THE IMPACT OF VACANCY DECONTROL IN NEW YORK CITY:

THE FIRST ESTIMATES FROM THE 1996 HOUSING AND VACANCY SURVEY*

By
Edgar O. Olsen
Department of Economics
University of Virginia
Charlottesville, VA 22903

November 1997

Introduction

The reauthorization of rent regulation in New York City was recently considered by the New York State Legislature and phasing out these regulations was one of the options seriously discussed. To help policymakers and citizens understand the likely effects of phasing out rent regulations, the New York City Rent Guidelines Board commissioned this study of the effects of vacancy decontrol.[1] This method has been used often to eliminate rent regulations whose costs were thought to outweigh their benefits. It became a leading proposal in the recent debate when Governor Pataki suggested it as a compromise between those who favored the continuation of the current rent regulations and those who wanted to eliminate these regulations completely within a few years.

It is important to realize at the outset that the distributional effects of vacancy decontrol are likely to be quite different from the distributional effects of the immediate deregulation of rents. Households currently living in the rent regulated units with the largest gap between market and actual rent (hereafter the rent discount) are likely to be among the biggest losers from immediate deregulation of rents. However, current occupants of rent regulated units would not necessarily incur any loss from vacancy decontrol. Indeed, they could benefit from it. If they did lose from vacancy decontrol, the current occupants of rent regulated units with the largest rent discounts would not necessarily incur greater losses than those who receive smaller discounts.

Some examples will make clear that the losses from vacancy decontrol do not depend on a household's current rent discount. Some households would leave New York City permanently the next time that they move whether the current regulations are continued or vacancy decontrol is implemented. The current rent regulations would apply to these households until they leave the City in either case. So vacancy decontrol would impose no direct cost on them, and it would indirectly benefit them for the remainder of their stays in the City. Vacancy decontrol would lead to higher levels of public services without increased tax rates because it would increase the market values of properties containing newly decontrolled units and hence ultimately increase their assessed values. Current residents of rent regulated units who would become homeowners in New York City for the rest of their lives on their next move and those who would move to publicly subsidized housing for the rest of their lives would gain from vacancy decontrol for the same reason.

Ideally this study would have estimated a wide variety of short-run and long-run effects of vacancy decontrol implemented on the expiration of the old law on June 15, 1997. Although the paper contains estimates of some long-run effects, the small amount of time between the availability of current data and the deadline for reauthorizing rent control dictated concentrating on a more modest goal, namely the short-run effect on rents of vacancy decontrol implemented in the spring of 1996 when data from the most recent New York City Housing and Vacancy Survey were collected. Since changes in the state of the housing market since then have been modest, this should provide an accurate picture of the short-run effect on rents of vacancy decontrol if it had been implemented on June 15,1997.

Estimating these rent increases requires predicting which rent regulated units will be vacated and reoccupied between 1996 and 1998 and the market and actual rents of these units under current rent regulations when they are reoccupied. In order to provide timely information, the simplest approach was initially used to provide estimates of the effect of vacancy decontrol based on the most current data available. Specifically, data from the 1991 and 1993 Surveys were used to predict how many households of various types would move in the two years after the date of the 1996 Survey and data from the 1996 Survey were used to estimate the market rents of regulated units at the time of the Survey.[2] The preliminary prediction of the increase in rent that would result from vacancy decontrol was simply the difference between the mean market rent of the units occupied by households who were predicted to move and the mean actual rent of these units at the time of the 1996 Survey. No account was taken of the increase in rent that was allowed under the prevailing rent regulations when an apartment is vacated and reoccupied or the increase in market rent due to the loss of the discount typically received by long-standing tenants and general inflation in housing prices.

Since the time that these preliminary estimates were produced, it has been possible to make more accurate estimates taking these and other factors into account. These refinements did not affect the major qualitative conclusion of the preliminary analysis. The majority of rent regulated units vacated over the two years have rents at or close to market levels. Large differences between market and actual rent are the exception rather than the rule. As a result, vacancy decontrol would lead to modest rent increases in most cases. The refinements did, however, enable us to make much more precise estimates of almost all magnitudes of interest, especially the number of extremely large rent increases and the distribution of these rent increases across different types of households.

Section 2 describes the methods used to predict the market rent of each rent regulated unit in 1996 and Section 3 the method used to predict which apartments occupied in 1996 would be vacated and reoccupied by 1998. Section 4 describes several methods for predicting the market and actual rents of these units after they are vacated and reoccupied under the rent regulations that existed prior to June 15, 1997. The excess of the predicted market rent over predicted actual rent is the predicted rent increase resulting from vacancy decontrol. The fourth section presents the average values of the alternative estimates of these rent increases for the entire city, the five boroughs, and thirty smaller geographical areas. These rent increases are the maximum losses from vacancy decontrol to new occupants of previously regulated units. Section 5 contains estimates of how the average loss varies with household income and size and the survey respondent's age and race. Sections 6 and 7 present estimates of the long-run effects of vacancy decontrol on the types of households living in the City and each of its boroughs and the market rents of unregulated units. Section 8 summarizes the results.

2. Estimation of the Market Rents of Regulated Units Prior to Vacancy

Vacancy decontrol allows the owners of previously regulated apartments to increase the rents of vacated units to market levels. These market rents must be predicted in order to determine the effects of vacancy decontrol. This is done in two steps. First, the market rent of each rent regulated apartment in the 1996 Survey is estimated. Second, the estimated market rent of each unit that is predicted to be vacated and reoccupied between 1996 and 1998 is adjusted for the change in market rent that would occur between the time of the survey and the time that the unit would be reoccupied. This change is due to the loss of the discount typically received by long-standing tenants and the general inflation in housing prices. This section deals with the first step, and the fourth section with the second step.

