Rent control and vacancies in Sweeden

Transcription

1 Workshop 11 - Housing Economics Rent control and vacancies in Sweeden Mats Wilhelmsson matsw@infra.kth.se Roland Andersson roland.andersson@infra.kth.se Kerstin Kllingborg kerstin.klingborg@infra.kth.se Paper presented at the ENHR conference "Housing in an expanding Europe: theory, policy, participation and implementation" Ljubljana, Slovenia 2-5 July

2 RENT CONTROL AND VACANCIES IN SWEDEN by Roland Andersson 1, Kerstin Klingborg 2 och Mats Wilhelmsson 3 Real Estate Economics Royal Institute of Technology Sweden 1 roland.andersson@infra.kth.se 2 kerstin.klingborg@infra.kth.se 3 matsw@infra.kth.se Draft: Word count: 7737 Abstract We estimate that the natural vacancy rate for rental apartment markets in Sweden is 1 percent on average a very low figure compared to the 4 to 12.1 percent estimated for residential markets in the US. While the natural vacancy rates are high in rapidly growing cities in the US we estimate that they are very low, even negative in some recent years, for rental apartments in the large, expanding Swedish cities. The natural vacancy rates for rental apartments in municipalities with an out-migration are also very low while the actual vacancy rates are high. But these results verify our hypotheses, reflecting the situation for rental apartments under the Swedish rent control regime. Thus, the very low or negative natural vacancy rates for large, expanding cities reflect the circumstance that queues for rental apartments have developed as a consequence of rent control. The actual high vacancy rates in regions with an out-migration reflect the fact that residences are immobile and durable. They are higher than they would be without rent control and therefore indicate welfare losses. In this paper we also make a major effort to estimate the influences of various determining factors on the actual vacancy rates, such as net migration, education level, employment rates, construction, demolition and rent levels. We find that net migration plays an important role in explaining the actual vacancy rates. 2

3 Introduction The issue of vacancies under rent control has not been dealt with very much in the literature. This is certainly due the fact that vacancies are not usually a major problem under a rent control regime for rental apartments. The major problem is rather housing shortages with queues as a consequence. Long queues for rental apartments have also developed under the Swedish rent control regime in the large, expanding cities. However, in smaller municipalities with out-migration, actual vacancy rates for rental apartments are often quite high and could remain so for a long time. Why? The aim of this study is to analyze this problem. We have organized the paper as follows. First, we present a review of the literature. Then we describe the Swedish rent control system briefly. We analyze the effects of rent control, distinguishing between expanding and contracting regions. We then formulate hypotheses accordingly for estimating the influences of various determining factors on vacancy rates. We estimate natural vacancy rates for rental apartments under rent control for a number of municipalities, in some cases also for labor markets. We also estimate the influences of various determining variables on the actual vacancy rates. In addition we investigate tendencies towards market orientation in some cities. Finally, we present our summary and conclusions. 1. Literature review There have been a number of analyses of rent controls in the literature; see for instance Lindbeck (1967), Olsen (1972), Fallis and Smith (1984), Arnott (1988, 1995 and 2002), Benjamin and Sirmans (1994), Glaeser and Luttmer (2001) and Hubert (2002). Turner and Malpezzi (2003) provide an extensive literature review. In general researchers claim that rent controls are unambiguously harmful. For instance, Lindbeck points out that a rent control such as the Swedish one leads to losses of welfare through a housing shortage, caused by the combination of a demand at the controlled rent level that is too high and the fact that too few new apartments are constructed due to the companies weakened incentives. Arnott (1988, 1995 and 2003) seems to be an exception, thinking that the analyses should be subject to some revision, then considering so-called second-generation rent controls. Glaeser and Luttmer show that the total welfare loss from rent control is potentially much larger than the standard textbook dead weight triangle due to the mismatch in willingness to pay between the 3

4 customers without apartments and the renters with apartments; see Andersson and Söderberg, (2006) for a similar analysis. In a world with perfect foresight without any matching problems and without rent control demand equals supply in equilibrium. But in the real world landlords and tenants do not have perfect foresight and matching problems occur as it takes a while for landlords to search for suitable tenants and for tenants to search for suitable residences. If demand is expected to rise in the future, a large inventory of vacant space may optimize the values of the properties for the landlords. The vacant space represents a call option on leased space and since option values increase underlying volatility, greater demand volatility will lead to a greater natural vacancy rate (Grenadier, 1995 and 1996, Gunnelin, 2000). Landlords are willing to accept the costs of leaving some of their units vacant, if they are convinced that they can recover these costs by charging higher rents on units contracted in the time being or later (Hendershott and Haurin, 1988). When a region is expected to grow landlords are willing to keep a big buffer of apartments by overbuilding. The greater the degree of heterogeneity in property characteristics, the more difficult the matching process between the rental space and tenants will be. Longer duration of vacancy (length of time that a unit remains in vacant status) and higher incidence of leases (probability that a unit becomes vacant) will lead to a higher buffer (Gabriel and Nothaft, 2001, Deng et al, 2003). Belsky (1992) defines the vacancy rate as the average duration of vacancies times the incidence. Due to these circumstances the concept of the natural rate of vacancy, V*, has been used in the literature as the optimal vacancy rate of buffer in a long-run equilibrium (Rosen and Smith, 1983): D = S(1-V*) (1) at the equilibrium rent R* where D represents the tenants demand and S the supply of residence space. Total demand could be defined to include not only the customers demand but also the landlords demand for overbuilding. In their model Rosen and Smith define the natural vacancy rate to be the rate at which real rent increases equal zero (see also Gabriel and Nothaft, 1988). At this equilibrium there is neither an excess demand for nor an excess supply of rental properties. The concept of natural vacancy rate for residences is used analogously to the concept of the natural unemployment rate used for employment markets resulting from search costs because of imperfect knowledge, transaction costs and similar factors among employers as well as employees. Thus an unemployment rate equal to the natural unemployment rate defines full 4

