaged groups in sociey. The 1990 s consiue a waershed in he radiional approach. Fuure subsidies and ousanding loans were canceled ou in 1995 ( Bruering ), leaving housing associaions on more disance from he sae and a full financial risk for managing he large social renal housing sock. Some minor sae suppor insrumens coninued o exis: hrough a sae guaranee srucure (WSW) housing associaions had access o cheap finance from a bank for governmen agencies (BNG), he sae supervising organizaion CFV may allocae direc financial suppor for disressed projecs or enire housing associaions and furhermore, some municipaliies had coninued o provide housing associaions wih cheap land for social housing consrucion (Eskinasi e al., 2012, p. 29). Social rens are sill subjec o sae regulaion, bu housing associaions also play a role in subsidizing renal housing. The Bruering coincided wih an unprecedened housing boom hroughou he 1990, when house prices doubled in 10 years. The equiy of housing associaions grew significanly, bu ren levels only increased abou 18%. Sae regulaion did no allow high ren increases, bu housing associaions also conribued by keeping rens srucurally below he sae given maximum. The equiy of housing associaions is in mormain, so hey lack incenives o srive for a marke yield on equiy invesed. The resuling yield compression largely eliminaed compeiion in he renal secor. In his siuaion of nearmonopoly, waiing liss coninued o exiss, as especially new middle income households could no afford owner occupied housing and he favorable rens discouraged higher income enans o move ou of social renal housing. Commercial invesors in renal housing, mosly backed by pension fund capial, proesed agains he apparen lack of level playing field and sared lobbying in he naional governmen and he European Union. Afer years of lobbying and debae, he European Commission approved he socalled Sae Suppor Regulaion proposed by he Duch governmen in December This regulaion defined social housing (i.e. dwellings owned by housing associaions wih monhly rens below 647 as a service of general economic ineres (SGEI) and is accompanying sae suppor arrangemens (menioned above). The arge group of sae suppored social renal housing was limied o households wih annual incomes below (2011 price level), abou 42% of he populaion 31. I was sipulaed ha 90% of all available social dwellings mus be allocaed o households wih incomes below he said No formal resricions apply for he allocaion of he remaining 10%, his is lef o he compeence of housing associaions. The approaching inroducion of he SSR spurred a ho debae, paricularly on he effecs for he lower middle income groups ( o roughly ; 13% of all households). These groups were said o face severe affordabiliy and morgage availabiliy problems for enry o he owner occupied housing secor, he former caused by he seep price increases in he 1990 s and he laer by he response of he banking secor o 31 Of which, for many differen reasons, a cerain share already lives in owner occupied or commercial renal housing.

98 Middle Incomes 95 he 2008 credi crisis. I was feared ha he SSR would no only resric heir enry o he social housing secor, bu also discourage propensiy o move house and so depress housing marke dynamics even more, on op of he unfolding effecs of he credi crisis (Arivé & OpMaa, 2011; Kromhou, Smeulders, & ScheeleGoedhar, 2010; RLI, 2011). Many paries herefore advocaed o exend he enry o he social housing marke also o middle income groups. The narraive of mos repors, however, was o depar ex ane from he difficulies of middle income groups, illusrae his deparure poin wih descripive analysis and hen o call for changes or even aboliion of he SSR. Bu no fundamenal impac analysis was made, comparing housing marke effecs ceeris paribus of scenarios wih and wihou he SSR. Several facors would highly complicae such an impac analysis. Firs, he SSR was inroduced only recenly so only very limied daa would be available. Second, is inroducion coincided wih he mos srenuous effecs of he 2008 credi crisis on he Duch housing marke, so ha available daa, if any, would conain influences of boh facors. Finally, he laes comprehensive survey daa se on housing preferences daed from 2008 (predaing he crisis), wih a new survey only available in April The PBL research saff herefore proposed o develop a simulaion model using daa of he 2008 survey. This simulaion would allow for proper impac analysis, bu no for prognosis of he acual course of evens or even deep and deailed insigh ino he dynamics of he period This was consisenly communicaed o all paries involved hroughou he enire projec in order o properly manage expecaions abou he model validiy. Furhermore, judging from previous housing sudies, he PBL research saff expeced large regional variance in he effecs of he SSR, wih he public debae mosly voicing common percepions of he housing marke in congesed Randsad regions. As one housing marke researcher was also a seasoned sysem dynamics modeler, i was decided o use sysem dynamics for he modeling pars of he research projec. VI.3 The sysem dynamics modeling projec Preprojec aciviies Preprojec aciviies consised of regular consulaions wih he Housing Secion of he Minisry of he Inerior. In hese consulaions, i was discussed and agreed ha PBL would make an impac analysis of he SSR. The managemen eam of PBL discussed and approved he projec proposal. A policy official of he Minisry and he responsible manager of PBL aced as gaekeepers for saffing he projec group and he research eam. Research aciviies The research projec followed a mixed mehodology approach. A discourse analysis of he hisory of he SSR revealed several blind spos in he public debae and idenified implici hypoheses on he effec of he SSR and miigaing policy measures for simulaion. Review of academic lieraure on housing preferences, housing economics provided heoreical cornersones for he simulaion model. Regional inerviews and documen