The 1996 Housing and Vacancy Survey collected information on about thirty housing and neighborhood characteristics of 3018 occupied apartments that were not subsidized or subject to any form of rent regulation. In explaining differences in the gross rent of different units, the analysis in this section uses fourteen of these housing and neighborhood characteristics: number of bedrooms, number of other rooms, condition of external walls, condition of windows, condition of floors, overall condition of building, number of units in the building, number of stories in building, existence of passenger elevator in building, year structure was built, presence of plumbing and kitchen facilities, existence of broken or boarded windows on the street, and location by subborough area. It also accounts for one characteristic of the occupant that is important for predicting the market rent of a unit that is vacated and reoccupied, namely the number of years that the tenant has lived in the unit. It is a well established empirical regularity that the mean rent paid by tenants who have lived in a unit longer is lower than the mean rent of tenants who have lived in units with the same observed characteristics for a shorter period. The other characteristics in the survey were not used in this preliminary analysis because the information on these characteristics was not reported in a significant number of cases. Time constraints precluded exploring the possibility of improving the predictions of the market rents of rent regulated units by incorporating these characteristics into the analysis.

The initial prediction of the market rent of each regulated apartment is an estimate of the mean market rent of unregulated units that are the same as this apartment with respect to the preceding fifteen characteristics. If regulated units that are the same as unregulated units with respect to these observed characteristics tend to be better than unregulated units in other respects, this procedure will tend to understate the market rents of regulated units and hence the increase in rents that will result from vacancy decontrol. However, the opposite seems more likely. That is, the regulated units are likely to be less well maintained in ways not captured in the data due to the incentives facing landlords. So the estimates herein are likely to overstate the increase in rents that will result from vacancy decontrol.

The most straightforward way to estimate the mean gross rent of unregulated apartments with a particular combination of the fifteen characteristics is to calculate the mean gross rent of the unregulated apartments in the sample with this combination of characteristics. Unfortunately, for the overwhelming majority of combinations of characteristics, there are too few unregulated apartments in the sample for the sample mean to be a reliable estimator of the population mean. Indeed, it is entirely possible that some combinations of characteristics that exist in the sample of rent regulated apartments in New York City are not represented in the sample of unregulated units, making it impossible to estimate the market rents of regulated units in this way.

The universal solution to this problem in statistical analysis is to estimate the means based on general assumptions about the relationship between the means for various combinations of characteristics. The analysis in this paper is based on the common assumption that the effect of any one variable on mean market rent is independent of the values of other characteristics. For example, the difference between the mean market rents of two and three bedroom apartments with a particular set of other characteristics is the same as the difference between the mean market rents of two and three bedroom units with any other combination of characteristics. In other words, the extra bedroom adds the same amount to the market rent of a unit with any set of other characteristics.

Table 1 reports the least squares estimates of the equation explaining how mean gross market rent of unregulated apartments varies with the fifteen characteristics. Mean market rent is assumed to be a linear function of dummy variables representing these characteristics. Dummy variables take on a value of 1 for certain values of a characteristic and 0 otherwise. If a characteristic is represented by a single dummy variable, the coefficient of that variable tells us the difference between the mean gross rent of apartments with a value of the characteristic corresponding to 1 and those with a value of the characteristic corresponding to 0, among apartments that are the same with respect to the other characteristics included in the equation. For example, it is estimated that the mean market rent of apartments that had no problems with external walls was about $12 per month greater than the mean rent of otherwise similar units that had such problems. Many characteristics are represented by a set of dummy variables. In these cases, estimation requires that one of the dummy variables be excluded from the equation and the coefficient of an included dummy variable tells us the difference between the mean market rent of an apartment in this category and in the omitted category, among apartments that are the same with respect to other characteristics included in the equation. For example, it is estimated that the mean rent of apartments that had been continuously occupied since 1981 to 1985 was about $71 per month greater than the mean rent of otherwise similar units that had been continuously occupied since before 1981.

In estimating this equation, we deleted 115 observations for which the exact gross rent was not reported and an additional 102 observations for which at least one of the housing characteristics was not reported.[3] This amounts to seven percent of the original sample. With more time, it would have been possible to include these observations in the analysis by using more complicated statistical procedures.

The reported gross rents of the unregulated apartments used in estimating this relationship varied from $80 to $2500 per month. Based on the estimated equation, the predicted mean gross rents for the combinations of the fifteen characteristics represented in the sample of unregulated units range from $130 to $1967 per month. The mean of the reported and predicted gross rents for all observations in the sample used to estimate the equation is $805 per month.

As usual in estimating statistical relationships, some estimated coefficients have unexpected signs or relative magnitudes. This is may be due to correlation between the variables included in the equation and other determinants of market rent. For example, it is estimated that apartments in buildings built prior to 1901 rent for more than otherwise similar apartments that were built between 1947 and 1959. This might be because the older buildings that were still used for residences tend to be the best built units of their vintage and have been substantially rehabilitated in recent years. They may be better than the more recently built units with respect to many characteristics not included in the equation. Even if there were no correlation between included and excluded determinants of market rent, it is to be expected that each estimated coefficient will have the opposite sign from the true coefficient in some samples. This is more likely to occur if the true coefficient is small, which might explain the negative estimated coefficients for the variables No Problems with Windows and No Problems with Floors.

Under the assumption that the relationship between mean market rent and the explanatory variables included in the equation is the same in the regulated and unregulated sector, good statistical practice dictates the inclusion of all variables in the prediction equation. One should not reestimate the equation deleting variables whose coefficients have unexpected signs or are statistically insignificant, as is sometimes done. Our estimate of the market rent of each rent regulated unit in 1996 is obtained by substituting the values of the dummy variables corresponding to its characteristics into the equation defined by the results in Table 1.

3. Estimation of the Propensity of Occupants of Regulated Apartments to Move

Only rent regulated apartments that experience a change in occupant are directly affected by vacancy decontrol. So to estimate the effect of vacancy decontrol over the two years after its implementation, it is necessary to estimate which units will change hands. The likelihood that different households will move can be quite different because, for example, some receive large rent discounts in their current regulated apartment while others receive no rent discount and some are young and restless while others are older and more settled. Therefore, we will want to estimate the probability of moving separately for households with different characteristics.

Since the preliminary version of the 1996 Housing and Vacancy Survey available for this analysis did not contain the identification numbers that would make it possible to match apartments in this survey with those in the 1993 Survey, it is impossible to use data from the two most recent surveys to estimate the fraction of households of each type living in rent regulated apartments at the time of the 1993 Survey who moved before the time of the 1996 Survey. Instead we use the 1991 and 1993 Surveys to estimate these fractions for households living in rent regulated apartments at the time of the 1991 Survey, and we assume that the fraction of households of a particular type who will move between 1996 and 1998 is the same as the fraction who did move between 1991 and 1993.