5 employment (Johnson and Layard, 1988). Marston (1985) found that excess labor in an area will vanish within a year owing to mobility. His conclusion is that any persistent geographic unemployment differentials are not evidence of uneven labor demand among regions but reflect the workers preferences for a region s attractive climate and other amenities as well as high wages and high unemployment insurance. Rosen and Smith (1983) propose that the rate of change in rent is a function of the deviation in the observed vacancy rate from the natural vacancy rate, assuming that the natural vacancy rate is constant over time. Several researchers have used main parts of their model, but with a consciousness of the fact that the natural vacancy rate may change over time (Gabriel and Nothoft, 2001). Rosen and Smith calculated the natural vacancy rates for rental housing in the 17 cities under study and found them to lie in a range from 5.6 to 23.2 percent, with a median of 9.8 percent. They also developed a model for determinants of the vacancy rates. Wheaton and Torto (1988) examine annual national vacancy and rental data for the period 1968 to They found that the national vacancy rate rose from 7 percent in 1968 to 13 percent in Gabrial and Nothoft (1988) found it to be 8 percent on average across the cities in their sample. They found a substantial variation among cities, highest in Houston with 12.1 lowest in Seattle with 3.9. They also found that the natural vacancy rate is higher in rapidly growing cities and cities with high rents. Jud and Frew (1990) report the natural rate to be 6.5 in the sample used in their study. In a more recent study Gabriel and Nothoft (2001) calculate the natural vacancy rate for 29 big cities in the US and find them nearly all to be in the range of 4 to 4.5 percent. This result is surprisingly low in comparison with the results they themselves as well as other researchers presented earlier. Tse and MacGregor (1999) investigated the Hong Kong housing market using a two-equation model. They found the natural vacancy rate to be 4.7 percent, but that it tends to be unstable over time. Tse and Webb (2003) investigate interactions between expected vacancy rate, office employment rate and the structural change in the services sector of the space absorption rate. They find that an increase in interest rate and labor productivity causes a positive change in the vacancy rates whereas an increase in unit labor cost and producer price lead to a negative change. Tse and Fischer (2002) explain the rent-vacancy behavior in the office market using a time-varying model and time-series data from Hong Kong, Sydney, Perth and London. Amy et al (2000) find the natural vacancy rate for office space in Singapore over the sample period to be between 10 and 12 percent. 5

6 In a recent study of the Swedish office market that is not under rent control Englund et al (2006) analyze the effects of an employment expansion on the adjustment processes towards a new equilibrium for rents, vacancies and the supply of office space, using an errorcorrection model. A temporary decrease in the vacancy rate is induced by the increased demand. The rent level increases in the short run, because of the inelastic supply of office space. Most of the total demand for office space is locked in into long-term lease contracts. Thus an excess demand occurs in the short run resulting in an overshooting increase in the rents due to the fact that the existing contracts were signed at a lower rent level than the current one. When these lease contracts expire, some of the tenants will not consider renewing them. Then hidden vacancies, i.e. the difference between space occupancy and demand at the current lease rate, will turn into open vacancies, corresponding to a negative shift in the demand, resulting in a decrease in the rent level. The vacancy rate increases towards the natural vacancy rate. New construction will make new office space available after an average period of three years. Then, the natural vacancy rate and a long-run equilibrium rent level will be reached; for the Stockholm commercial rental market the natural vacancy rate was found to be The Swedish rent control system The main principle in for controlled rents in Sweden is that the historical production costs should determine the rent level. Thus, no consideration is taken of the varying willingness to pay for apartments at different locations when determining the controlled rents. The rents for apartments of a given vintage were put at nearly the same level not only within a certain municipality but also over the whole country. Queues developed in the big cities such as Stockholm with increasing demand for rental apartments because of the rent control. The companies incentives to build new apartments in the big cities are very limited because of the rent control, particularly since the subsidies for construction were gradually abandoned starting in The queues in such cities have become even longer as their populations increase over time through in-migration. In municipalities with an out-migration the opposite situation has developed with high actual vacancy rates as a consequence in many cases. In particular, smaller municipalities in the north of the country (Norrland) show high actual vacancy rates. This is due to the fact that residences are durable goods and hard to move from one municipality to another. 6

