Table of Contents. MLS Home Price Index (MLS HPI) Methodology Page 1 of 24

MLS HPI Methodology

Table of Contents Introducton... 2 Partnersh... 2 Hghlghts... 2 MLS HPI... 2 Benchmark Prces... 3 Relatve Benchmark Prces... 3 Markets... 4 Market Segmentaton... 4 Data nclusons and exclusons... 5 MLS HPI Methodology... 5 Data... 5 Market Segmentaton... 6 Modelng Aroach... 7 Model secfcaton... 8 Varables... 9 Regresson... 10 Aggregates and Comostes... 11 MLS HPI... Error! Bookmark not defned. Benchmark Prce... Error! Bookmark not defned. Relatve Benchmark Prces... 13 Index Mantenance... 15 Governance... 15 HPI Contact Informaton... 16 Dsclamer... 16 Aendx A... 17 Aendx B... 19 Aendx C... 22 MLS Home Prce Index (MLS HPI) Methodology Page 1 of 24

Introducton The MLS HPI s desgned to be a relable, consstent, and tmely way of gaugng changes n home rces. It s calculated each month and covers fve major housng markets (Greater Vancouver, Fraser Valley, Calgary, Greater Toronto, and Greater Montreal, wth addtonal markets to come). The MLS HPI s also aggregated for the collecton of these markets. The MLS HPI tracks rce levels at a ont n tme relatve to rce levels n a base (reference) erod for one- and two-storey sngle famly homes, townhouse/row unts, and aartment unts. A comoste MLS HPI s also calculated for the collecton of these housng categores n each of the fve housng markets tracked by the ndex. Partnersh The MLS HPI s generated and ublshed under agreements between The Canadan Real Estate Assocaton, Greater Vancouver Real Estate Board, Fraser Valley Real Estate Board, Calgary Real Estate Board, Toronto Real Estate Board, Greater Montreal Real Estate Board, and Altus Grou. The MLS HPI model was develoed by a desgn team at Altus Grou that ncludes Professor Franços Des Rosers, the 2011 recent of the Internatonal Real Estate Socety Achevement Award. He has been teachng Urban and Real Estate Management snce 1976 wthn the Faculty of Busness Admnstraton of Laval Unversty n Quebec Cty, Canada. Reresentatves from Statstcs Canada, Canada Mortgage and Housng Cororaton, the Bank of Canada, Fnance Canada and Central 1 Credt Unon have also revewed and endorsed the MLS HPI methodology, and rovded valuable contrbutons n suort of ts develoment. Hghlghts MLS HPI The MLS HPI s avalable for sngle famly homes (whch are further slt nto 1-storey, and 2-storey sngle famly homes), townhouse/row unts, and aartment unts. These sub-ndces are used to calculate a comoste or overall MLS HPI n each market beng tracked. The MLS HPI for each market s also used to calculate an aggregate MLS HPI for the collecton of Metrooltan markets. MLS HPI values track relatve rce levels by comarng rce levels at a ont n tme to rce levels n a base (reference) erod. Because the base (reference) erod has a value of 100, t s ossble to uckly nfer the extent to whch rces have changed relatve to the base erod. For examle, f the base (reference) erod for the HPI s the month of January 2005, and the HPI value for Aartment unts n Setember 2011 s 135.1, ths ndcates that Aartment unts n Setember 2011 were u 35.1% comared to January 2005. The MLS HPI s calculated usng multvarate regresson analyss, a commonly used statstcal technue. Usng a hybrd modelng aroach that merges the Reeat-Sales and Hedonc Prce aroaches, the MLS Home Prce Index (MLS HPI) Methodology Page 2 of 24

MLS HPI model reflects contrbutons made by varous uanttatve and ualtatve housng features toward the home rce, ncludng: Number of rooms above the basement level Number of bathrooms & half-bathrooms Suare footage for man lvng & basement areas Whether t has a frelace and/or fnshed basement Lot sze The age of the roerty Parkng How the home s heated Foundaton, floorng, sdng & roofng tyes Whether the roerty has waterfront or anoramc vew Whether the roerty has been sold revously (newly constructed and revously unsold, or reeat sale) Proxmty to shong, schools, hostals, olce statons, churches, sorts centres, golf courses, arks, and transortaton (ncludng the tran staton, ralways, and arorts) Detals on MLS HPI calculatons aear n the MLS HPI Methodology secton below. Benchmark Prces The MLS HPI model s used to calculate Benchmark Prces. A Benchmark home s one whose attrbutes are tycal of homes traded n the area where t s located, one beng generated for each suorted Subarea. Benchmark roerty descrtons are based on medan values for uanttatve roerty attrbutes (e.g. above ground lvng area n suare feet), and the most commonly occurrng value (.e. modal value) for ualtatve attrbutes (e.g. basement s not fnshed). Benchmark Prces are avalable for each housng category tracked by the MLS HPI n each market. Comoste and Aggregate Benchmark Prces are also avalable, reresentng an aggregaton of Benchmark categores and Metrooltan markets tracked by the Index. Ths enables Benchmark Prces and ther rce changes to be comared across areas, and to the overall market. Detals on Aggregate and Comoste Benchmark home rce calculatons aear n the MLS HPI Methodology secton below. Relatve Benchmark Prces Relatve Benchmark Prces (RBP) show the ercentage by whch a Benchmark Prce n a artcular area and category s above or below the Benchmark Prce for the overall market at a ont n tme. The RBP for the overall market s 100 at every ont n tme for each housng category tracked by the HPI. Ths enables uck dentfcaton of market areas where Benchmark Prces are above (or below) the overall market for each Benchmark housng tye, and by what ercentage. MLS Home Prce Index (MLS HPI) Methodology Page 3 of 24