99 96 chaper VI analysis indicaed he large regional variaion in effecs and formed a realiy check for simulaion resuls. Saisical daa analysis enriched he regional inerviews and provided inpu for he simulaion model. The research eam me weekly o discuss projec managemen and o share preliminary insighs and conclusions. The modeler monhly consuled one of he academics in he projec group o discuss modeling aspecs and o review simulaion resuls for validaion. The oal ime expendiure for PBL saff amouned o ca working hours, of which approximaely 500 were invesed in he consrucion of he simulaion model. In order o assess he magniude of he regional variaion, he research eam seleced six widely differen regions. The region around Amserdam is he mos ensed housing marke in he counry, rural bu sable Friesland and declining ZeeuwsVlaanderen are he oher opposies. Mos ineresing were hree inermediary regions: Eindhoven, Roerdam and ArnhemNijmegen, as we had no srong inuiions on he effec of he SSR in hese regions. VI.4 The resuling model The simulaion model is solidly grounded in heories on housing marke economics household behavior in relaion o residenial mobiliy. The simulaion iself was buil in Powersim Sudio 9 SR1. Inpu daa were generaed in SPSS from he housing survey. Inerfacing wih he simulaion model was done in Microsof Excel. Overall model srucure The overall feedback srucure of he model (see figure 29) connecs o he worldview of sysem dynamics. Decisions of acors influence he housing sock or marke. Acors, however, base heir decisions on informaion abou prices, rens, availabiliy. Households base heir decision o move no only on socioeconomic and demographic variables, bu also on marke informaion abou prices and availabiliy. Developers base decisions o buy land, and sar consrucion on price rends, local circumsances eceera. Housing associaions consanly balance financial and social goals hrough ren level selling, sale of housing and new consrucion on basis of needs perceived, i.e. marke informaion. The model gains is dynamic complexiy from his consan feedback. The acual simulaion model also draws concepually from he four quadran model of Di Pasquale and Wheaon (1996), which connecs hree parial markes (housing services marke, housing propery marke and consrucion marke) ino a balancing or equilibriumseeking feedback loop. I also inheried several modificaions of he 4QM from housing marke model Houdini (Eskinasi, 2011b) (see also chaper V). On he demand side, households exer demand boh on basis of semisaic household properies, bu also of marke informaion. On he supply side, wo main ypes of acors exis in he model. Developers srive o maximize profis upon sale of newly consruced housing. Landlords (housing associaions and commercial invesors) le dwellings, order new consrucion and someimes demoliion, sell exising renal housing and se ren levels. The commercial renal secor is open o exernal capial, so hese acors are driven by reurn on invesmen. The Duch social renal secor is a closed sysem. Housing

100 Middle Incomes 97 associaions srive o fulfil social objecives wihin he bounds of financial sabiliy. In he model, heir social objecives consis of consrucion, ren seing and sale of dwellings. The ineres coverage raio (ICR) is he main financial variable. A low ICR signifies financial problems, simulaes ren increases and sale and decreases consrucion. Higher ICR s exer converse influences. Figure 29 Middle incomes: overall feedback srucure Household secor The household secor of he model is based on he dynamic lifecourse approach (Clark, 2012) where boh macro and micro facors influence he housing decisions (Van Ham, 2012). In sysem dynamics erms, he main srucure is an ageing chain, where households evolve hrough five sages or sock variables: young households in he formaion phase (up o 30 years), he family phase (3054 years) and a parallel caegory for households wihou children, he senior phase (5574) and he elderly phase (75). Figure 30 shows he main srucure of his model secor. Inermediary flow raes are based on duraion of he disinc phases. Family households moving o he senior or empynes phase produce new young households, hus creaing a feedback loop. This model secor was calibraed o fi he 2011 naional and regional demographic housing forecas PEARL (De Jong e al., 2005). The households are caegorizes ino wo dimensions: educaion level and curren enure. We discern hree educaion levels and assume i consan from household formaion onwards. Educaion level is relevan as a srong predicor for he income career of a household. Households wih higher educaion have larger probabiliy of aaining a high income mid and endcareer and when pensioned. Groups wih lower educaion end o have lower incomes hroughou heir full life cycle. This is relevan as he argeed middle income group is very heerogeneous as o age, educaion level and income dynamics. Households are labeled by curren enure or housing ype. The model discerns four ypes of renal housing, hree ypes of owner occupied (based on ren and price levels