7 In Sweden, municipalities own a large share of the rental apartments. Municipalities are not permitted to go into bankruptcy. To avoid bankruptcy, they do not reduce the rents to the decreased level of market rents resulting from decreased demand caused by out-migration. Thus, the rather curious situation has developed that controlled rents have become higher than market rents in such municipalities. This situation, owing to this rent control regime, has resulted in even higher vacancy rates in municipalities with an out-migration. In order to reduce high unemployment rates in several municipalities, in particular in the northern part of the country, the Swedish government put forward an extensive program for employment and housing policy with big subsidies for residential production. This subsidization policy ended in the early 90s. However, even subsequently, the government has subsidized companies that establish themselves in regions with high unemployment rates as part of an extensive regional policy program (Andersson, 2005). Because of the large subsidies for residential construction, a great deal of housing was built at the end of the 80s, mainly outside the big cities. Consequently, vacancy rates in municipalities lacking employment opportunities became even higher at the end of that period. From the end of the 90s the rent level has to some extent been adapted to the fact that the willingness to pay varies with the location of the apartments in some of the big cities. However, such an adaptation has not occurred in Stockholm. In municipalities with very high vacancy rates some rebates from the controlled rent level have been given. 3. Rent control and vacancies The vacancy situation is quite different under rent control for rental apartments in Sweden as compared to a situation with market rents, as in the US. The effects of rent control on the vacancy rates will depend on whether the region is expanding or contracting. Let us first consider an equilibrium situation, represented by the equilibrium rent level R 0 * in Figure 1A corresponding to a natural vacancy rate V0*. Then we assume that an expansion shock takes place, illustrated by a positive shift in the demand curve from D 0 to D 1 in Figure 1A. Since it takes some time to build new residences the supply curve, S 0, is quite inelastic in the short run. If the expansion shock is unexpected, a rather big increase in the rent level, to R 1, could take place in the short run. After a time lag corresponding to the time it takes to build new residences, a new long-run equilibrium will occur at rent level, R 1 *, where the new demand 7

8 curve cuts the long-run marginal cost (LRMC). The natural vacancy rate V1* corresponds to this new equilibrium rent level. Let us now assume that a controlled rent R c lower than the equilibrium rent R 0 * is introduced before an expansion takes place. Then queues will develop. No new apartments will be built, since the landlords are not permitted to lease their apartments at a rent level that covers longrun marginal costs (unless production costs are subsidized). Consider that an expansion takes place under the rent control regime. No new apartments will be built, as the controlled rents do not cover LRMC. Instead, the queues increase and simultaneously, so do the shadow values for the apartments. In such a situation, landlords have low incentives for overbuilding. Therefore, their buffer will be low or non-existent. Consequently tenants search costs for apartments will increase. Figure 1. Vacancy rates and queues for apartments under a rent control. D 1 Supply Supply D 0 LRMC D 0 R 1 R 1 * R 0 * R C Demand R 0 * D 1 R C SRMC Demand S 0 * Queue Vacancies S 0 * A. Expanding Cities B. Contracting Cities The Stockholm area is an example of an expanding region with rental apartments under rent control. There are no vacancies in older rental apartments in the Inner City. Instead, a long queue has developed there over time. It takes several years for a customer in this queue to get an older rent controlled apartment. Historical construction costs determine the controlled rent level. Since inflation has been rather high during the years, rents for older vintages of rental apartments in the Inner City are very low in relation to market rents while the rents for newly 8

9 built apartments are considerably higher. Because of rent control, the market for rental apartments will never reach equilibrium and optimum. Thus, rent control causes welfare losses as pointed out by researchers from Lindbeck up to Glaeser and Luttmer. Let us now consider a region in contraction. When a contraction shock takes place, the equilibrium rent level in a free market decreases, as illustrated by a negative shift in the demand curve in Figure 1B. If the rent level becomes lower than short-run marginal cost (SRMC), some apartments should be demolished. Then a new equilibrium rent level equal to SRMC will be established. At such an expected rent level, landlords incentives for a buffer of apartments are very low. Marston (1985) found that excess labor in an area will vanish within a year owing to mobility. But while workers are mobile, properties are not. An actual high unemployment rate could vanish rather quickly from out-migration from a municipality. But a high actual vacancy rate may persist for a long time since properties are immobile and durable. The property market in a municipality with out-migration can be plagued by overbuilding for a long time when actual demand for residential space becomes lower than expected. Thus in such regions the actual vacancy rate could lie at a higher level than the equilibrium natural vacancy rate for a long time, contrary to what is the case in expanding regions, unless apartments are demolished. Let us now assume that the rent control is introduced in the first equilibrium situation, i.e. before a contraction shock takes place. Then, a queue develops in the same way, as illustrated in Figure 1A. When a contraction shock takes place, vacancies develop, illustrated by a negative shift in the demand curve in Figure 1B. Because of rent control the rather curious situation occurs that rents will be higher than market rents. Then actual vacancies will be greater than in a situation with no rent control. The reason for the sticky high rents under rent control is that the municipalities own a large share of the rental apartments. They are not allowed to go into bankruptcy. Because of this very special situation in Swedish municipalities with out-migration another type of welfare loss than pointed out by Lindbeck, Glaeser and Luttmer, etc., occurs. These losses can be measured by the consumer surplus illustrated by the shaded triangle in Figure 1B. The following hypotheses can be derived from the analysis presented above: The natural vacancy rates in rental apartments under the Swedish rent control in expanding municipalities are very low, perhaps even negative. 9