Detals on RBP calculatons aear n the MLS HPI Methodology secton below. Markets The MLS HPI, Benchmark Prces, and Relatve Benchmark Prces are avalable for Greater Vancouver, Fraser Valley, Calgary, Greater Toronto, and Greater Montreal. Housng markets ncluded n the MLS HPI System meet a number of crtera based on ther contrbuton to rovncal and natonal sales actvty. The MLS HPI wll be exanded to nclude the followng markets, based on the followng crtera: Where rovncal home sales actvty accounts for x% of natonal actvty, and x s: Board/Assocaton home sales actvty must account for y% of rovncal MLS res. sales actvty, where y s: Real Estate Boards/Assocatons meetng crtera for ncluson n an exanded MLS HPI: Less than or eual to than 5% Greater than or eual to 25% Wnneg, Fredercton, Moncton, Sant John, St. John s, Halfax-Dartmouth, Regna, Saskatoon Greater than 5% and less than or eual to 15% Greater than or eual to 10% Edmonton, Quebec Cty Greater than 15% and less than or eual to 25% Greater than or eual to 5% Okanagan-Manlne, Vancouver Island, Vctora Greater than 25% Greater than or eual to 3.5% Hamlton-Burlngton, Mssssauga, Durham Regon, Ottawa, London Market Segmentaton To generate consstent ndces, markets are dvded nto areas and sub-areas for whch sales n MLS HPI categores have smlar attrbutes (homogenous). Sub-areas have the same geograhcal boundares as those used by Real Estate Boards/Assocatons, whch are well known as neghbourhoods. They are used to set MLS HPI sub-ndces, Benchmark Proertes, and Benchmark Home Prces. Each sub-area s MLS Home Prce Index (MLS HPI) Methodology Page 4 of 24

tested to confrm that t s small enough to ensure homogenety and large enough to ensure that there are suffcent sales volumes to model the MLS HPI throughout housng market cycles. Detals on market segmentaton aear n the MLS HPI Methodology secton below. Data nclusons and exclusons The MLS HPI ncludes transactonal data for home sales va MLS Systems at artcatng Canadan Real Estate Boards and Assocatons. These data nclude sale rce and addtonal nformaton that s added to suort the MLS HPI model, ncludng nformaton from a Geograhcal Informaton System (GIS) to cature addtonal neghbourhood characterstcs (roxmty factors) relatng to schools, man streets, water, and others. To mantan data consstency, transactonal data are fltered to nclude records above 0.5% and below 99.5% of the medan for the dstrbutons of Sale rce, Age, Lvng Area, Land Area, number of rooms, and number of bathrooms. Should a transacton record aear to nclude nternally nconsstent data, t s manually revewed and amended (scrubbed). Transactons for whch data dscreances cannot be reconcled wthout a feld vst are excluded. The scrubbng rocess results n excluson of less than fve er cent of transacton records. Detals on data aear n the MLS HPI Methodology secton below. MLS HPI Methodology Data Transactonal Data collected and used n the MLS HPI must frst be reformatted, analysed, sorted, and n some cases, amended; ths rocess s commonly referred to as scrubbng. Transactonal data are reformatted to nclude addtonal felds necessary to suort the MLS HPI. These new felds nclude calculated, estmated or nferred attrbutes from other avalable nformaton. For examle, Floor Area Above Man and Floor Area Man are created n the database, and are more useful than a unue Global Lvng Area feld. Detaled lvng areas by floor are aggregated and comared to the Global Lvng Area n MLS HPI regressons. For markets where Transactonal Data ncludes detaled Lvng Area nformaton, t s rortzed over the sngle Global Lvng Area n modelng tests. In keeng wth best ractces, results are fltered to nclude records wth values above 2.5% and below 97.5% of cumulatve Normal dstrbutons; other results are treated as outlers and automatcally removed. To mtgate volatlty, a movng fve-year erod s used, snce the use of a shorter samle horzon may result n an nsuffcent number of sales over the erod and cause ndex naccuraces. Cook s Dstance s used to estmate the nfluence of an observaton when dong least suares regressons, and hels detect outlers or dentfy a sub-area where t would be recommended to have MLS Home Prce Index (MLS HPI) Methodology Page 5 of 24