101 98 chaper VI respecively) and a remaining enure caegory for households no having a home ye. The disaggregaion by curren enure form is required for simulaing residenial mobiliy, and filering and limiing enry of moving households o cerain enure ypes allows o simulae effecs of he SSR. Young o mediors Mediors Ageing mediors <Family duraion> New Households Household reproducion facor Young households Young o family Family facor Youngser duraion Families Seniors Empy nesing families Ageing seniors Senior duraion Family duraion Elderly Ouflow of elderly Elderly duraion Figure 30 Middle incomes: simplified household secor By means of probabiliy ables, he simulaed household evoluion (by household sage and educaion level) yields he developmen of households by sage, educaion and income. Demand and maching secor The purpose of his model secor is o ranslae household dynamics o demand in he housing marke by means of using mobiliy and housing preferences and by applying affordabiliy and insiuional resricions. In his secor, a cenral variable is he sock of acively house huning households. The household ype and enure specific average occupancy ime (based on he 2009 housing survey) deermines he inflow of households ino his sock. The sock has wo ouflows for successful and unsuccessful house huners. Unsuccessful households exi afer a consan disappoinmen ime. Successful house huners exi upon finding a new house maching heir preferences, aking ino accoun financial and insiuional resricions. The srucure here is similar o he Haaglanden model (Eskinasi e al., 2009). Dividing he number of successful house huners by he sock of searching house huners yields he average search ime, he cenral indicaor for chances of differen income groups on he housing marke. Figure 31 illusraes he overall srucure of his model segmen. The menioned insiuional and financial resricions include an array of facors like income developmen, budge shares for housing, maximal loan o income raios, age limis for morgages, barriers in allocaion sysems for social renal housing and he relaive pricequaliy levels of curren and available houses, influencing subsiuion Assuming a subsiuion elasiciy of 1.

102 Middle Incomes 99 from he given housing preference paern in he 2009 housing survey. For deermining housing budges, he model akes ino accoun all componens of he user cos approach (Conijn, 1995; Donders e al., 2010; Renes e al., 2006), including he effecs of morgage deb on he income. Moreover, households moving from one house o anoher and elderly households flowing ou of he housing marke vacae houses which are added o he available housing sock. income price qualiy curren vs available available houses mached houses & huners housing preferences housing allocaion resricions Moving households search ime Acive house huners new house huners Unsuccessful househuners Disappoinmen ime Figure 31 Middle incomes: simplified demand and supply maching Housing sock secor The housing sock secor regisers he changes o he housing sock. Is main srucure is a shor producion chain wih houses under consrucion, vacan and occupied houses and i is disaggregaed by enure ype and ownership: boh housing associaions and commercial invesors can own houses in he renal enure forms. Changes o he housing sock reflec sale of houses, ransfers beween owners and renal enure ypes, consrucion and demoliion. The supply side acors conrol hese flows. Houses move beween occupied and vacan sae on basis of he muaion and absorpion raes 33, boh of which are linked o he house moves of households in he demand and maching secor. The model regisers he dynamics of house prices and rens (or user coss for owner occupied housing) by means of socalled coflow srucures (Serman, 2000). In accordance wih Di Pasquale and Wheaon (1996), he absorpion rae of he owner occupied 33 The muaion rae conrols ransfer from occupied o vacan, he absorpion rae he ransfer from vacan o occupied.

103 100 chaper VI secor (i.e. he inverse of sale ime) exers a pressure on house prices 34. Furhermore, rens are under he influence of he policies of he supply side acors. consrucion absorpion rae B sale acive managemen cashflow operaion cashflow B ren increases B ICR ne overall cashflow R ineres paymens ineres rae Deb Deb changes Figure 32 Middle incomes: simplified diagram of housing associaions Supply secor This secor depics he aciviies of supply side acors, like consrucion, sale of renal housing and ren level policies. I discerns hree ypes of acors. Developers (as defined in his model) operae for he owneroccupied secor only. They srive for adding 2% annually o he owneroccupied sock, bu reac quie srongly o changes developmen profis and he absorpion rae. Developmen profis resul from dynamic house prices and residual developmen coss as in chapers V and Eskinasi e al. (2011). Commercial invesors and housing associaions operae in he renal marke. These acors build, sell and follow ren policies. The main crieria for commercial invesors 34 Afer saisical esing, he effec of he absorpion rae on he change of house prices was modeled as a cumulaive normal disribuion wih a range of 5% o 5% price change per year. A daase from Huizenzoeker.nl yielded monhly figures for number of houses sold, number of houses for sale and price changes from Sepember 2008 o November 2011, in oal 38 daa poins. The absorpion is calculaed by dividing he number of houses sold by number of houses for. We esed linear and loglinear models and found a correlaion of r 2 = 0,4627 beween independen variable absorpion rae and dependen variable yearoyear price change. Price changes for his period average minus 2,04% wih a sandard deviaion of 2,07%. The average absorpion rae is abou 80% wih a 25,5% sandard deviaion.