10 The natural vacancy rates in rental apartments under rent control in contracting municipalities are also very low while the actual vacancy rates may be relatively high. The actual vacancy rates are mainly determined by factors such as unemployment rates, migration, rent levels, and construction and demolition of residences. These hypotheses will be empirically tested in this paper. 4. Modeling natural vacancy rates 4.1. The models The natural vacancy rates on the Swedish residential housing market are estimated using the following model (see Voith and Crone, 1988). The model is an autoregressive model where the vacancy rate (V) is a function of past vacancy rates. V i,t ε i,t = α i i = ρ ε + ε i,t 1 i,t + β t + µ i,t (2) By substituting, we get V = α (1 ρ ) + β + ρ V + µ (3) i,t i i t i i,t 1 i,t If α i is jointly insignificant, there is no difference in natural vacancy rates across housing markets in Sweden, and if β i is jointly insignificant there is no difference in natural vacancy rates over time. Our hypotheses are that both can be rejected, i.e., there is a difference in natural vacancy rate both across markets and over time. If ρ i is equal to zero, the shocks occur in one single period, i.e., there is no persistence, which is not very likely. We will also test the hypothesis that persistence differs across housing markets. There is nothing in the theory that says that the persistence should be different across markets. However, as the markets can be imperfect to different degrees, persistence can differ. As a second step, we will explain the variation in natural vacancy rates in the same manner as in Rosen and Smith (1983). V*=f(X) (4) where V* is equal to natural vacancy rate and X is equal to determinants of the natural vacancy rates. In their model, the determinants of the natural vacancy rate are mobility rate, rent dispersion, average rent, change in housing stock and change in population. Rent 10

11 dispersion is defined as the percentage standard deviation of the mean rent. Rosen and Smith also tested whether racial segregation and the size of the market affected the natural vacancy rates. However, they did not find such evidence. Although Sweden has a rent control policy in all municipalities, the degree of market orientation varies from municipality to municipality. In some municipalities, market orientation has gone far and the negotiated rents are close to the market rents, even if the rent level is a result of negotiation between the local tenant organization and the municipal public housing company. Our hypothesis is that the dispersion between the actual rents and the market rents will affect the natural vacancy rate; specifically, rent control will reduce the natural vacancy rate (see e.g. Benjamin and Sirmans, 1994) Data We hypothesize that the following variables are important as explanations for the variation in the municipalities vacancy rates: net migration, residential construction and demolition, and rent levels. Data for most of the variables presented above are required for 280 Swedish municipalities for a certain period of time such as for the period or alternatively for 100 working place areas. Most of the data are already available or relatively easy to obtain. Data for the municipalities population changes are available. Thus, data for the net migration could be calculated. Also data for the municipalities unemployment rates are available as well as for residence constructions and the demolitions. It will of course be important to get hold of the rent levels for the apartments in the different municipalities for the different years. The data available are based on a national survey conducted by Statistics Sweden for the average rent level in a number of municipalities (45 66 depending on year) and in all counties. The rent level we are using is measured on the municipality level if possible, otherwise county level. Table 1 shows the data set. 11

12 Table 1. Data set Variable Unit Abbreviation Average Standard Deviation Vacancy Rate Percent Vac-rate Construction Rate Percent Constr Demolition Rate (constructed)* Percent Demo Population Number Pop Net Migration Percent Net-migr Taxable income per capita Kronor Inc Rent per square meter Kronor Rent *Demolition (constructed) = new construction change in housing stock The average vacancy rate across all municipalities and over time is as low as 5.8 percent with a standard deviation in the same size range. That is, the variation is considerable. The maximum actual vacancy rate is as high as 56 percent and the lowest 0 percent. Construction of new apartments on average is as low as 0.3 percent (measured as new construction per housing stock). However, the maximum residential construction is 4 percent of the housing stock. Naturally, demolition of apartments is also low but the highest demolition rate is as high as 8 percent of the housing stock. The average population size is approximately 30,000 ranging from 2,575 to 765,044. The average taxable income per capita is almost SEK 115,000 but the variation is considerable, with a maximum of SEK 400,000. As you can see, there is some variation in the macroeconomic variables, but when it comes to the rent level per square meter, variation is very low. The average rent level is SEK 691 per square meter and year and the standard deviation is only 8 percent around its mean value. The minimum rent level is SEK 521 and the maximum SEK 900. As the average apartment is around 65 square meters in area, the average total rent per year is SEK 45,000 or 40 percent of the taxable income per capita. Table 2 shows correlations between the variables. Vacancy rate is positively related to demolition. That is, demolition is high in municipalities where the vacancy rate is high. The opposite is true for construction. The rent level and income per capita are also negatively related to the vacancy rate. Hence, in municipalities with high rent level and high income per capita, vacancy rate is expected to be low. The correlation between income and rent level is positive and high, but we perhaps expected an even higher correlation. 12