The first step in estimating the probability of moving for households of each type is to determine which households in the sample moved between 1991 and 1993. This cannot be done with certainty. It must be inferred from the answers to two questions. Each survey reports whether an apartment is occupied at the time of the survey and the respondent's answer to the question of when he or she moved into the unit. If the apartment was occupied in 1991 and vacant in 1993, we can be certain that the household living in it in 1991 moved. If the apartment was occupied in both years, there are two possibilities. At least one occupant remained in the unit over this period, or it was vacated and reoccupied. If the apartment was occupied in 1991 and 1993 and the respondent in 1993 moved into the unit before 1991, then it must have been occupied continuously over this period. If the apartment was occupied in 1991 and 1993 and the respondent in 1993 said that he or she moved into the unit in 1991, the conclusion is ambiguous. It is possible that another person lived in the apartment at the time of the 1991 Survey and vacated the unit during the year and that the respondent in 1993 occupied it before the end of 1991. However, since this is likely to be a rare occurrence, I assumed that all units of this sort were continuously occupied. If the unit was occupied in 1991 and 1993 and the respondent moved into the apartment in 1992 or 1993, I assumed that the unit was vacated and reoccupied. However, this was surely not true in all cases even if the information reported was accurate. It is possible that the respondent in 1991 still lived in the apartment in 1993 but that the respondent in 1993 was another person who moved into the unit later. If the unit was occupied in both years and the respondent in 1991 moved into the apartment before 1991 and the respondent in 1993 moved into the apartment in 1991, I assumed that it was vacated and reoccupied over this period, even though it is possible that it was continuously occupied by the 1991 respondent. The 1993 respondent might have moved into the apartment in 1991. Obviously, errors in the reported answers to these questions will create additional errors in determining which households moved. With more time, it would have been possible to use other information in the surveys to determine more accurately which households moved and to test the sensitivity of the results to reasonable alternative answers to this question. Based on the reported answers and the assumptions made to deal with ambiguous cases, I create a dummy variable that is 1 if it appears that the 1991 occupant of the apartment had moved within two years and 0 otherwise.

Regression analysis is used to estimate the probability of moving for households of various types for the same reason that it is used to estimate the mean market rent of apartments of various types, namely the sample mean for many combinations of characteristics is a highly unreliable estimator of the population mean due to small sample sizes. It is assumed that the proportion of the population of households living in rent regulated units that moved over this period is a linear function of the dummy variables in Table 2. These dummy variables represent six characteristics of the household and apartment that might be expected to affect the propensity to move.

The reasons to expect age of a head of the household, the magnitude of the rent discount, length of occupancy, and household size to affect this propensity are fairly obvious.[4] The inclusion of the other two characteristics requires some explanation. The estimated rent discount used in the estimation is based on a predicted market rent of each unit that is the same for all units that have the same observed characteristics listed in Table 1. It seemed plausible that rent regulated apartments that are the same with respect to these characteristics but have a higher actual rent would be better with respect to unobserved characteristics and hence have a rent discount greater than the estimated rent discount. If this were true, we would expect the propensity to move to be lower for units with higher rents among units with the same estimated rent discount. It also seemed plausible that stabilized and old-style controlled apartments with the same current rent discount would provide different expected future rent discounts due the different rules concerning rent changes as long as the person remains in the unit. Specifically, it is reasonable to expect the stabilized units to have smaller future rent discounts for the present tenants and hence for their occupants to be more likely to move. The results in Table 2 provide little support for these hypotheses. However, nothing is lost by including these variables in the prediction equation.

For the other variables, the results of the estimation of the linear probability model reported in Table 2 are usually in accordance with expectations. The probability of moving was less for households with older heads except for the oldest age group where deaths and moves to nursing homes become a major factor. In almost all cases, households that received the largest rent discount in their current apartment were the least likely to move. There was a tendency for the households that had lived in their apartments for the longest time in 1991 to have the lowest probability of moving, though the relationship is not as clearcut as expected. The probability of moving was smallest for the largest households except for households consisting of five or more persons.

To predict the fraction of households with a particular combination of characteristics living in rent regulated apartments in 1996 who would move within two years, the values of the dummy variables corresponding to these characteristics are substituted into the equation defined by the results in Table 2.[5] For example, the predicted probability of moving within two years for a household of one person in his or her twenties who has been living in a post-1947 rent stabilized unit for less than a year is .55. The weighted mean of the predicted probabilities for the households used in the analysis was .31 and the weighted median .29.[6] The predicted probabilities ranged from .02 to .71, which certainly makes clear the importance of accounting for these differences.

Since these estimates are based on the experience in the absence of vacancy decontrol, they assume that vacancy decontrol will have no effect on turnover rates. It is reasonable to believe that this policy will affect turnover rates in several offsetting ways. First, it might increase turnover by leading to increased landlord harassment of tenants. Vacancy decontrol increases the landlord's financial incentive to get sitting tenants to move. Second, it might decrease turnover by making one of the alternatives to staying put less attractive, namely the alternative of moving to an apartment whose rent is greater on account of vacancy decontrol. Since there is no systematic evidence on the effect of vacancy decontrol on turnover rates and time constraints precluded producing evidence, the analysis is based on the assumption that the effect is zero.

In estimating the effects of vacancy decontrol, it is neither necessary nor desirable to attempt to predict which specific households in the sample will move between 1996 and 1998. Since each household in the Housing and Vacancy Survey is selected at random from a subset of the population of the City, each represents a certain number of other similar households. To estimate how many of these households will move between 1996 and 1998, the number of households represented by a sample household is multiplied by the estimated fraction of these households who will move within two years.

4. Effects of Vacancy Decontrol on Rents of Regulated Units

This section reports the results of three methods for calculating the rent increases for regulated units that would have resulted over a period of two years from the implementation of vacancy decontrol in the spring of 1996. The simplest method assumed no change in the market or actual rents of these apartments under the prevailing regulations between the spring of 1996 and the time that they were reoccupied after being vacated.[7] The second method accounts for changes in stabilized rents of these units allowed under the rent regulations in effect at that time and changes in market rents due to the loss of the tenure discount and the general inflation in housing prices. The third method attempts to improve upon the second method by accounting for the fact that the actual rent of an apartment cannot exceed its market rent.