104 Middle Incomes 101 is he gross yield on renal housing, defined as he raio of rens o prices. Their ren policy focuses a maximizing rens wihin legal bounds. They hold fixed preferences for new consrucion and sale of renal houses, bu reac o he acual gross yield and he absorpion rae. Unlike housing associaions, commercial invesors are open o he capial marke and can arac exernal equiy. This allows hem o reac o gross yield raher han o solvency or ineres coverage. The equiy of housing associaion is earmarked for social renal housing. They can arac exernal finance, bu mus always balance heir ne cash flows wih ineres paymens. The ineres coverage raio is he main financial crieria for decision making. Safe ineres coverage raio s enice housing associaion o se lower ren increases, o sell less and o build more new houses. A simplified diagram of housing associaions finance is presened in figure 32. I should be noed ha all aciviies of hese acors influence he overall balance on he housing marke and indirecly he acions of households (see figure 29). Auxiliary secor Nex o he secors described above, he model conains many auxiliary variables for connecing o saring daa and for making differen aggregaions for graphing purposes. VI.5 Validaion As he model is relaively complex, several approaches are needed o validae is working and oucomes. Forreser and Senge (1980), Coyle and Exelby (1999), Vennix (1996) and many oher auhors cover echniques for validaing sysem dynamics models. Firs, he model is based in common real esae and housing lieraure, so ha srucure validaion was made more or less implicily during model consrucion. On he echnical side, issues like uni consisency are safeguarded in he sofware. Model srucures and oucomes were rigorously scruinized by he academic members of he projec group. Theil saisics of inequaliy (Serman, 1984) were also calculaed for several variables 35. Bu mos imporan, he regional inerviews and daa analysis poined in he same direcion as he simulaion resuls. In oher words, his conribued o behavior esing of he model. This facor significanly solidified he conclusions drawn. Furhermore, all projec eam members, involved academics and managers of PBL consisenly communicaed he scope of he model: is purpose is o make an impac analysis of he SSR, hree miigaing policies and regionally differen housing marke and populaion variables. I was srongly emphasized ha he model oucomes could no 35 The fi of simulaion oucomes o empirical daa and exising forecas hrough he Theil saisics of inequaliy (Serman, 1984) was done for nine higherlevel model variables, i.e. he oal housing sock, he oal populaions in number of households, four household subgroups available in he demographic forecass, oal new consrucion, new consrucion and sale by housing associaions. Table a1 presens he Theil saisics for hese variables. Noe ha variables wih high errors (RMPSE) have only few daa poins n. No much value was aribued o he Theil saisics for several reasons. Firs of all, oher validaion echniques had already convinced he projec eam. Second, here are sufficien daa poins only for comparing he demographic dynamics wih oher forecass and finally, he model was consisenly communicaed no o forecas evens, so ha a hisorical fi would seem relaively useless.

105 102 chaper VI be inerpreed as a deailed prognosis of he acual course of evens wih he financial crisis unfolding. This disclaimer was wellreceived by he policy officials and no discussions on he scope or validiy occurred ye. This indicaes ha he model was acceped as valid for is specific purpose, namely an impac analysis of he measures menioned above. The crossvalidaion of model oupu, inerviews and saisical analysis and he proper framing of he model purpose srongly increased confidence in he model. VI.6 Base run and policy alernaives Afer model esing and validaion, we simulaed he baseline (wih SSR) and alernaive (wihou SSR) policies o see he effecs of he SSR on he average search ime for lower, middle and higher income groups. Bu o our iniial surprise, average search imes ended o decrease srikingly in boh runs. Upon furher scruiny, we concluded ha he ouflow of elderly households mus be he main cause: he poswar generaion is srong in numbers and has a significan higher level of home ownership han prewar generaions. Wihin 15 o 20 years, his ouflowing poswar generaion will vacae large numbers of owner occupied singlefamily houses, resuling in coninuous downward pressure on house prices o he benefi of middle and even lower income groups (De Groo & Eskinasi, 2013; De Groo e al., 2013). These findings are similar o e.g. Mankiw and Weil (1988) and Myers and Ryu (2008). Table A Theil inequaliy saisics for nine main variables Households Housing sock Toal consrucion Consrucion by HA s Sale by HA s n r2= 0,9970 0,9960 0,8513 0,1075 1,0000 MSE= 2,63E09 3,64E09 8,46E07 1,13E06 2,25E06 RMSPE= 0,63% 0,77% 13,23% 0,19% 10,23% U m = 0,65 0,56 0,02 0,02 0,07 U s = 0,13 0,23 0,82 0,23 0,05 U c = 0,22 0,21 0,16 0,75 0,87 Households Young households Families & mediors Seniors Elderly n r2= 0,9970 0,6196 0,9705 0,9834 0,9966 MSE= 2,63E09 1,49E09 1,48E10 2,51E10 2,35E09 RMSPE= 0,63% 3,66% 3,76% 6,3% 5,20% U m = 0,65 0,24 0,89 0,93 0,10 U s = 0,13 0,04 0,07 0,01 0,77 U c = 0,22 0,72 0,04 0,06 0,12

106 Middle Incomes 103 Simulaed search ime for low income groups 8 7 Search ime (years) Base wih SSR Al wihou SSR Simulaion ime (years) Simulaed search ime for lower middle income group 8 7 Search ime (years) Base wih SSR Al wihou SSR Simulaion ime (years) Figure 33 Middle incomes: main simulaion resuls for (a) lower incomes and (b) lower middle incomes