more data onts. Cook s Dstance s also used to dscard outlers that may exert a sgnfcantly detrmental mact on the MLS HPI. When the Cook s Dstance for an observaton s hgh, the observaton s redrected to the scrubbng rocess for manual valdaton. To ensure the full otental to extract knowledge from outlers, observatons wth a hgh measurement of Cook s Dstance are manually revewed and valdated before beng removed. Market Segmentaton After revewng the data, sub-areas are tested to ensure they are small enough to be homogenous and large enough to be statstcally sgnfcant. Dummy varables are created for each sub-area and ntroduced n the modelng rocess. Vsual valdaton usng trend mas of resduals, sale rce/suare foot of lvng area, and average ncome er household are used to further valdate sub-area delneatons. Sub-areas must have a mnmum level of sales actvty to be statstcally sgnfcant; accordngly, where sales volumes fall short of the mnmum, sub-areas may be groued nto sub-area sets for samlng uroses. These sub-areas are also examned to suggest alternatve geograhc boundares when a gven attrbute among roerty records lacks suffcent homogenety. The use of dummy varables n models usng sub-area sets enables each subarea n a groued samle to be reorted searately wth ts own unue value. Sub-areas themselves reman ntact, wth ther own ndvdual Benchmark Proertes and sub-ndces once MLS HPI models are comlete. Sub-areas wth nsuffcent data are excluded from subseuent calculatons. The frst valdaton of sub-area defntons reles on a cartograhcal analyss of the homogenety of two demograhc characterstcs, average ncome and educaton levels. Results show that average ncome s a key contrbutor wth regard to demograhc homogenety. A vsual nsecton s erformed to dentfy adjacent sub-areas for whch dsarate average ncome and/or educaton levels for households would reclude groued statstcal rocessng of ther resectve transactonal data. Statstcal dstrbutons for lvng areas, age of roertes, and sale rces are also analysed to valdate sub-area defntons, and to suggest otental sub-area groungs. To reduce the mact of tme on dstrbutons, transactonal data sannng the years 2009 and 2010 are used. Sub-areas are further valdated by addng each sub-area nto a general model. A hedonc regresson s erformed whereby sale rce s modelled as the deendant varable and all sub-areas but one are used as ndeendent varables, wth the remanng sub-area servng as a reference or base sub-area. The model then assgns a value to each sub-area. On a cartograhcal bass, sub-areas are revewed to determne f sub-areas should be groued. When runnng a regresson wth sub-areas as exlanatory varables, the calculated coeffcents reresent the comaratveness of each sub-area to the base subarea. To determne whch sub-areas can be groued, results are llustrated cartograhcally and subject to vsual valdaton to determne f sub-areas wth relatvely comarable weghts are adjacent to one another. MLS Home Prce Index (MLS HPI) Methodology Page 6 of 24

In cases where sub-areas wth relatvely comarable weghts are adjacent to one another, sub-area homogenety s subjected to further valdaton, whereby each sub-area s geograhcally analyzed to determne f t should be groued or slt nto smaller sub-areas. Geograhcal dstrbutons for lvng areas, roerty ages and sale rces are vsually analyzed. Ths revew ncludes the use of Google mas to valdate breaks between sub-areas, and confrmaton that neghbourhoods on each sde of sub-area lmts are hyscally smlar. Usng the knowledge ganed though each of these valdatons, markets are segmented for each roerty tye. Models of emergng communtes wthn sales terrtores are taken nto account from the date that the number of Transactonal Data roerty records acheves a mnmum bound (tycally ten er month over a erod of at least twelve months). Analyss of these sales must also satsfy varous dagnostc testng crtera. In the ntal confguraton of sub-areas, new communtes are dentfed and modeled accordngly. The treatment of new communtes s also taken nto account as art of annual revew of the MLS HPI system. As art of the annual revew, changes to names and boundares for market segments n use by the Real Estate Board/Assocaton are also taken nto account, together wth dentfcaton of new sub areas that come nto beng. Modelng Aroach The MLS HPI s based on a hybrd model that merges Reeat-Sales and Hedonc Prce aroaches. Usng multvarate regresson analyss, a commonly used statstcal technue, the MLS HPI model reflects the contrbuton that varous housng features make toward the home rce, and ncludes a dummy varable n the hedonc model secfcaton to dstngush sngle and reeat sales. The MLS HPI s concetually smlar to the Consumer Prce Index (CPI), whch measures the value of a basket of common goods and servces. Smlarly, the HPI measures the contrbuton toward a home s rces that each attrbute or feature makes as art of a basket of housng features. The aroach used to construct the MLS HPI s sueror to the Reeat-Sales aroach that has ganed meda attenton over the ast few years n Canada and the Unted States: The Reeat-Sales aroach omts useful nformaton and samle sze s reduced because only homes that have been sold at least twce are used. The Reeat-Sales aroach may be ncaable of relably trackng home rces for sub-areas wthn a market. Prce ndces calculated usng the Reeat-Sales aroach may be roduced wth a consderable tme lag due to data collecton and avalablty. MLS Home Prce Index (MLS HPI) Methodology Page 7 of 24