The calculations in this paper are based on all rent regulated units.[8] The rent regulations considered by the State Legislature applied only to rent stabilized units. The City has complete jurisdiction over old-style rent control. If an apartment currently under the older form of rent regulation is vacated, its landlord is allowed to charge whatever the market will bear on the new lease, but future leases are subject to the restrictions of rent stabilization. Vacancy decontrol applied only to stabilized apartments will have no effect on the rents paid by tenants who move into units previously subject to old-style rent control. These new tenants will pay market rents with or without the change in the rent stabilization law. Although future occupants may face higher rents under vacancy decontrol applied to rent stabilized units, this is irrelevant for an analysis of short-run effects. Since my calculations assume that no apartment has more than one change in tenant during the two years under consideration, the calculated increase in rent for any unit under old-style rent control that is vacated during this period should have been zero. However, the calculations in this paper ignore this distinction between rent stabilized and old-style rent controlled units and assume that the latter would experience rent increases equal to the excess of their current market rent over their actual rent. So my estimate of the effect of vacancy decontrol on all rent regulated apartments overstates the percentage increase in the aggregate rent of these units that would result from the vacancy decontrol of stabilized units. It also overstates the increase for all rent stabilized units because rent discounts are much larger for apartments under old style rent control than those under rent stabilization. However, since old style rent control accounts for less than seven percent of all rent regulated units, these overestimates are likely to be small.

Data from the 1991 and 1993 Surveys permitted reasonable estimates of how many rent regulated units occupied by households of each type would be vacated between the spring of 1996 and the spring of 1998. The exact dates at which units are vacated and occupied are not reported. In the interest of producing results within a reasonable time, I assumed that all of these units would be reoccupied in May 1997. When the second and third methods are used, this leads me to overstate the actual and market rents of units reoccupied earlier and understate the rents of units reoccupied later . However, it should not lead to any perceptible bias in the estimates of the average actual and market rents.

The simplest approach to estimating the effect of vacancy decontrol is to compare an estimate of the market rent at the time of the 1996 Survey with the actual rent at that time for each regulated unit that is vacated over the two years. Although this difference does not account for changes in the market and actual rent that would occur under the prevailing ordinance between the time of the survey and the time that the vacated unit is reoccupied, it provides a reasonable first approximation to the increase in rent due to vacancy decontrol. The market rent of each apartment is estimated based on the results in Table 1. The actual rent is reported in the survey.

The median and mean absolute increase for apartments that will be vacated between 1996 and 1998 in the entire city and each of its boroughs based on these measures of market and actual rent are reported in the first column of Tables 3 and 4.[9] The fourth column in Table 3 reports the median of the percentage increases. The fourth column in Table 4 reports the percentage by which the mean market rent exceeds the mean actual rent. Since the rent increase cannot be negative, a median of zero is reported when the median of the estimated differences is negative, in other words, when less than half are positive.[10] These results lead to the conclusion that vacancy decontrol would result in small increases in rent for the majority of rent regulated apartments vacated over the two years after its implementation, except in Manhattan. In part, this is because the apartments with the largest rent discounts are much less likely to be vacated.

More precise estimates of the effects of vacancy decontrol would account for several factors that would change the market and actual rents of units that are vacated. The market rent of a vacated unit will increase by the amount of the tenure discount at the time that it is vacated. This is equal to the current tenure discount plus the increase in the rent of identical newly occupied unregulated units due to the general inflation in housing prices. The estimated current tenure discount for each unit is calculated based on the coefficients of the dummy variables in Table 1 and the current tenant's length of occupancy. For example, if the current tenant moved into the apartment between 1986 and 1990, it is estimated that a new tenant of the unit at the time of the 1996 Survey would pay an extra $32.78 per month in rent.[11] It is assumed that the gross market rent of newly occupied unregulated apartments with any combination of characteristics increased by 2.5 percent in the year after the 1996 Survey. This is the increase in the Consumer Price Index for the New York City metropolitan area between April 1996 and April 1997.

The actual rent of a regulated unit can increase whenever a new lease is signed by a continuing tenant and even more when a new tenant is involved because the ceiling rent is increased on these occasions. Under the regulations in effect between October 1, 1996 and June 15, 1997, the ceiling rent of a rent stabilized apartment was increased by 14 percent when it was occupied by a new tenant who signs a one-year lease. Since this increase applies to contract rent and the ratio of mean contract to mean gross rent in the unregulated sector was .92, it is estimated that the actual rent of a regulated apartment that is vacated a year after the 1996 Survey would increase by 12.9 percent under the rent regulations in effect at that time.

The rent discounts based on the adjusted market and actual rents are reported in columns 2 and 5 of Tables 3 and 4. Obviously, these more accurate estimates only strengthen the conclusion of the preliminary analysis. Vacancy decontrol would result in small increases in rent for the majority of apartments vacated over the two years after its implementation, except in Manhattan.

The calculations up to this point ignore an important aspect of reality. The actual rent of a regulated apartment cannot exceed its market rent because tenants will not pay more for a regulated unit than for an identical unregulated unit. However, the actual rent often exceeds our estimate of market rent, sometimes by a large amount. The estimated market rent of an individual regulated unit with a particular combination of observed characteristics can be much greater or less than its true market rent for the same reason that the actual rent of an unregulated unit with these observed characteristics deviate from the mean rent of such units. Some units are much better than average with respect to unobserved characteristics and others much worse.

Since the actual rent of each apartment is known with considerable certainty while our estimate of market rent clearly involves substantial prediction errors in many cases, a simple approach to using this information about the relationship between actual and market rent is to set the predicted market rent equal to the actual rent in cases where the actual exceeds the predicted market rent based on the results in Table 1. However, increasing the predicted market rents of some apartments and leaving the other predicted market rents unchanged would increase the mean of the predicted market rents, and we have no reason to believe that our procedures lead to an overestimate of the mean market rent of regulated units. If anything, there is reason to believe that we have underestimated this mean because regulated units are likely to be less well maintained in ways not captured in the data.