107 104 chaper VI Simulaion resuls on he shor erm effec of he SSR confirmed iniial fears (see figure 33). Especially house huners from he lower middle income groups jus above he income limi suffered mos from he SSR. Income groups beween and were also impaced. Higher income groups winessed smaller changes in search imes due o indirec crowding effecs in he owner occupied secor. Lower income house huners, on he oher hand, were found o profi significanly from he SSR 36. Even hough in hindsigh, his is a logical effec, i had no been signaled le alone emphasized in he public debae, which had been biased owards he effecs for middle income groups. Figure 33 presens he developmen of average search imes for he wo main focus groups under he baseline and alernaive scenario s. Miigaing policies Furhermore, hree miigaing policies were simulaed, based on exising policy lines and he public debae on he SSR. Some paries advocaed selling former social renal houses or increasing he rens of suiable houses in order o ransfer hem o he commercial renal secor 37. These paries argued ha middle income groups should no have subsidized housing and ha hese measures would increase he availabiliy of houses for hem. The alleged effecs, however, did no maerialize in he simulaion (see figure 34). In he firs five o en simulaion years, owner occupied and commercial renal housing are financially sill ou of reach for house huning middle income families. Income increases in he longer run would improve affordabiliy, bu hen he effecs of he measures would no differ significanly from he overall relaxaion of he housing marke due o he ouflow of elderly households. Furhermore, he measures have negaive impac on he lower income groups for obvious reasons: suiable vacan houses would be moved away from hem. A hird miigaing policy consiss of ransferring some of he social renal houses ino a porfolio exemp of he SSR, for which housing associaions do no have sae suppor. As opposed o ransferring hem o he commercial segmen, i is no necessary neiher o increase rens ino he commercial segmen, nor o wai unil vacancy. The simulaion indicaes modes posiive effecs for middle incomes and modes negaive effecs for lower income groups, as his measure allows for compeiion beween hese group in a par of he social housing marke. 36 The graphs sugges ha he negaive effec for lower middle incomes is much larger han he posiive effec for low income groups. This applies on a per household basis. There are, however, many more house huning low incomes han house huning middle incomes. Measuring in oal search years over he enire marke, he simulaion wih SSR is slighly more favorable. 37 Houses wih monhly rens over 634, (2011 price level) are considered commercial renal housing.

108 Middle Incomes 105 Simulaed search ime for low income groups 8 7 Search ime (years) Base wih SSR Al wihou SSR Sale Commercial rens Transfer Simulaion ime (years) Simulaed search ime for lower middle income groups Search ime (years) Base wih SSR Al wihou SSR Sale Commercial rens Transfer Simulaion ime (years) Figure 34 Middle incomes: effec of miigaing policies for (a) low and (b) lower middle income groups

109 106 chaper VI Regional simulaions There is, however, significan regional variaions in housing markes. In order o simulae regional variance of he impac of he SSR, we ook a selecive sample of six regions, ranging from he highly ensed housing marke around Amserdam, hrough he urbanized regions ArnhemNijmegen, Eindhoven, Roerdam, o rural bu sable Friesland and declining ZeeuwsVlaanderen. Inuiively, we expeced he SSR o increase ension in Amserdam and probably also in Eindhoven, which was labeled as a region of scarciy in one change of ren regulaion legislaion. For Friesland and ZeeuwsVlaanderen, we expeced virually no complicaions as houses are much cheaper and people are generally oriened more owards home ownership in hese regions. We had no srong inuiions for ArnhemNijmegen (no a scarciy area) and Roerdam (low house prices, ageing economic srucure and selecive oumigraion). We double checked our simulaion resuls wih inerviews and analysis of daa and policy documens for hese regions. Amserdam came ou as he mos ensed region, as expeced. Immigraion of groups wih higher educaion plays a significan role: hese groups ener as low incomes, make a career and pass hrough he middle income segmen upward. Even hough he SSR cerainly adds o he ension, is impac is relaively small in regards o he overall pressure. The effec of ageing was much smaller in Amserdam, bu he miigaing policies worked relaively well because of he higher upward mobiliy of he well educaed populaion. In all oher urbanized regions, inraregional differences in pressures dominaed he oucome of he inerviews and lieraure sudy. In mos cases, no complicaions for middle incomes were found in he cenral ciies, bu pressure and effecs of he SSR concenraed in he suburbs, where he housing sock is generally more in line wih he preferences of middle income groups. Opposed o our inuiion, ArnhemNijmegen demonsraed he second mos ensed housing marke wih a relaively large sensiiviy o he SSR. The miigaing policies showed mixed resuls: In ArnhemNijmegen, sale had beer performance as house prices are somewha lower. In Eindhoven, he effec of sale was smaller, laer and more concurren wih he demographic effec. The simulaion for Roerdam showed a faser relaxaion of he housing marke han generally expeced. The resuls for Friesland and ZeeuwsVlaanderen mached our expecaions: very low pressure on he housing markes, no problems for middle incomes whasoever and consequenly, no addiional complicaions hrough he SSR. VI.7 Followup aciviies The final repor was published in Ocober 2012 and sen o Parliamen. On several occasions, MPs quesioned he Miniser of Housing abou he effecs of he SSR on basis of he repor. Some of he mos sriking findings were published in professional magazines: he expeced fuure impac on he housing marke when he baby boom generaion will sar leaving he marke around 2020 (Eskinasi & De Groo, 2013) and he fac ha real complicaions of he SSR are mos probably confined o he mos ensed housing markes in he Norhern Randsad only (Van Middelkoop e al., 2013). Furhermore, he analysis of baby boom ouflow was also published in De Groo e al. (2013) and on basis of he dynamic heory, more deailed daa analysis was published in De Groo and