13 Table 2. Correlation Matrix ( ). Vac-rate Constr Demo Net-mgr Rent Inc Vac-rate Constr Demo Net-migr Rent Inc As can be seen in Table 3, vacancy rate varies considerably over time. The variation across municipalities is also large. The highest average vacancy rate was in The highest construction rates were prior to this and the highest demolition rates were after this. Table 3. Descriptive Statistics. Average and coefficient of variation over time across municipalities. Vacancy Construction Demolition Net migration Average Coeff.Var. Average Coeff.Var. Average Coeff.Var. Average Coeff.Var % % % ,52% 6, % % % ,08% -31, % % % ,17% -14, % % % ,36% -1, % % % ,32% -2, % % % ,28% -2, % % % ,09% -8, % % % ,02% 29, % % % ,19% 3, % % % ,24% 2, % % % ,20% 2, Estimations of natural vacancy rates The empirical model of equation 3 is equal to (following Voith and Crones notation): 4 T 4 i,t = α j(1 ρ j)d j + β tm t + ρ jvj,t 1 + i,t (5) j= 1 t= 2 j= 1 V µ where: D j is a vector of dummy variables that =1 if municipality=i, 0 otherwise M t is a vector of dummy variables that=1 if time=t, 0 otherwise. Table 4 shows the estimated natural vacancy rates on the Swedish residential housing market. The parameters α+β equal the natural vacancy rate. The first model assumes that the α do not differ across markets and have no time-specific component. In the second model, we relax the 13

14 assumption about a time-specific component. The third and fourth models assume that the natural vacancy rates differ across markets. Here we test whether the natural vacancy rates are different across 280 municipalities and 100 labor markets, respectively. Table 4. Least Square Estimates of α, ρ and β. Model 1 Model 2 Model 3 Model 4 ρ (78.47) (78.35) (28.68) (54.06) α(1-ρ) (12.32) (-0.39) β (8.53) (3.54) (6.86) β (5.91) (2.65) (4.83) β (7.08) (4.46) (6.21) β (9.08) (7.53) (8.56) β (8.28) (8.17) (8.24) β (5.66) (6.20) (5.81) β00 - Default Default Default β (1.66) (-0.67) (0.90) β (0.95) (-2.76) (-0.26) β (-0.01) (-5.08) (-1.68) β (1.83) (-4.46) (-0.25) R2-adj Moran s I t-statistics within parentheses. The market-specific estimates are not shown in the table. Almost 70 percent of the variation in vacancy rates can be explained by the vacancy rates in the previous period. Approximately 80 percent of the vacancy rate will persist in the following period in the models lacking market-specific and time-specific components. However, its statistical significance is reduced if we introduce market-specific and timespecific components. That is, shocks do not occur only in a single period. However, no difference in persistence of shocks across markets can be detected (not shown in the table). Furthermore, differences in natural vacancy rates over time are clearly significant. However, 14

15 the differences in natural vacancy 1 rates across markets are very small, especially among the three largest labor markets (see figure 2). That is, the variation in natural vacancy rates is not significant among Stockholm, Gothenburg and Malmö. In all three metropolitan areas, the natural vacancy rates are close to zero and in some years even negative. The natural vacancy rates in the rest of the country are significantly higher. Most surprisingly, the overall natural vacancy rate is very low. Compared to e.g. Gabriel and Nothaft (1988), Jud and Frew (1990) and Rosen and Smith (1983) our results indicate that the natural vacancy rates are much lower in the Swedish housing market than in the residential market in U.S. The American studies estimated the average natural vacancy rate to be in the range of percent compared to our estimate of only 1 percent. Our fixed effects model indicates that the natural vacancy rates in Sweden are slightly higher compared to a model without fixed effects for municipality and labor market. We estimate the average natural vacancy rate in the Swedish rental housing market (with a fixed effects model) to be around 2 percent, with a spread from 0 to 4 percent. However, it is reasonable to expect that the natural vacancy rate is lower in a nation with rent control (Muller, 1991). Figure 2. Natural Vacancy Rates across markets and over time ( ). 8% 6% 4% 2% 0% -2% -4% Stockholm Göteborg Malmö Rest Why are the natural vacancy rates so low for rental apartments? The market for rental apartments is under a rent control regime. 1 Natural vacancy rates are constructed with the estimates from model 3 (Nat_vac1) and 4 (Nat_vac2), that is, with both market-specific and time-specific components. The natural vacancy rates are naturally highly correlated to the actual vacancy rate. They are higher when we estimate the natural vacancy rates per municipality instead of per labor market. 15