The Reeat-Sales aroach assumes that ualtatve and uanttatve attrbutes of homes reman constant; however, the sgnfcance of Canadan home renovaton exendture each year makes ths assumton unrealstc. Model secfcaton Desgnng a relable MLS HPI reures that the regresson model be adeuately secfed. Model mssecfcaton can arse n a number of ways. A rgorous set of statstcal tests s used to dentfy and resolve otental roblems arsng from model mssecfcaton. In a lnear regresson, one of the man assumtons s that there are no remanng multcollnearty 1 henomena. Stewse regresson s emloyed to remove excessve multcollnearty by selectng only those exlanatory varables that contrbute sgnfcantly to exlanng rce varatons. As a dagnostc test, varance nflaton factors (VIF) are used to hghlght and remove varables wth a hgh degree of multcollnearty. The Akake Informaton Crteron (AIC) allows comarng models that dffer wth regard to ther functonal form, varable secfcaton, or both; as such, t can ad n model selecton based on how close values redcted by the model are to the real data. The AIC s used to test whch of the Lnear or Semlog functonal forms rovdes the best ft. To accommodate nonlneartes, the lvng area, lot sze and age of roertes are transformed nto non-lnear forms. Results of the AIC suggest the use of the semlog form over the lnear form. Addtonally, the Ramsey RESET Test s used to determne f some form of non-lnear transformaton s reured wthn the model secfcaton (wthout ndcatng how to amend the secfcaton). The RESET test estmates an auxlary regresson usng the estmated Y from the orgnal regresson: Yˆ = ˆ β + ˆ β X +... + ˆ β X + γyˆ + δyˆ + ωyˆ =1, 2, N 1 1 n n 2 3 4 where Ŷ s rased to the 2 nd, 3 rd and 4 th owers and re-nserted n the ntal hedonc euaton as addtonal ndeendent varables. The test then comares the orgnal and the auxlary regressons va F statstc test. The hedonc functon s shown to be non-lnear f at least one of these emerges as statstcally sgnfcant. n Ŷ added terms In cases where the euaton fals the Ramsey RESET test, the AIC confrms the functonal form. That the age of a roerty cannot be non-lnearly transformed may exlan the falure at the thrd and fourth degree for markets where roerty age s modelled as a bnary varable denotng age range. 1 Multcollnearty s a statstcal henomenon n whch two or more exogenous varables n a multle regresson model are hghly correlated. MLS Home Prce Index (MLS HPI) Methodology Page 8 of 24

Demand for one- and two-storey sngle famly homes s sgnfcantly dfferent, as reflected n ther sales rces. Accordngly, they are modelled searately, wth suffcent sales actvty to mantan searate and statstcally vald categores. An aggregate Sngle Famly Home sub-ndex s calculated usng the weghted ndex of one- and two-storey sngle famly homes. Detals on how the Sngle Famly Home subndex s calculated aear n the Aggregates and Comostes secton below. Sngle famly homes nclude both attached and detached structures, snce analyss shows that the behavour of a combned detached/attached ndex tracks congruently wth a detached ndex (confgured by extractng sales records of attached homes whle mantanng comlance wth test crtera). Detached and combned detached/attached ndces are montored to ensure that the congruency of ther resectve trends suorts a combned ndex. New communtes wthn a sales terrtory are consdered as art of an annual revew of the MLS HPI system. When accumulaton of Transactonal Data results n adjustments to market segmentaton of a Sales Terrtory, MLS HPI models are re-run to take account of geograhc revsons whle ensurng that homogenety s mantaned for each groung. Varables All avalable nformaton and data that descrbes land, buldngs and locaton amentes s consdered n the MLS HPI model secfcaton. Soco-demograhc attrbutes (namely, Educaton Level and Average Income) also contrbute to the determnaton of sub-areas and ther groung for samlng uroses. Addtonally, a Geograhcal Informaton System (GIS) s used to cature addtonal neghbourhood characterstcs (roxmty factors) such as those relatng to schools, man streets, water and other factors. Data are valdated before beng used n the modelng rocess. Each varable s analyzed (mnmum, maxmum, dstrbuton, form), resultng measurements are stored, and key varables are montored on an ongong bass. Varables for Lvng area, Land area, roerty characterstcs and dummy tme varables are ncluded n the model, and key varables (e.g. Lvng Area, Land Area) are transformed to ft the data (a lst of varables used n the MLS HPI aears n Aendx A). To cature the margnal contrbuton of each varable, tests are erformed wth the suare and the cube of varables, as well as wth ther resectve suare and cubc roots. Statstcal tests show that the suare root and cubc root transformatons best cature the margnal contrbuton of each transformed varable, and have greater statstcal sgnfcance than the suare and the cube of the varables. Accordngly, the suare root and cubc root of key varables are used. To mantan homogenety, outler records are fltered out so that data nclude records above 0.5% and below 99.5% of the medan for dstrbutons of Sale rce, Age, Lvng Area, Land Area, number of rooms, and number of bathrooms. MLS Home Prce Index (MLS HPI) Methodology Page 9 of 24