110 Middle Incomes 107 Eskinasi (2013). Findings were also presened on several naional conferences. Near he end of 2013, he Minisry requesed addiional simulaions based on moions voed for in Parliamen. VI.8 Evaluaion of he projec Opinion of he projec group and research eam The projec group acceped he model oucomes as valid and relevan o he discussions on he SSR, even hough a ha ime, oher housing marke issues were dominaing he public opinion. On basis of he sudy, he Minisry of he Inerior fel confiden o oppose changes o he SSR proposed by some MPs bu encouraged municipaliies and housing associaions o use he possibiliy of he hird miigaing policy 38. The academic projec group member mos closely involved in model consrucion and validaion changed his percepion of sysem dynamics from iniial skepicism owards a sufficienly posiive aiude o reques sysem dynamics modeling assisance for a followup projec (see also chaper VII). The managemen of PBL found he conclusions on he regionally diverging impac of he ouflow of elderly relevan for policy and furher research. The dynamic heory of ouflow of elderly was elaboraed furher in De Groo and Eskinasi (2013) and De Groo e al. (2013). Members of he research eam valued he closekni cooperaion beween he modeling and he oher pars of he research projec. I was noed ha a projec incepion more emphasis was on he modeling, bu when resuls sared o emerge, he relaive weigh of he regional case sudies increased. They fel confiden ha his PBL research projec was more comprehensive han all previous sudies on he SSR. Even hough hey were no very deeply involved in he model iself, hey perceived i as reliable because i produces an inernally consisen sory and because is oucomes mached findings of he oher pars of he research projecs. Opinion of he modeling eam In he opinion of he modeling eam, in casu he sysem dynamics modeler, he middle income model increased he curren level of modeling skill and experiences. The model incorporaes experiences and building blocks from previous modeling projecs bu also includes several exensions. I is consisen wih common housing heories, provide answers o a complicaed quesion in a seing where oher approaches migh no have succeeded due o daa limiaions. The model compleely avoids driving exernal ime series. The modeler suggess ha he curren model size is close o he limis a single modeler can reasonably handle in he given work seing. The model does ake a significan se of sarup values and has a cerain amoun of disaggregaion i.e. in he housing sock, 38 Even if i should be admied ha he Minisry was relucan a priori o change he SSR, hough mosly because of he burdening procedures wih he EU and no because of any assessmen of impac on he housing marke.

111 108 chaper VI educaional levels of households. The array size of eigh a mos, however, did no escalae ino he problemaic dynamics experienced wih ITS (Eskinasi & Fokkema, 2006). In he ligh of Forreser (2007b) sandards, he model clearly demonsraes why expeced sagnaion of residenial mobiliy does no occur and why neiher he proposed changes o he SSR nor he miigaing policies have high leverage on marke dynamics. Moreover, he model oucomes sugges ha facors ha do maer (i.c. he nascen rising ouflow of elderly ) are no ye on he reina of policy makers and ha curren policies miss a crucial poin. On he oher hand, no high leverage policies were found ha miigae derimenal effecs of he ouflow. Overall and aking ino accoun ha he elderly ouflow finding sill reverberaes among policy makers, he modeler considers middle incomes o be a relaively successful sysem dynamics projec. VI.9 Conclusions The simulaion projec on he middle incomes has demonsraed ha naionwide measures can have largely differen effecs in differen regional housing markes. On he one hand, i confirmed ha he SSR can be problemaic for middle incomes in he shor run. On he oher hand, he projec has helped o pu he effecs ino perspecive. Firs, he posiive effec for lower incomes had been underemphasized in he public debae. Second, he ouflow of he baby boom generaion becomes very dominan in he long run and overpowers he impac of he SSR. Even hough he ouflow of baby boomers is presen in inernaional lieraure, he curren policy debae in he Neherlands is focused on adaping exising and building suiable houses for elderly people, bu disregards he ouflow effec as ye. Third, he simulaion unveiled he large regional variance in he impac of he SSR and he miigaing policies. The use of he sysem dynamics mehodology enabled he research eam o develop a relaively complex model for impac analysis of a paricular measure in a raher shor ime frame. Furhermore, i allowed o isolae he impac of he SSR from exernal effecs like he credi crisis, bu neverheless o pu he SSR ino a comprehensive picure of all ineracions on he housing marke. Crossvalidaion of conclusions and proper communicaion of he scope and limiaion of he model conribued o successful landing of he conclusions a he Minisry of he Inerior. Building he simulaion required significan sysem dynamics skill and experience, bu is sill worh he effor.