16 The mobility from a rental apartment is low, i.e., the probability that tenants vacate their rental apartments is low (low incidence). A low expected rent level due to rent regulation will lead to a low natural vacancy rate on average in the big cities. Higher expected absorption rate will lead to lower natural vacancy rates, i.e., reduce both the duration and the incidence Explaining the vacancy rates Our basic empirical model is similar to Rosen and Smith s model the variation in natural vacancy rates will be a function of a number of determinants: V * i,t = α + βx + ε (6) i,t where V* is the natural vacancy rate and X includes the variables construction rates, demolition rates, net migration rates, the degree of market orientation, and population size. We introduce the concept of market orientation, which is the degree to which the controlled rent is market oriented. The market orientation equation is: R i,, t = α + βinci t + ε (7) Table 5. Rent and income per capita Without Fixed Effects Fixed Municipality Effects Fixed Labor Market Effects α (138.76) (132.84) (148.94) β (37.60) (61.45) (37.68) R2-adj Moran s I t-statistics within parentheses. The absolute difference between actual rent level and predicted rent level (residual, ε, in the equation) is a measure of market orientation (variable name: InverseMO). A large difference between the actual rent level and what can be explained by the income per capita indicates that in the particular market controlled rents are far from markets rents, that is, the market orientation will be low and InverseMO will be high. However, the estimate of market orientation only measures the relative differences among municipalities, not the actual rent level deviation from market rents. 16

17 We have also designed a binary variable to indicate whether the actual rent is above the estimated market rent (variable name: RAMR) and an interaction variable between RAMR and InverseMO. Table 6. Correlation between vacancy rates and market orientation. InverseMO Vac Nat_vac1 Nat_vac2 InverseMO 1.00 Vac Nat_vac Nat_vac The degree of market orientation has a positive relationship with both actual and natural vacancy rate. That is, a market with low degree of market orientation will be associated with low vacancy rates. Table 7 (a and b) provides estimates for equation 6. The first model uses natural vacancy rate as the dependent variable (municipality-specific components) and the second uses actual vacancy rate. Table 7a Determinants of Natural Vacancy Rates (Dependent variable=nat_vac1) Model A1 Model A2 Model A3 Model A4 Model A5 Net-migr (-11.78) (-10.34) (-9.51) (-6.32) (-6.98) Rent (-23.11) (-18.98) (-14.74) (-23.80) InverseMO (-8.75) (-4.53) (-14.10) Constr (-18.90) (-17.84) Demo (7.75) (9.33) Pop (-8.50) (-3.19) RAMR (0.21) RAMR*InverseMO (15.57) Constant (80.78) (30.07) (26.31) (23.50) (28.82) R2-adj Moran s I t-statistics within parentheses. 17

18 Table 7b Determinants of Actual Vacancy Rates Model C1 Model C2 Model C3 Model C4 Model C5 Net-migr (-8.50) (-7.27) (-6.29) (-2.95) (-3.15) Rent (-14.01) (-10.67) (-6.13) (-11.55) InverseMO (-9.45) (-5.41) (-9.05) Constr (-17.56) (-16.63) Demo (8.78) (9.56) Pop (-7.99) (-4.90) RAMR (1.23) RAMR*InverseMO (7.61) Constant (61.45) (19.06) (16.19) (12.76) (15.10) R2-adj Moran s I t-statistics within parentheses. Almost 40 percent of the variation in natural vacancy rate can be explained by net migration, rent level, market orientation, construction, demolition and whether or not the controlled rent is above or below the estimated market rent. However, only 25 percent of the variation in actual vacancy rates can be explained by the same variables. A higher degree of market orientation will lead to a higher vacancy rate (both actual and natural) if the controlled rent is higher than the expected market rent. Controlled rents will lead to lower natural vacancy rates; however they will be higher if the controlled rents are higher than the market rents. If net migration increases, natural vacancy rates will be lower. If net migration increases by 1 percent, the expected result is that natural vacancy rate will decrease by 0.15 percent. The level of rents will have a negative effect on natural vacancy rate. 18

19 Today s construction and demolition is a consequence of the vacancy rate. However, construction during a previous period may be a determinant of vacancy rate. For example, the correlation between construction rate and vacancy rate is negative in the case of no lags, but the correlation between them is positive with a lag of 4 years. That is, previous construction has a positive effect on today s vacancy rate. On the other hand, net migration is more directly a determinant of the vacancy rate. If inmigration is larger than the out-migration today, vacancy rate is expected to be lower. For example, vacancy rate is expected to decrease by 0.15 percent if net migration increases by 1 percent. Since Rosen and Smith estimated the mobility effect to be in the range , our results are in accordance with their results. Municipalities with a high vacancy rate have a low construction rate and a high demolition rate and vice versa. In municipalities where they have increased the demolition by 1 percent, vacancy rate is approximately 1 2 percent higher. However, the key policy question in the municipality is to increase net migration. In the next model we explore whether the determinants affect the vacancy rate differently depending on whether the municipality has grown or not over the whole period. We use the same models as for the previous estimation but now we introduce a new dummy variable equal to 1 if the municipality has had a population growth over the period, otherwise 0. Table 8. Vacancy rates with population growth. Model D5 Model E5 Net-migr (-4.91) (-1.15) Rent (-21.26) (-8.47) InverseMO (-14.11) (-8.55) Constr (-7.51) (-7.45) Demo (7.64) (7.37) Pop (0.97) (-0.41) RAMR (1.02) (1.08) RAMR*InverseMO (15.71) (8.15) Popgr (-7.90) (-3.35) Popgr*Rent (6.56) (2.19) Popgr*InverseMO (3.65) (2.60) Popgr*Constr (3.10) (3.94) Popgr*Demo (-1.59) (1.12) Popgr*RAMR (-1.06) (-0.81) Popgr*RAMR*InverseMO (-4.06) (-2.82) Constant (25.57) (11.48) R2-adj Moran s I t-statistics within parentheses. 19

20 What do these results suggest? The major conclusion is that the determinants do affect the vacancy rate differently depending on whether the municipality is growing or not. For example, the rent level has a negative effect on vacancy rate if the municipality has had a negative population growth over the period. That is, a high level of controlled rents decreases the vacancy rate. That is also true if the municipality has had a positive population growth but to a lower degree. The degree of market orientation has a positive effect on vacancy rate; that is, if the controlled rents differ greatly from expected markets rents, we expect that the actual and natural vacancy rates would be lower. However, its impact is larger in municipalities with no population growth. The estimation also suggests that if the controlled rent level is above the market rents, vacancy rates will be much lower in areas with a positive population growth than in a region with no growth. 5. Summary and Conclusions We have estimated that the natural vacancy rates for rental apartments under the Swedish rent control regime are 1 percent on average, a very low figure compared to the 4 to 12.1 percent estimated for residential markets in the US. While the natural vacancy rates were found to be high in rapidly growing cities in the US, they are very low in large, expanding Swedish cities, even lower than in municipalities with net out-migration. The actual vacancy rates, however, are relatively high in municipalities with net out-migration. The results are in line with our hypothesis that rental apartments are under a rent control regime in Sweden. Thus, the low or negative natural vacancy rates for large, expanding cities reflect that queues for rental apartments have developed under the rent control. The high actual vacancy rates in municipalities with an out-migration reflect the fact that residences are immobile and durable. Furthermore, rent control has resulted in the rather curious situation where rents in such municipalities are higher than market rents. This means that actual vacancies are more than in a situation with no rent control. The reason for these sticky high rents under rent control is that the municipalities own a great share of the rental apartments and that they are not allowed to go into bankruptcy. The high vacancy rates resulting from rent control cause welfare losses. Another main task in this paper has been to estimate the influences of various determining factors on the actual vacancy rates, such as net migration, education level, employment rates, demolitions, residence construction and the rent levels. In line with our hypothesis, we find that these factors play an important role in explaining the actual vacancy rates, in particular migration. 20

21 What are the policy implications of our study? First, rents should be deregulated to allow the market to be able to reach equilibrium and an optimum. Second, the policy makers in municipalities with high vacancy rates should consider measures to discourage out-migration. For this the employment rate is a key policy parameter. Glaeser et al (2001) and Graves (2003) have shown that various amenity variables may play an important role in attracting both companies and people to a city. The attention of policy makers as well as researchers should be directed towards possibilities of enhancing various amenity variables, such as the quality of schools, existence of colleges, low crime rate, dwellings with nice views and pleasant parks, as well as attractive cultural and sports events. Table 8. Simulation results with full market orientation (percent). Growth Contraction Natural vacancy rates RAMR Predicted Simulation RBMR Predicted Simulation Actual vacancy rates RAMR Predicted Simulation RBMR Predicted Simulation Predicted averages are based on coefficients in table 6 (Model D5 and E5) and table 8. Simulation averages are based on coefficients in table 7 (Model A6 and C6) and table 8 with InversMO=0 and RAMR=0. RAMR is equal to controlled rents above market rents and RBMR is equal to controlled rents below market rents The simulation results of Table 8 indicate that the natural and actual vacancy rate is higher than in a situation with full market orientation if the controlled rents are higher than the expected market rent (RAMR). The result is the opposite if the controlled rents are below market rents (RBMR). As Wiltshaw (1996) argues, if the resulting spatial pattern is not random, the conclusions are likely to be flawed. Spatial dependency 2 in a regression model can be detected with either the spatial weight matrix or the direct specification of the covariance matrix (Dubin, 1998). We used the weight matrix for urban and real estate analysis (e.g. Pace et al., 1998; Dubin et al., 1999, and Wilhelmsson, 2002). Among different statistical tests with weight matrix, we chose Moran s I, which is empirically estimated in e.g. Bogdon and Can (1997), de Graaff et al. (2001), and Wilhelmsson (2002 and 2004). To detect spatial autocorrelation, Moran s I can be calculated using the row standardized weight matrix of inverse square distances. Importantly, 2 There are two types of spatial effect: spatial dependency and spatial heterogeneity. However, according to De Graaff et al. (2001), due to difficulties in separating them as well as for other reasons, they should be handled jointly. 21

22 if the values of the Moran s I statistic are large, it suggests that neither model captures the underlying spatial relationship very well. This merits further research. 22

23 References Amy, Khor, Ming, Yu Shi and Yan, Lim Lan (2000). The natural vacancy rate of the Singapore office market. Journal of Property Research, Vol 17, No 4. Andersson, Roland (1999). Attractive Cities. An Economic Analysis, Byggforskningsrådet. Andersson, Roland (2005). The Efficiency of the Swedish Regional Policy. Annals of Regional Science. Andersson, Roland and Bo Söderberg (2006). Deregulation of the Swedish Rent Control. Why and How? Working paper, Dept of Infrastructure, Royal Institute of Technology, Stockholm. Arnott, Richard (1988). Rent Control: The International Experience. Journal of Real Estate Finance and Economics, Vol. 3, No 3. Arnott, Richard (1995). Time for Revisionism on Rent Control? Journal of Economic Perspectives, Vol. 9, No 1. Arnott, Richard (2003). Tenancy Rent Control. Swedish Economic policy Review, 10. Belsky, Eric (1992). Rental Vacancy Rates. A Primer. Housing Policy Debates, Vol.3, No 3. Benjamin, J. D. and G. S. Sirmans (1994). Apartment Rent: Rent Control and other Determinants. Journal of Property Research, Deng, Y., S. Gabriel and Frank Nothoft (2003). Duration of Residence in the Rental Housing Market. Journal of Real Estate Finance and Economics, 26: 2/3, Englund, Peter, Åke Gunnelin, Patric Hendershott and Bo Söderberg (2006). Adjustments in Property Space Markets. Estimates from the Stockholm CBD Office Market. Fallis, G. and B. L. Smith (1984). Uncontrolled Prices in a Controlled Market: The Case of Rent Controls. American Economic Review, Vol. 74, pp Gabriel, Stuart and Frank Nothoft (2001). Rental Housing Markets, the incidence and Duration of Vacancy, and the Natural Vacancy Rate. Journal of Urban Economics, 49, Gallagher, Mark and Antony Wood (1999). Fear of overbuilding in the office Sector: How Real is the Risk and Can We Predict It? Journal of Real Estate Research 23

24 Glaeser, Edward and Erzo Luttmer (2001). The Misallocation of Housing under Rent Control. American Economic Review, Glaeser, E., J. Kolko, and A. Saiz, (2001). Consumer City. Journal of Economic Geography, Vol. 1, pp Graves, Philip and Peter Linneman, (1979). Household migration: Theoretical and Empirical Results. Journal of Urban Economics 6, pp Graves, Philip (1979). A Life-Cycle Empirical Analysis of Migration and Climate, by Race. Journal of Urban Economics, Vol. 6, pp Graves, Philip (2003). Linking regional science and urban economics. Long-run interactions among preferences for amenities and public goods. Grenadier, Steven (1995). Valuing lease contracts: A Real Option Approach. Journal of Financial Economics, 38, pp Grenadier, Steven (1996). The Strategic Exercise of Options: Development Cascades and Overbuilding in Real Estate Markets. The Journal of Finance, 51 (5), pp Gunnelin, Åke (2000). Optimal Capacity Choice and Overbuilding. published in Real Options in Real Estate (PhD-thesis). Royal Institute of Technology, Stockholm. Hendershott, Patric and Donald Haurin (1988). Adjustments in the Real Estate Market. AREUEA Journal, 16: Hubert, Franz (2002). Rent Control: Academic Analysis and Public Sentiment. Swedish Economic Policy Review, 10. Johnson, G. and P. Layard (1986). The natural rate of unemployment: explanation and policy. In Handbook of Labor Economics, 2, (O. Asarton, Stephenhenfelter and P. layard, Eds), Elsevier, Amsterdam/New York. Jud, Donald, Joh Benhamin, and Stacy Sirmans (1996), What Do We Know about Apartments and Their Markets? The Journal of Real Estate Research, Vol. 11, No 3. Lindbeck, Assar (1967). Rent Control as an Instrument of Housing Policy published in The Economic Problems of Housing, (editor A. A. Nevitt) International Economic Association, MacMillan, pp Linneman, Peter and Philip Graves. (1983), Migration and Job Change: A Multinomial Logit Approach. Journal of Urban Economics 14, pp

25 Marton, Stephen (1985). Two Views of the Geographical Distribution of Unemployment. The Quarterly Journal of Economics, Vol. 100 No 1, Muller, R. (1991). Rent Control and Vacancy Rates in Canada. QSEP Research Report 275, McMaster University, Canada. Rosen, Kenneth and Larry Smith (1983). The price adjustment process and the natural vacancy rate. American Economic Review. 73, Tse, Raymond and Bryan MacGregor (1999). Housing Vacancy and Rental Adjustment. Evidence from Hong Kong. Urban Studies, Vol 36, No 10, Tse, Raymond and Dominique Fischer (2002). Estimating Natural Vacancy Rates in Office Markets Using a Time-varying Model. Journal of Real Estate Literature. Tse, Raymond and James Webb (2003). Models of Office Dynamics. Urban Studies, Vol 40, No 1, Turner, Bengt and Stephen Malpezzi (2003). A review of empirical evidence on the costs and benefits of rent control. Swedish Economic Policy Review, 10, Voith, R. and T. Crone (1988). National Vacancy Rates and the Persistence of Shocks in U.S. Office Markets. AREUEA Journal, Vol.16(4), pp