A random control samle s then created usng 10% of the remanng Transactonal Data records to run through the same rocess as the ntal model to valdate varables. Regresson Usng a stewse regresson rocedure, ndeendent varables are successvely forced nto the model and then removed from the hedonc euaton based on ther statstcal sgnfcance va a Student t-test. Varables ket n the model are fully analyzed and nterreted. It s ensured that tme dummy varables are ncluded and that key varables satsfy logcal rules (e.g. number of rooms cannot be negatve). Also, varables wth data occurrence greater than 5% wthn the database are ncluded n the model secfcaton 2, and a random control samle s confrmed as vald. Afterwards, Cook s Dstance s aled to dentfy and dscard outlers that may exert a deleterous mact on hedonc coeffcent estmates. Dagnostc statstcal tests (as below) are then erformed to determne f assumtons underlyng ordnary least suare (OLS) regresson modellng are volated. If test results ndcate that these assumtons are volated, or that the model s ms-secfed (e.g. omsson of an mortant varable) or subject to a functonal form desgn flaw, then the results and the samle are analysed, and correctve actons are taken at the data, scrubber, market delneaton or functonal level as arorate. One of the man assumtons for the (OLS) regresson method s that errors have the same varance throughout the samle. If true, the model s sad to be homoskedastc. If not, the data are sad to be heteroskedastc. As long as the assumton of homoskedastcty s not volated, OLS s consdered to be the best lnear unbased estmator (BLUE). When the assumton s volated, OLS regresson estmates are deemed neffcent and OLS s not the best regresson method. One or a combnaton of addtonal measures and strateges are used to detect heteroskedastcty, and when reured, correct for t (e.g. Whte test, Weghted Least Suares regresson technue, addtonal data transformatons). Moran s Index Test, often referred to as the Moran s I test, s used to measures the degree of satal deendence among resduals. A model can be consdered adeuate f ts resduals are not related n sace. If they are, ths s consdered to be evdence of satal autocorrelaton. Lke heteroskedastcty, the resence of satal autocorrelaton volates the OLS method assumton that resduals are ndeendent from each other. The resence of satal autocorrelaton s tycally marked by unstable regresson arameters and unrelable nference tests. Several solutons are avalable to correct for the resence of satal 2 For examle, f the number of roertes that have arkng s greater than 5% but the arameter Parkng s not n the model, the arameter s forced nto the model. MLS Home Prce Index (MLS HPI) Methodology Page 10 of 24

autocorrelaton, ncludng Casett s exanson method, satal autoregressve technues and Peer effect models. The Chow Test s also used to determne whether the coeffcents n a regresson model are the same n searate subsamles. As a test for structural change, t s manly used n tme seres analyses where the assumton of homoskedastcty s vald. Test results for break onts each month suggest that a structural change occurred n 2008 (lkely due to the global fnancal and economc crss). Benchmark Prces and Sub-Indexes Followng the generaton of regresson euatons, each subarea s benchmark roerty attrbutes are nserted n the euaton to calculate ther resectve benchmark rces. Each roerty tye suorted n the sad subarea s attrbuted a benchmark roerty, gnorng other roerty tyes. These ndvdual benchmark rces are calculated for each month. Monthly sub-ndexes are calculated usng the benchmark rce of the reference erod (January 2005) as the denomnator and rces n other erods as numerators to calculate corresondng monthly subndexes. Aggregate and Comoste Benchmark rces The MLS HPI System calculates a set of rce ndexes and sub-ndexes, Benchmark Prces and Relatve Benchmark Prces. Aggregate Benchmark rces for areas are based on the weghted 3 contrbuton of sales actvty n consttuent sub-areas for each Benchmark category (1-storey sngle famly, 2-storey sngle famly, townhouse/row unt, and aartment unt), whereby the MLS HPI model calculates Benchmark home rces for each sub-area usng alcable Benchmark home attrbutes n each sub-area. P crea = W, j* P, j j where P reresents HPI category Benchmark rce, reresents Benchmark category, j reresents consttuent sub-area, and w reresents the roorton of Benchmark category actvty for the sub-area. Several levels of Aggregaton exst and vary from board to board, deendng on ther secfed reurements. The next level s Area and the level above ths s the Sales Terrtory of the Real Estate Board, followed by Provnce and then the aggregate of artcatng boards n Canada. 3 Weghts based on roortonal values for a movng three-year erod of sales actvty. MLS Home Prce Index (MLS HPI) Methodology Page 11 of 24