112 VII Morgage Model 39 VII.1 Inroducion In addiion o he srong role of housing associaions in he renal secor, he Duch sae has been supporing home owners for decades. Fiscal faciliies allow deducing morgage ineres from he axable household income base and o exemp home relaed equiy from equiy axes, boh equiy invesed in he home as special savings schemes for morgage deb service. The 1990 s winessed decreasing morgage ineres raes and abundan capial for morgage financing. Housing prices boomed, new morgage schemes maximized ineres deducions and he oal morgage deb grew from ca. 295 billion in 2000 o over 650 billion in The financial crisis in 2008, however, necessiaed banks and he governmen o finally face and manage he risk of his mounain of morgage deb. The Amserdam School of Real Esae developed a sysem dynamics model of he morgage growh in cooperaion wih a sysem dynamics modeler of PBL Neherlands Environmenal Assessmen Agency. The model demonsraes ha i is virually impossible o decrease he morgage deb significanly over he nex en o fifeen years. VII.2 Conex of he sysem dynamics inervenion The sysem dynamics projec was carried ou in 2012 and 2013, when he financial crisis had necessiaed he Duch governmen o inroduce auseriy policies and o resric he widely criicized morgage ax reducion. From January 1 s, 2013, households aking ou a new morgage are eniled o fiscal suppor only when he morgage is annuiy based. The sysem dynamics projec se ou o explore he possibiliies of reducing he oal morgage deb by means of furher policy inervenions. Hisory of he problem conex Hisorically, he Neherlands has a long radiion of allowing morgage ineres paymens o be deduced from household axable incomes. I is, however, only since he 1990 s ha his fiscal morgage suppor has become a deerminan facor in housing marke dynamics. The 1990 s winessed a shif in sae housing policy. In order o reduce financial pressure on he sae budge, financial arrangemens wih housing associaions were drasically reduced (see also IV.2). Furhermore, agreemens beween he sae and lower auhoriies on new housing consrucion increasingly emphasized owner occupied consrucion. Household and income growh had simulaed demand and he main policy paradigm was o build beer and more expensive housing in order o simulae vacancy chains. More affluen households should move ino owner occupaion and hus vacae he abun 39 This chaper is based on Schilder, F., Conijn, J. B. S., & Eskinasi, M. (2012). De Nederlandse hypoheekschuld in 2025: de (on)mogelijkheid om de sijging van de hypoheekschuld e beperken. Amserdam: Amserdam School of Real Esae.

113 110 chaper VII dan social renal housing for less affluen groups and young households. Beween 1995 and 2012, he owner occupied sock grew from around 3 million houses o nearly 4.4 million houses, whereas he oal renal sock decreases from 3.2 million o slighly under 3 million. Global economic growh led o decreasing morgage ineres raes. Housing prices boomed and banks developed many new morgage saving schemes in order o maximize he fiscal suppor and exend ineres paymens. In mos cases, households ook ou ineresonly loans wih a linked savings schemes, based on cash savings and/or sock invesmens. Ineres paymens could hen be deduced for he full amoun and duraion of he loan, and he axexemp savings scheme would amorize he loan fully and a once a he end. The financial engineering minimized he monhly coss of morgages for households or, in oher words, maximized morgage loan volumes and ineres deducions. This suppored furher price growh (SaenGeneraal, 2012), so i is reasonable o assume he presence of a financial acceleraor in house prices (see also Anundsen & Jansen, 2013). Boh price and volume growh led o a rapid increase of he oal amoun of morgage deb from approximaely 330 billion in 2001 o nearly 630 billion in 2011 (DNB, 2012), provoking concerns from he Naional Bank, inernaional organizaions and he naional governmen abou sabiliy of he naional economy and he governmen budge. Afer years of debae on reforms and incremenal measures, he governmen decided in 2012 o resric fiscal suppor for new morgages o annuiy based schemes only. Meanwhile, he housing marke sared o suffer he impacs of he Grea Financial Crisis of 2008 onwards: consrucion and house sales plummeed and prices decreased by abou 20%. Wih decreasing housing prices, more and more households faced negaive equiy (Van Middelkoop, 2010), especially younger households who bough homes a he price peak financed wih he more risky ypes of loans, in hose regions sruck mos by prices decreases from 2008 onwards. There is an unequal disribuion of LTV raios beween age groups: older home owners generally bough for sill low prices, have more and iniially low selfamorizing morgages on low and have been amorizing for a longer period, saw increasing house prices. Newer home owners bough for much higher prices, have iniially high morgages wih increasing ineresonly componens, did no ye buil up much savings or invesmens and face decreasing house prices. VII.3 The sysem dynamics modeling projec Preprojec aciviies Before he sar of he morgage model projec, several auhors had made empirical analyses of he problem of negaive home equiy. Schilder and Conijn (2012b) esimae ha approximaely half a million families had negaive home equiy in They also found ha negaive equiy is concenraed wih younger age groups and more recen home purchases. Van Middelkoop (2010) adds a regional perspecive and demonsraes also significan geographical concenraion around new own Almere and in he region near Roerdam. The absence of reliable daa on he amoun of equiy in savings schemes, however, complicaed analysis. Esimaes widely varied from 30 billion o 220 billion.