To insure that the predicted market rent of each regulated unit is at least as great as its actual rent without affecting the estimated mean market rent, we decrease the estimated market rent of each unit by the same percentage and then increase the estimated market rent of each unit up to its actual rent in cases where actual rent exceeds estimated market rent. The percentage is chosen to insure that the mean of the new estimated market rents is the same as the mean of the original estimated market rents.

Figure 1 illustrates the procedure. This figure refers to apartments that are the same with respect to observed characteristics. They differ with respect to unobserved characteristics which is why we are unable to accurately estimate the market rents of all apartments. The actual rent of each apartment is measured along the horizontal axis; the predicted market rent along the vertical axis. The six apartments with this combination of observed characteristics have different actual rents A1 through A6. Since these apartments have the same observed characteristics, our initial prediction of the market rent of each is the same, in this case M4. After making the preceding adjustments, the estimated market rent of apartments 1, 2, and 3 is reduced to M3, apartment 5 is increased to M5, and apartment 6 to M6.

The results of making these further adjustments in predicted market rents are reported in columns 3 and 6 of Tables 3 and 4. With respect to medians, the adjustments affect only the results for Manhattan. They lead us to conclude that even in Manhattan the majority of apartments that would be vacated within two years of vacancy decontrol would experience no rent increase as a result of adopting this policy. The adjustments have a different effect on the estimates of mean rent increases. They increase the estimated means to levels somewhat greater than the crudest estimates. However, the qualitative conclusion is unaffected. Vacancy decontrol would result in small increases in rent for the majority of rent regulated apartments vacated over the two years after its implementation, except possibly in Manhattan.

Although it appears that the majority of the apartments would experience little or no rent increase as a result of vacancy decontrol, some apartments would experience substantial rent increases. Based on the third method, vacancy decontrol would lead to rent increases in excess of $233 per month for 10 percent of all vacated units and in excess of $369 per month for 5 percent. A few apartments in the sample would experience increases somewhat greater than $1000 per month, and there are almost surely others not in the sample for which the rent increases would be even greater.

Tables 5 and 6 report the results for thirty subareas of the City based on the crudest and the most refined methods.[12] The crudest method suggests that that the majority of the apartments vacated over the two years in sixteen of the thirty areas will experience no rent increase as a result of vacancy decontrol and that the median rent increase will exceed 10 percent in only seven of the thirty areas, the majority in Manhattan. The most refined method suggests that the majority of the apartments in each area vacated over the two years will experience no rent increase as a result of vacancy decontrol. In terms of means, the crudest method leads to the conclusion that only six of the areas will experience an increase in the mean rent of these apartments in excess of 10 percent; the more refined method to the conclusion that only four areas would experience increases of this magnitude.

Some indirect evidence supports the view that many rent regulated apartments are currently renting at or close to market levels. The Housing and Vacancy Survey not only indicates whether an apartment was rent regulated but also the respondent's answer to a question concerning the rent regulation status of his or her apartment. It is reasonable to believe that tenants in rent regulated units renting for substantially less than market rent would be aware that their units are regulated. More than 50 percent of the households living in rent stabilized apartments and 27 percent living in old-style rent controlled units did not realize that their apartments were subject to rent regulation. The difference between the percentages for occupants of apartments under rent stabilization and old-style rent control supports this argument. The mean rent discount in 1996 for apartments under old-style rent control was about $100 a month greater than for apartments under rent stabilization. So we would expect more occupants of apartments under old-style rent control to be aware of the rent regulation status of their apartments.

5. Average Losses from Vacancy Decontrol by Household Characteristics

Since the losses from vacancy decontrol are incurred by households that move after its implementation and these losses are less than the rent increases reported in the previous section for reasons to be explained, it is considerably more difficult to determine the magnitudes of the losses to households of various types than the magnitudes of the rent increases for apartments in different locations. This section explains why the rent increase overstates the loss and presents estimates of the maximum average losses from vacancy decontrol to different types of households.

The excess of the market rent over the actual rent of a newly occupied apartment subject to rent regulation overstates, probably by a large margin, the loss from vacancy decontrol for four reasons. First, it is reasonable to believe that new tenants typically spend more time and money to obtain rent regulated apartments, especially units with a large rent discount, than they would have spent to obtain an unregulated apartment. The amortized value of this extra time and money should be subtracted from the rent increase in calculating the loss of vacancy decontrol. Second, it is reasonable to believe that tenants of rent regulated apartments spend more on the maintenance of their apartments than occupants of unregulated units in order to offset the reduced landlord maintenance that results from rent regulation. This extra expense should be subtracted from the rent discount in calculating the loss from vacancy decontrol.[13] Third, the rent discount induces many households to accept or stay in an apartment whose characteristics differ from the characteristics of the unit that the household would choose if it were given a cash grant equal to its rent discount and required to live in unregulated housing. As a result, the benefit of rent regulation, and hence the loss from vacancy decontrol, to such households is less than their rent discount.[14] Finally, the available evidence, including recent evidence for New York City reported in Section 7, indicates that rent regulation results in higher rents in the uncontrolled sector for apartments with each set of characteristics. Vacancy decontrol will gradually reduce rent levels in the unregulated sector, thereby reducing any loss suffered by the new occupants of previously regulated apartments and benefiting all occupants of previously unregulated units.

Tables 7 through 13 report the maximum mean monthly losses to different types of households that would have resulted from vacancy decontrol implemented at the time of the 1996 Housing and Vacancy Survey due to the higher rents of apartments vacated over the subsequent two years. These are the means of the rent discounts after making all of the adjustments described in the preceding section. They are greater than the true losses for the reasons mentioned in the preceding paragraph. Except for Table 13, these tables report estimates of the maximum mean loss to all households of a particular type who might lose from vacancy decontrol, the percentage of regulated apartments that will be vacated in the two years following the implementation of vacancy decontrol, and the maximum mean loss to all households of that type living in regulated apartments. Since tenants who remain in their rent regulated apartments do not lose from vacancy decontrol, the mean loss to all occupants of rent regulated apartments is less than the mean loss to occupants who moved into their apartments after the implementation of vacancy decontrol. Table 13 reports only the mean losses to all occupants of rent regulated apartments.