Comoste Benchmark rces n each area are based on the weghted contrbuton of sales actvty n consttuent sub-areas er benchmark housng category, wth the Sngle Famly Benchmark rce analogously calculated based on weghted contrbutons of just 1- and 2-storey sales actvty: = j P crea W, j* P, j where P reresents HPI Comoste Benchmark rce, reresents Benchmark category, j reresents consttuent sub-area, and w reresents the Benchmark category s roorton of total sales actvty for the sub-areas. Smlarly, Metrooltan Comoste Benchmark rces are based on the weghted contrbuton of sales actvty n consttuent sub-areas er benchmark housng category. Aggregate and Comoste Indexes Snce Benchmarks are the only tem n the consumer basket, Paasche, Laseyres ndex 4 values do not change whle calculatng sub-ndexes er Benchmark category, snce uanttes cancel themselves out. P L P = = where P L and P P reresents Laseyers and Paasche Index resectvely, reresents Benchmark category, j reresents the subject erod, and 0 reresents the reference erod. Snce the Fsher ndex P F s obtaned by takng the geometrc mean of Laseyres and Paasche, uanttes also cancel themselves out. It s mortant to understand that ths statement s only true on sub-ndexes er tye; j, 0, j, 0, 0, 0, j, j, P = P * P F Unlke the Laseyres Index whch overestmates the varaton n rces, and the Paasche Index whch underestmates t, the Fscher Prce Index s more relable n the estmaton of actual rce change over tme. L P 4 Research and Innovatve Technology Admnstraton, Use of the Chaned Fsher Ideal Index to roduce the Aggregated Transortaton Servces Index, Economcs and Fnance, htt://www.bts.gov/rograms/economcs_and_fnance/transortaton_servces_ndex/methodology/df/methodolo gy_chaned_fsher_deal_ndex.df MLS Home Prce Index (MLS HPI) Methodology Page 12 of 24

The Chaned Fsher Index s used to calculate aggregate and comoste ndexes to conserve the drect month-to-month lnk that kees recent sale rces non-obsolete. Accordngly, the results of calculatons used n dervng the Metrooltan Comoste and Aggregate Comoste MLS HPIs also serve n ts calculaton: P FC = 0, 0, 1, 0, * 1, 1, 1, 0, * 1, 1, 2, 1, * 2, 2, 2, 1, *...* j 1, 1, j j, 1, j * j, j, j, j 1, where P FC reresents the HPI Chaned Fsher Index, reresents Benchmark category j reresents the subject erod, and j-1 reresents the reference erod. Relatve Benchmark Prces Relatve Benchmark Prces (RBP) show the ercentage by whch a Benchmark Prce n a artcular market and category s above (or below) that for the overall market at any secfc ont n tme. The RBP s calculated for each Benchmark category, wth market aggregatons as the numerare 5. For the Natonal RBP reort, the Benchmark Prce for the Aggregate of Metrooltan markets ncluded n the ndex serves as numerare for each Metrooltan market. For examle, the RBP for a 1-story sngle famly home n Toronto s calculated by dvdng the Benchmark Prce for a 1-story sngle famly home n Toronto by the Benchmark Prce for 1-story sngle famly home for the aggregate of all Metrooltan markets, wth the result multled by 100. Ths aroach s used for each Benchmark housng category, and for comoste Benchmark home rces. Analogously, ths aroach s also used n Metrooltan market reorts, wth the Benchmark rce for the overall Metrooltan market servng as the numerare. For examle, a tycal the RBP reort for Toronto would nclude the RBP for a 1-story sngle famly home n an area of nterest, calculated as the area s Benchmark Prce for a 1-story sngle famly home dvded by the Benchmark Prce for 1-story sngle famly home for overall Toronto market, wth the result multled by 100. Ths enables Benchmark Prces for an area or sub-area to be comared to those n other areas or sub-areas or for the overall Metrooltan market. In the natonal RBP reort, the Aggregate RBP for each category at every ont n tme has a value of 100, snce ts numerare s eual to ts comarator n the numerator. Ths enables uck dentfcaton of 5 Whle aggregatons are normally used as numerares, the flexblty of the MLS HPI System enables the use of other Benchmark rce numerares. MLS Home Prce Index (MLS HPI) Methodology Page 13 of 24

the ercentage by whch Benchmark home rces are above or below the overall market, and easy calculaton of the ercentage by whch Benchmark home rces n a Metrooltan market are above or below other markets. Usng RBPs rather than Benchmark Prces to comare Prces between and wthn Metrooltan markets enables uck dentfcaton of the ercentage by whch rces n Metrooltan markets are above the overall market, and ease of calculaton for ercentage dfferences n rces between markets. MLS Home Prce Index (MLS HPI) Methodology Page 14 of 24