114 Morgage Model 111 Discussions beween researchers and officials of he Duch Naional Bank on he underlying assumpions led o he sar of he morgage modeling projec. The purpose was o simulae he fuure rends of he oal morgage volume and o es he impac of differen facors and policy measures. The modeling projec A small eam of housing researchers carried ou he morgage modeling projec: he auhors of one of he esimaes of he oal equiy in savings schemes (Schilder and Conijn (2012b) and a sysem dynamics modeler in heir nework. A preliminary model was buil, esed and fineuned in a relaively shor period. Afer model validaion and calibraion, several scenarios were repored in Schilder e al. (2012) and Schilder and Conijn (2012a). VII.4 The resuling model Overall srucure The overall model srucure consiss of hree main sock variables: home owners, oal ousanding morgage deb and he oal amoun of equiy in he said saving schemes. Developmen of he owner occupied sock and of house prices is modeled in sock variables as well, bu hese are governed by exogenous ime series and are no par of he feedback srucure. The model is basically an ageing chain of home owners wih morgage deb and equiy in savings schemes as is coflows. Figure 35 clarifies ha he model has no complicaed feedback srucure. Exogenous housing sock growh drives he ageing chain of home owners, he inflows of morgage deb depends on inflow of new home owners and house prices. Saving schemes dynamics are closely linked o morgage deb and have a fixed duraion. In fac, he only reinforcing loop in he model is he savings scheme equiy growh loop R1 wih he savings scheme ineres or yield as a parameer. The overall dynamics of he model are dominaed by he ineria and he hisory of he socks. Home owners according o age groups The full model adds more deail o he simplified srucure in figure 35. Firs of all, he sock of home owners is disaggregaed ino fiveyear age groups. Inflow and ouflow raes are age specific. Home owners move hrough he ageing chain on basis of simple firs order maerial delays. Furhermore, residenial dynamics are added for home owners moving from one house o anoher wihin he owner occupied secor, again based on age specific move raes. When moving o beer and larger houses, home owners ake ou addiional morgages. The coflows for morgage deb and saving scheme equiy are also disaggregaed by age groups.

115 112 chaper VII B1 Home owners New home owners Ouflow of home owners housing sock growh average morgage annuiy serviced B2 Morgage deb New morgage deb Debs serviced house price Scheme paymens B4 Equiy in saving schemes R1 equiy growh B3 Finished schemes Scheme duraion scheme yield Figure 35 Morgages: simplified model srucure Morgage deb by age group and ype The model discerns hree main ypes of morgages: selfamorizing or annuiy morgages, ineresonly morgages wih a linked saving scheme and pure ineresonly morgages wihou saving schemes. For ype 1, a par of he morgage paymens is used o amorize during he morgage duraion. The ousanding morgage deb decreases gradually and so do ineres and he enilemen o fiscal suppor. Type 2 consiss of paymens for he ineresonly morgages and for he saving scheme. Only a he end of he duraion, he morgage deb is amorized in one go by he saving scheme, which is properly dimensioned for his purpose. This ype of morgage maximizes ax deducions

116 Morgage Model 113 and profis from he savings ineres on he scheme. Some schemes are cash based and guaranee 100% accumulaion, oher schemes are sock invesmen based, offer higher yields bu also he risk of building up no enough capial, leaving he home owners wih a parially uncovered ineresonly loan. Type 3 are rue ineresonly loans wihou saving schemes. The sock of morgage deb is disaggregaed by age group and morgage ype. The disribuion of morgage deb over ypes for new and moving home owners is given exogenously. This parameer was used for simulaing differen policy scenarios. Anoher parameer models credi raioning for new home owners. Equiy in savings/ invesmen scheme The hird sock is only relevan o he second ype of morgages. I regisers he accumulaion of equiy in he savings scheme hrough periodical paymens and ineres accruemen. I forms a firs order maerial delay wih a fixed duraion. Sandard sysem dynamics maerial delays wih average duraion did no yield opimal resuls, so a discree conveyorbel delay was used in he final model. When savings schemes finish, hey auomaically amorize he conneced morgage. Iniial values and parameers Iniial values of he socks were aken from he 2002 housing survey. Parameers for e.g. residenial mobiliy raes are based on a ime series of four subsequen housing surveys i.e. 2002, 2006, 2009 and VII.5 Validaion The reference mode of behavior for he model is he developmen of oal morgage deb in he Neherlands (see figure 36). Differen sources for oal morgage deb give somewha differen values 40. The error level beween hese sources is abou 4%. Furhermore, he number of house sales was available as a second reference mode of behavior. The projec eam succeeded in calibraing he model o closely mach boh reference modes of behavior (see figure 37 and figure 38). Theil saisics for boh reference variables have errors on he 3% level, i.e. lower han he error level beween empirical sources for morgage deb The saisical office CBS and EMF/DNB have full ime series for , housing survey only four years. Theil saisics beween CBS and EMF/DNB are: RMPSE = 3,8%, U m = 0,70, U s = 0,18, U c =0, Theil saisics for morgage deb: RMPSE = 3,0%, U m = 0,04, U s =0,80, U c =0,16; Theil saisics for house sales: RMPSE = 3,2%, U m = 0,13, U s = 0,10, U c = 0,76

117 114 chaper VII Figure 36 Morgages: reference mode of behavior from hree sources 800 Toal morgage deb 700 Morgage Deb (Billion ) CBS Simulaion Years Figure 37 Morgages: oal deb RMoB and simulaion