These calculations are based in part on the assumption that vacancy decontrol has no effect on which household occupies each vacated apartment. Under this assumption, the new occupant of a previously regulated apartment would pay a higher rent for the same unit. Obviously, vacancy decontrol could effect who lives where. Indeed, Section 6 contains crude estimates of the long-run effects of this policy on the types of households living in the City and each of its boroughs. If vacancy decontrol does affect who lives in previously rent regulated apartments that are vacated under this policy, then the rent discount does not necessarily tell us anything about the loss to the new occupant of a previously regulated unit. For example, suppose that the new occupant of a previously rent regulated apartment under vacancy decontrol would have lived in an unregulated apartment if the old rent regulations had been continued. This tenant would not lose a rent discount on account of vacancy decontrol. Time constraints precluded attempting to account for the effects of vacancy decontrol on who lives where in making the calculations in this section.

The analysis in this section is also based on the assumption that each unit that would have been vacated between the spring of 1996 and the spring of 1998 is occupied by a household with the same characteristics as the previous tenant. [15] Although luck surely plays an important role in determining whether a particular household gets a regulated apartment with a large discount, systematic factors are also at work. For example, more desirable tenants such as those who are likely to move more frequently or who are expected to cause less damage to the apartment will have a better chance of getting a particularly good bargain.[16] While it is not reasonable to expect each currently regulated apartment that is vacated to be occupied by a household with the same characteristics as its previous occupant, it is reasonable to believe that the same sorts of households would tend to occupy the same sorts of apartments if existing rent regulations were continued. So to the extent that vacancy decontrol has little effect on who lives where, we might reasonably expect the average rent discount for new occupants with a particular combination of characteristics to be about the same as it would have been had each household moving out of a currently regulated apartment been replaced by a household with the same characteristics.

Table 7 reports the maximum mean loss for four income groups. Each contains a fourth of the households in New York City. [17] Among households that lose from vacancy decontrol, the poorest seem to incur the largest losses. However, the differences between their mean loss and the losses to other income groups among all households living in regulated units is much smaller because they are less likely to move and hence incur any loss.

Table 8 indicates that the maximum mean loss to one person households is about twice as large as the maximum mean loss to households of other sizes among those who incur losses and among all occupants of rent regulated apartments.

Table 9 suggests that, among households that lose from vacancy decontrol, those headed by a person over 61 years of age incur a mean loss that is considerably greater than the mean loss to households with a younger head. The estimated difference appears to be much smaller among all households living in rent regulated units because the elderly are much less likely to move. It is important to recognize the possibility that the assumptions underlying our calculations may be especially far from the mark in this case, resulting in highly misleading conclusions. It is entirely possible that the overwhelming majority of elderly households who vacate rent regulated units die or move into a nursing home, retirement village, or house in a warmer place. It may be rare for such households to move into rent regulated apartments. The fraction of rent regulated apartments occupied by the elderly could stay approximately the same over time due to the aging of sitting tenants.[18] As a result, the elderly may incur much smaller losses from vacancy decontrol than younger households contrary to the reported results.

Table 10 indicates that, among households that lose from vacancy decontrol, whites incur somewhat larger losses than blacks and that blacks incur larger losses than other nonwhites. These differences are smaller among all occupants of rent regulated apartments due to differences in propensity to move.

Tables 11 through 13 allow us to see how the maximum mean loss varies with one household characteristic among households that are the same with respect to one or two other characteristics. Some of the results in these tables are similar to those in Tables 7 through 10. For example, one person households incur larger losses than households of other sizes among households with the same income, the same income and age of the head of the household, and the same income and race. This is the case whether we consider only households that incur a loss or all households in rent regulated housing. However, some of the results of the simple correlations reported in Tables 7 through 10 appear to be due to the influence of other household characteristics that are correlated with characteristic under consideration. For example, it is far from the truth to say that the poorest households incur the largest losses from vacancy decontrol among households that are the same with respect to other characteristics. For example, Table 11 indicates that the mean loss among households of two and four persons that incur a loss is greatest for the richest households. The poorest households incur the largest loss in less than half of the cases considered in Tables 12 and 13. The overall impression from these tables is that the mean losses to almost all types of households are small, especially when it is recalled that these numbers are likely to be substantially larger than the true losses.

It is important to remember that the preceding calculations ignore the benefits of vacancy decontrol such as the higher level of public services that will result from the higher property tax revenue and the lower rents of previously unregulated apartments. These benefits will reduce the losses to households who occupy previously rent regulated apartments after the implementation of vacancy decontrol and provide gains to others. Therefore, if we expanded our consideration to all households of a particular type, we might easily find that some or most types of household experience a net benefit from vacancy decontrol, even if we ignore the gains to the owners of regulated units.

6. Effect of Vacancy Decontrol on Who Lives Where in New York City

In the debate over the continuation of rent regulation, a frequently expressed concern about vacancy decontrol was that it would lead to a massive rearrangement of the population of the New York metropolitan area, with the richest households displacing most low and middle income households in Manhattan and these households moving into adjoining boroughs and driving up rents there. Evidence from the 1996 Housing and Vacancy Survey makes clear that this will not happen.

If these arguments were correct, we would expect almost all apartments in Manhattan that were not subject to rent regulation to be occupied by rich households. In fact, about half of the households in these apartments have annual incomes less than $55,200. Twenty two percent have incomes less than $30,000. Even if we limit consideration to the more expensive areas of Manhattan (Greenwich Village, the Financial District, Chelsea, Clinton, Midtown, Styvesant Town, Turtle Bay, Upper West Side, Upper East Side), the numbers are not much different. Forty six percent have household incomes less than $55,200; nineteen percent less than $30,000.

This reflects the fact that individual tastes differ enormously. Some people are willing to make great sacrifices to live in Manhattan. Others with the same income and household composition are not willing to make these sacrifices. Obviously, the wealthiest renters in the metropolitan area who live outside of Manhattan could have outbid these low and middle income households for their apartments. They did not do it because they prefer their current apartments, given the market rents of units of various types in Manhattan and elsewhere.