Examle: RBP: Townhouse/row unt All Areas Area A Area B Jan 2011 100 136.3 105.8 The general formula for calculatng the ercentage dfference between X & Y s: (X/Y - 1) * 100. Accordng to the above table, n January 2011: The Benchmark Prce of a townhouse n Area A s 36.3% above the Benchmark Prce of a townhouse for the overall market --.e. (136.3/100-1) * 100 = 36.3%. The Benchmark Prce of a townhouse n Area A s 28.8% above the Benchmark Prce of a townhouse n Area B.e. (136.3/105.8-1) * 100 = 28.8% In ths examle, Areas may be defned as Metrooltan markets, wth All Areas reresentng the aggregaton of all Metrooltan markets ncluded n the MLS HPI. Alternatvely, Areas may be defned as sub-markets wthn a Metrooltan market, wth All Areas reresentng the aggregaton of all subareas wthn a Metrooltan market. Index Mantenance The MLS HPI System s revewed annually. The annual revew ncludes re-testng model secfcatons wth a vew to otentally strengthenng the model. If revews result n models beng re-secfed, hstorcal data are revsed. Data exclusons are also revewed and udated as necessary. Governance Polcy decsons on the use and crculaton of MLS HPI nformaton are the urvew of the MLS HPI Steerng Grou, whch conssts of reresentatves of CREA, Real Estate Boards and Realtor Assocatons takng art n the MLS HPI. MLS Home Prce Index (MLS HPI) Methodology Page 15 of 24

HPI Contact Informaton For techncal enures, or enures about ndex oeratons or busness develoment regardng the MLS HPI, lease contact Gregory Klum, CREA s Chef Economst at gklum@crea.ca For news meda enures regardng the MLS HPI, lease contact Perre Leduc, CREA s Meda Relatons Offcer, leduc@crea.ca. Dsclamer Data and reorts regardng the MLS HPI are rovded for nformatonal uroses only. Products and nformaton regardng the MLS HPI are not ntended for nvestment uroses. The nformaton and any statstcal data regardng the MLS HPI are obtaned from sources that CREA beleves to be relable, but does not reresent that they are accurate or comlete. All estmates and onons exressed by CREA regardng the MLS HPI consttute judgments as of the date of ths reort and are subject to change wthout notce. 2011 The Canadan Real Estate Assocaton. All rghts reserved. Unauthorzed use, dstrbuton, dulcaton or dsclosure wthout the ror wrtten ermsson of CREA s rohbted by law and may result n rosecuton. MLS Home Prce Index (MLS HPI) Methodology Page 16 of 24

Aendx A Varables used n the Model Parkng access Tangble or ntangble benefts that ncrease attractveness or value Proerty s servced by muncal aueduct Proerty s near a shong mall Method of heatng Source of energy for heatng Floorng tye Foundaton materal Proerty s eued wth a frelace Garage has two arkng saces Proerty s eued wth geothermal energy Proerty buldng s sem-detached Land sze n suare feet Proerty sdng materal Proerty has undergone major renovatons Only a art of Proerty s renovated Proerty s eued wth a roughed-n frelace Basement s fnshed Parkng lot has a shelter or carort Garage s located below man floor Roofng materal Proerty has a crawlsace Proerty has a vew of water Proerty has a anoramc vew Number of bathrooms Number of half-bathrooms Proerty s n roxmty to an elementary school or a hgh school Hydro lne neghbours Proerty lot Proerty has a vew of ower lnes Proerty s n roxmty to a tran staton Proerty s n roxmty to a church Proerty s n roxmty to an arort Proerty s n roxmty to a boulevard Proerty s adjacent to a boulevard Proerty n roxmty to a sorts center Proerty s n roxmty to a ralroad Proerty s n roxmty to a hostal Proerty s n roxmty to a olce staton Proerty s n roxmty to a rson Proerty s n roxmty to a golf course MLS Home Prce Index (MLS HPI) Methodology Page 17 of 24

Proerty s n roxmty to a ark Proerty s adjacent to a ark Basement lvng area n suare feet Tme dummy varable month and year Number of rooms above basement level Man lvng area n suare feet Number of rooms at basement level Age of roerty MLS Home Prce Index (MLS HPI) Methodology Page 18 of 24