If we want to estimate the effect of vacancy decontrol on who lives where, we might reasonably assume that, under this policy, the apartments in each borough that are currently rent regulated will ultimately be occupied by households with the same characteristics as the current occupants of unregulated units in that borough. For example, if 15 percent of the current occupants of apartments in Manhattan that are not subject to rent regulation are occupied by households with annual incomes between $13,440 and $30,000, we might reasonably assume that 15 percent of the units that are currently rent regulated will ultimately be occupied by households in this income category as a result of vacancy decontrol. If anything, this will overstate the extent to which richer households will move into, and poorer households out of, Manhattan because the typical rent regulated apartment is in worse condition than the typical unregulated unit and hence fewer of the rent regulated than the unregulated apartments will be attractive to the richest renters.[19]

Tables 14 through 20 contain calculations based on the preceding assumption. They indicate that vacancy decontrol would have a modest effect on who lives where within the metropolitan area. At most it would eventually reduce the number of households in New York City with incomes below $30,000 from 50% to 45% (Table 14). Since the median household income in the suburbs is considerably greater than in New York City, this would reduce the income gap between the suburbs and the city, albeit by a small amount.

As usually argued, Manhattan would experience the largest influx of affluent households. The number of households in Manhattan with incomes in excess of $55,200 would increase from 35% to at most 46% (Table 17). At the other extreme, the effects on the distribution of income in Brooklyn, Queens, and Staten Island would be miniscule (Tables 16, 18, 19). The largest change would be a decrease from 30% to 27% in the number of households in Brooklyn with annual incomes below $13,440. No other income class in these boroughs would experience a change of more than one percentage point.

The fear that vacancy decontrol will lead to increases in the rents of unregulated units in the boroughs adjacent to Manhattan ignores the fact that it will induce some households to leave these boroughs while others enter them. Rent levels depend on the net effect of these moves. Unless there is some reason to expect that vacancy decontrol will lead to fewer households wanting to live in Manhattan or the suburbs and hence more households wanting to live in the other boroughs, there is no reason to expect it to lead to higher market rents in these boroughs.

The preceding method can be used to estimate the distribution of other household characteristics in New York City resulting from vacancy decontrol. The results are reported in Table 20. These calculations indicate that vacancy decontrol will gradually lead to extremely modest decreases in the numbers of small and elderly households in the City and corresponding increases in the numbers of large and young households. It will have virtually no effect on the racial composition of the City.

My conclusion from these calculations is that New York City would remain wonderfully diverse with or without vacancy decontrol. Vacancy decontrol would gradually lead to some changes in who lives where, but these changes would not be massive. In part, this is because it would have little effect on the locational decisions of households living in owner occupied or publicly subsidized housing. In part, it is because the characteristics of households currently living in rent regulated units are surprisingly similar to the households in apartments renting at market rates.

7. Effect of Vacancy Decontrol on Market Rents

Vacancy decontrol has a direct effect on the rents of regulated apartments that are vacated. It also has an indirect effect on the rent of unregulated units with each combination of characteristics. This section discusses the short-run effect of vacancy decontrol on the rents of units with different combinations of characteristics and presents evidence on its long-run effect on the level of rents.[20]

In the short-run, vacancy decontrol cannot possibly lead to rent increases or rent decreases for currently unregulated units of all types. Decontrol adds the same number of apartments and households to the unregulated sector. It will undoubtedly increase the demand for units of some types more than it will increase the supply of these units in the unregulated sector, thereby driving up their rents. If so, it must increase the supply of other types of units in the unregulated sector more than it increases their demand, leading to lower rents for these units. Only in the totally implausible case where the increase in the demand and supply of units of each type in the unregulated sector are the same will this pattern of rent increases for some types of units and rent decreases for other types be avoided. Unfortunately, there are no credible estimates of these short-run effects of vacancy decontrol on the market rents of various types of housing.

The lower vacancy rates and higher rents for units of some types will lead landlords to alter other types of apartments that are less profitable to make them similar to the units in high demand and to rehabilitate structures that are not currently in residential use to provide apartments of these types. Later, new construction will provide more units to compete with the better existing units in excess demand. These supply responses will drive down the rents of these units from their temporarily high levels. The conversion of existing apartments from one type to another will reduce the supply of units of each type that had experienced a short-run decline in rent, thereby increasing the rents of units of those types.

Over the longer haul, the effects of vacancy decontrol on the rents of currently unregulated apartments would depend on whether the current occupants of regulated units would want to live in better or worse housing if they must pay market rents and the magnitude of the higher rents that owners of unregulated rental housing currently receive to compensate them for the higher risk of operating in a market with a history of rent regulation that has from time to time reduced the value of their property.

If current occupants of rent regulated apartments would want to live in worse housing at market rents, this would decrease the overall effective demand for housing in the long run and hence reduce rents for units of all types as well as the overall quality of the housing stock. If they would want to live in better housing at market rents, this would increase the overall demand for housing in the long run and hence increase rents for units of all types. Obviously, there are some current occupants of rent regulated apartments in each category. The overall demand would increase if the total amount that these households would want to spend on housing at current market rent levels exceeds the sum of the market rents of their rent regulated units. Two equally plausible methods for estimating the effect of vacancy decontrol on demand led to conflicting results. With additional effort it probably would have been possible to obtain an estimate worth reporting. However, the magnitude and nature of this change in demand are unlikely to produce substantial changes in the prices of the inputs involved in the production, rehabilitation, and maintenance of housing and hence unlikely to produce a substantial change in rent levels in the long run.

The direction of the effect of vacancy decontrol on the willingness to supply housing is unambiguous. As fears of the reimposition of rent regulation subside, investors would be willing to provide rental units of each type at a lower rent, or to put it another way, they would be willing to provide more apartments of any type at a given rent. The only issue is the magnitude of the effect. The evidence presented below suggests that the market rents of apartments of all types would eventually fall 8 percent on this account. This would probably be manifested in a lower rate of inflation in rent levels rather than an absolute decrease in rents. The rate at which it would occur depends on the rate at which investor fears of the reimposition of rent regulation subside. In light of the history of rent regulation in New York City, it is plausible to believe that these fears would subside very little in the early years after vacancy decontrol is implemented.

The estimate of the long-run effect of vacancy decontrol on the market rent of an apartment with specified characteristics is based on a previously estimated equation explaining the mean rent of uncontrolled units with specified characteristics in terms of determinants of the demand and supply of housing in the uncontrolled sector.[21] Specifically, the quality-adjusted index of rent in the uncontrolled sector was assumed to be a linear function of mean real household income, the percentage of households headed by a black, and the percentage of households headed by a female in the uncontrolled sector, the price of land and an index of the prices of other inputs used to produce housing, and the percentage of rental units in the metropolitan area that are subject to rent regulation. The equation was estimated using data on the 44 metropolitan areas in the 1985 through 1988 American Housing Surveys, supplemented with data from other sources.[22] The New York metropolitan area was surveyed in 1987. The predicted market rent of units with the specified characteristics in the presence of the existing rent regulations in New York City was obtained by substituting the values of the explanatory variables for New York into the estimated equation. The predicted market rent in the absence of rent regulations was obtained by the same procedure except that the percentage of units subject to rent regulation was set equal to zero, which is the ultimate effect of vacancy decontrol. The predicted rent with rent control exceeded the predicted rent without it by about 8 percent.

8. Conclusions

Had vacancy decontrol been implemented on June 15, 1997, it would have had resulted in extremely small increases in rent for the majority of rent regulated apartments vacated over the following two years, except possibly in Manhattan. In part, this is because the apartments with the largest rent discounts are much less likely to be vacated. However, some of the vacated apartments would experience substantial increases. Our best estimates suggest that vacancy decontrol would lead to rent increases in excess of $230 per month for about 10 percent of all vacated units and in excess of $370 per month for about 5 percent. Similar results applied to the overwhelming majority of the thirty subareas into which the City was divided for purposes of this study. Few would have experienced increases in the mean rent of vacated units in excess of 10 percent.

It is difficult to determine the loss from vacancy decontrol to any household because (1) the loss from vacancy decontrol of an apartment is less than the excess of its market rent over its regulated rent by an amount that is probably large but difficult to determine and (2) who loses depends on which households will move into which units after its implementation. We estimate the average loss to households of different types using the rent increase of the previously rent regulated apartment occupied by each household as the measure of loss and assuming that vacancy decontrol has no effect on which household occupies each vacated apartment and that each unit that would have been vacated between the spring of 1996 and the spring of 1998 is occupied by a household with the same characteristics as the previous tenant.

Among households that lose from vacancy decontrol, the poorest seem to incur the largest losses. However, the differences between their mean loss and the losses to other income groups among all households living in regulated units is much smaller because they are less likely to move and hence incur any loss. Furthermore, it is far from the truth to say that the poorest households incur the largest losses from vacancy decontrol among households that are the same with respect to other characteristics. For example, the mean loss among households of two and four persons that incur a loss is greatest for the richest households. The poorest households incur the largest loss for less than half of the household types considered.

The mean loss to one person households is about twice as large as the mean loss to households of other sizes among those who incur losses and among all occupants of rent regulated apartments. This is true even among households that are the same in other respects. One person households incur larger losses than households of other sizes among households with the same income, the same income and age of the head of the household, and the same income and race. This is the case whether we consider only households that incur a loss or all households in rent regulated housing.

Although our results suggest that households with an elderly head incur larger losses on average than other households, it is reasonable to believe that the assumptions underlying the calculations are so far from the mark in this case that the truth is the opposite of the reported results.

Among households that lose from vacancy decontrol, whites incur somewhat larger losses than blacks and blacks incur larger losses than other nonwhites. These differences are smaller among all occupants of rent regulated apartments due to differences in propensity to move.

The overall impression from these calculations is that the mean loss over the two years after the implementation of vacancy decontrol would have been small to households of almost all types. Furthermore, it is important to remember that the preceding calculations ignore the benefits of vacancy decontrol such as the higher level of public services that will result from the higher property tax revenue and the lower rents of previously unregulated apartments. These benefits will reduce the losses to households who occupy previously rent regulated apartments after the implementation of vacancy decontrol and provide gains to others. Therefore, if we expanded our consideration to all households of a particular type, we might easily find that some or most types of household experience a net benefit from vacancy decontrol, even if we ignore the gains to the owners of regulated units.

In the debate over the continuation of rent regulation, a frequently expressed concern about vacancy decontrol was that it would lead to a massive rearrangement of the population of the New York metropolitan area, with the richest households displacing most low and middle income households in Manhattan and these households moving into adjoining boroughs and driving up rents there. Under the assumption that apartments in each borough that are currently rent regulated will ultimately be occupied by households with the same distribution of characteristics as the current occupants of unregulated units in that borough that are rented at market rents, evidence from the 1996 Housing and Vacancy Survey makes clear that this will not happen.Vacancy decontrol would gradually lead to some changes in who lives where, but these changes would not be massive. In part, this is because it would have little effect on the locational decisions of households living in owner occupied or publicly subsidized housing. In part, it is because the characteristics of households currently living in rent regulated units are surprisingly similar to those of the households in apartments renting at market rates.

Vacancy decontrol has not only a direct effect on the rents of regulated apartments that are vacated but also an indirect effect on the rent of unregulated units with each combination of characteristics. In the short-run, vacancy decontrol cannot possibly lead to rent increases or rent decreases for currently unregulated units of all types. Over the longer haul, the effects of vacancy decontrol on the rents of currently unregulated apartments would depend on whether the current occupants of regulated units would want to live in better or worse housing if they must pay market rents and the magnitude of the higher rents that owners of unregulated rental housing currently receive to compensate them for the higher risk of operating in a market with a history of rent regulation that has from time to time reduced the value of their property.

The changes in the types of housing that the current occupants of regulated units would want to occupy if they had to pay market rents are unlikely to produce substantial changes in the prices of the inputs involved in the production, rehabilitation, and maintenance of housing and hence unlikely to produce a substantial change in rent levels in the long run. The direction of the effect of vacancy decontrol on the willingness to supply housing is unambiguous. As fears of the reimposition of rent regulation subside, investors would be willing to provide rental units of each type at a lower rent. It is estimated that the market rents of apartments of all types would eventually fall 8 percent on this account. This would probably be manifested in a lower rate of inflation in rent levels rather than an absolute decrease in rents. The rate at which it would occur depends on the rate at which investor fears of the reimposition of rent regulation subside. In light of the history of rent regulation in New York City, it is plausible to believe that these fears would subside very little in the early years after vacancy decontrol is implemented.