Aendx B Proerty Tyes Consdered n MLS HPI Models Legend Used by the Board and modeled Analyss of ts characterstcs determnes how roerty s categorzed. Used by the Board but not ncluded n MLS HPI models Not used by the Board Fraser Valley Greater Vancouver Greater Montreal Calgary Greater Toronto Benchmark Category Two storey sngle famly home Two storey sngle famly home (Attached) Two storey sngle famly home (Detached) 1½ Storey Two storey /basement 2½ storey 3 Level slt 3 Storey 3½ storey 4 Storey Mult-storey 4 Level Slt 5 Level Slt Backslt 5 Mult-level Sdeslt 5 House wth Acreage - Two or more Storeys Country Resdence (mult-storey) MLS Home Prce Index (MLS HPI) Methodology Page 19 of 24

Fraser Valley Greater Vancouver Greater Montreal Calgary Greater Toronto One storey sngle famly home One storey sngle famly home (Attached) One storey sngle famly home (Detached) Bungalow Bungalow w/basement Bungaloft Bungalow - rased Country Resdence (one-storey) 1-Storey, basement entry 1-storey, slt entry B-level Slt-level 2-storey slt 3 level slt Backslt Backslt ALL Backslt 3 Backslt 4 Frontslt Sdeslt 3 Sdeslt 4 Sdeslt All House wth Acreage - 1 storey Rancher Rancher w/basement Rancher/Bungalow w/loft Hllsde Bungalow Hllsde Slt Townhouse Half dulex Aartment Bachelor/Studo Loft Stacked Townhouse Mult-level Condo (Bungalow) Condo (2-Storey) Condo (3-Storey) Sem-det Condo Condo Townhouse Detached Condo MLS Home Prce Index (MLS HPI) Methodology Page 20 of 24

Fraser Valley Greater Vancouver Greater Montreal Calgary Greater Toronto Aartment Sngle level aartment Multlevel Multlevel aartment Loft Bachelor/Studo Stacked Townhouse Studo Studo Sute Penthouse Condo Phased Condo Leasehold Condo Det Condo Co-O At Co-Ownersh At Comm Element Condo Condo At Dulex Trlex Quadrlex/Fourlex Fvelex Other Proerty Tye Manufactured Manufactured wth Land Moble Home Floatng Home Modular Home Carrage/Coach House Farm Recreatonal Rental Tmeshare Vacant lot Other Unknown MLS Home Prce Index (MLS HPI) Methodology Page 21 of 24

Aendx C Benchmark Home Defntons Benchmark homes are reresentatve of standardzed homes for secfc sub areas. Ther hyscal characterstcs reman fxed over tme. Benchmark roerty attrbutes are formulated for each sub area for Benchmark housng categores that have a sgnfcant resence n a sub-area. Benchmark roertes attrbutes are determned usng the medan value for each non-bnary feld (e.g. lvng area above ground), and the most freuent (.e. modal) value for each avalable feld that s a bnary. The followng descrbes general characterstcs for each Benchmark housng category: Two-storey sngle famly homes: A roerty wth two, or more, above ground floors. Ths tye of roerty s characterzed by the dstrbuton of bedrooms on the uer floor and a ktchen, lvng room and other day-to-day rooms on the man floor. Ths category ncludes Proerty Styles submtted by Partcatng Boards labeled as: 4 Level Slt, 5 Level Slt, One-and-a-Half Storey, Two- Storey, Two-and-a-Half Storey, and Three-Storey. One-storey sngle famly homes: A roerty wth one floor above ground. Ths tye of roerty s characterzed by the bedrooms, ktchen and dnng rooms beng on the same floor; the utlty room and laundry room are generally located below ground. Secal attenton s made to rased bungalows, where the basement s artally above ground and where the room dstrbuton rovdes crtera for ts assgnment to the arorate Benchmark housng category. Ths ncludes Proerty Styles submtted by artcatng Real Estate Boards labeled as: Back Slt, B-Level, Bungalow, Hllsde Bungalow, Hllsde Slt, 2 Storey Slt and 3 Level Slt. Townhouse/row unts: Townhouses have confguratons whch lay between aartment unts and freehold non strata buldngs, such as bungalows and two-storey houses. Owners tycally ay co-ownersh fees for mantenance and enjoy exclusve access to a art of the lot. Ths category ncludes Proerty Styles submtted by Partcatng Boards labeled as any of the submtted Styles, wth a note that the roerty s a Townhouse. Aartment unts: Aartment unts are characterzed by beng art of a mult-unt buldng. Occuants of aartment unts may or may not have drect access to the lot from ther unts. There are also no arts of the lot whereby access s reserved for only one of the co-owners or aartment occuants. Ths category ncludes MLS Home Prce Index (MLS HPI) Methodology Page 22 of 24

Proerty Styles submtted by Partcatng Boards labeled as: Sngle Level Aartment, Mult-Level Aartment, Loft, Penthouse and Studo Sute. MLS Home Prce Index (MLS HPI) Methodology Page 23 of 24

ERROR: undefned OFFENDING COMMAND: STACK: