The Asset Price Incidence of Capital Gains Taxes: Evidence from the UPREIT Structure and the Taxpayer Relief Act of 1997

Transcription

1 The Asset Price Incidence of Capital Gains Taxes: Evidence from the UPREIT Structure and the Taxpayer Relief Act of 1997 Todd Sinai and Joseph Gyourko Real Estate Department The Wharton School University of Pennsylvania This draft: October 15, This research was supported by the Research Sponsor Program of the Zell/Lurie Real Estate Center at The Wharton School of the University of Pennsylvania. Donghoon Lee provided outstanding research assistance. We are grateful to Bill Gentry, Barbara Murray, Doug Shackelford, and Joel Waldfogel for helpful comments and discussions.

2 Copyright: Todd Sinai and Joseph Gyourko, October 1998.

3 Abstract In this paper, we examine the asset price incidence of the capital gains tax cut in the Taxpayer Relief Act of By comparing two organizational structures in the real estate industry that differ only in how they should be affected by a change in capital gains tax rates, we isolate the effect of the tax cut from industry trends and firm-level fixed effects. Our estimates indicate that, in this industry, the benefit of a capital gains tax deferral when selling appreciated property accrued almost totally to the buyer of the asset, not the seller. Using share price data, we find that real estate firms that experienced a reduction in their tax subsidy in TRA97, called UPREITs, had 8.6 percent less price appreciation in 1997 relative to 1996 than did the companies that had no tax change, known as REITs. Firms that appeared most likely to be purchasers of property and which thus received the most benefit from the tax subsidy before it was cut endured the largest relative decline in share prices. We also find suggestive evidence that prices for new acquisitions rose for UPREITs relative to the REITs after TRA97. We conclude that up to 25 percent of UPREITs share values is due to the remaining tax subsidy that they enjoy.

4 Asset price models of tax incidence predict that a change in the tax treatment of the return on an asset should be captured to some degree in the price of that asset. While the theoretical underpinning of these models is well-developed [e.g., Summers (1981), Poterba (1984)], empirical research on asset price incidence is relatively rare and the results are mixed. An early study by Cutler (1988) concludes that the market inefficiently prices tax information. Goolsbee s (1998) study of changes in the investment tax credit concludes that much of the investment subsidy accrues to capital suppliers, at least in the short run. Lang and Shackelford (1998) examine whether the decrease in the capital gains tax rate in TRA97 affected share prices since investors would receive a higher after-tax return. They find that over one week in April and May when a budget compromise was finally reached, share prices for companies that pay low dividends rose more than share prices for companies that paid high dividends, the latter benefitting less from the cut in the capital gains tax rate. Our paper provides evidence that in at least one industry suppliers of capital bore the incidence of the capital gains tax cut in the Taxpayer Relief Act of Good measures of incidence are necessary to evaluate the impact of potential tax reforms and the distribution of costs of the current tax system. By comparing two organizational structures in the real estate industry that differ only in how they should be affected by a change in capital gains tax rates, we are able to estimate a measure of incidence that is better identified than in previous work. The empirical strategy employed results in an uncontaminated estimate of the asset price incidence of the capital gains tax cuts in TRA97 at the expense of some loss of generality due to the focus on the real estate industry. Specifically, we examine a corporate structure unique to some publicly traded companies 1

5 in the real estate industry called an Umbrella Partnership Real Estate Investment Trust (UPREIT). Building owners who sell properties to companies with this structure can defer paying capital gains taxes if they take operating partnership units in exchange. Partnership units are convertible one-for-one to common shares, with the seller of appreciated property able to defer paying capital gains tax without penalty until she chooses to sell her stake in the UPREIT. Since an UPREIT has no significant structural differences from a traditional real estate investment trust (REIT) other than its ability to purchase property with its operating unit currency, the effect of the change in the tax code is isolated from general real estate industry conditions by comparing UPREITs to REITs. To the extent that REITs bear the incidence of the capital gains tax, they will have to pay higher prices for properties relative to UPREITs. The incidence question is investigated with two sources of data. We first analyze property-transactions data to determine whether the prices paid by UPREITs for properties rose relative to REITs after the reduction in the capital gains rate in TRA97. This provides a direct test of whether the incidence is on UPREITs and their existing shareholders. The findings suggest that UPREITs did pay higher prices, but the point estimates are too large to be taken at face value and have wide confidence intervals. We suspect that deficiencies in the property-level data the absence of good building quality controls especially make this a relatively weak test of the incidence question. When we examine the incidence question using much less noisy firm share price data we obtain stronger results. Comparing UPREIT to REIT price changes over the period of the tax legislation, we again find evidence that UPREITs and their current shareholders bore most of the burden of the change in the capital gains rate. Potential self-selection of companies choosing to 2

6 incorporate as UPREITs is controlled for by comparing the excess return to the UPREIT form in 1997 (relative to REITs) to the excess return over the same time period in In our preferred specification, UPREIT share price growth is found to be 8.6 percent lower than REIT share price growth around the passage of TRA97, relative to the same time period in The corresponding estimated tax elasticity of!0.93 (0.39) indicates that the bulk of the incidence of the implicit subsidy to the UPREIT structure from sellers being able to defer gains taxes is borne by the UPREIT. Further investigation reveals that share prices for relatively highly acquisitive UPREITs respond more to the tax change than do highly acquisitive REITs, further reinforcing our basic incidence conclusion. Finally, this result provides insight into the contribution of the capital gains subsidy to UPREIT market value. Assuming that the statutory capital gains tax rate on appreciated real estate is 21 percent, we calculate the remaining benefit from being able to defer gains taxes accounts for as much as 25 percent of UPREIT value, and possibly much more for higher growth firms. Hence, a significant fraction of UPREIT value presently derives from its ability to capture the benefits of allowing building sellers to defer taxes on sales to the UPREIT. The rest of this paper proceeds as follows: Section 1 describes the REIT and UPREIT structures for readers who are not familiar with the real estate industry. In section 2, we provide an analytical framework that explains why UPREITs may receive a tax subsidy when they purchase appreciated property. The circumstances surrounding the Taxpayer Relief Act of 1997, which we use to identify our estimates, are outlined in section 3. In section 4, we examine data on property purchases to see if there was a price response to changes in capital gains tax rates while section 5 uses firm-level data to see if stock prices moved. Section 6 concludes. 3

7 I. Background on the REIT industry and the UPREIT structure REITs are publicly-traded real estate companies that elect a special tax status. What distinguishes a REIT from a typical public corporation is that REITs do not pay income or capital gains taxes at the corporate level. In exchange, they must pay out at least 95 percent of their taxable incomes in the form of dividends to avoid entity-level taxation. 1 Equity REITs, which are firms that own and operate properties, are the focus of this research. These firms make up over 90 percent of the market capitalization of the REIT industry. 2 Even though the legislation enabling REITs to be created was passed in 1960, only in the last decade has the industry seen much growth. Figure 1 shows that the total market capitalization of all REITs was just over $15 billion at year-end 1992, while five years later the figure had risen to $140 billion. Nearly all of that growth was due to equity REITs, which ended 1997 with $128 billion in market capitalization. Figure 2 documents that the total number of REITs has also grown, but not as quickly as market capitalization. Between 1992 and 1997 the number of equity REITs doubled from 89 to 176. The UPREIT structure itself is a recent 1992 creation. Effectively, it consists of a publicly-traded REIT melded with an operating partnership. Figure 3 illustrates how a simple UPREIT structure might be set up. The operating partnership owns all the properties directly while the publicly traded portion owns at least some of the operating partnership. The UPREIT s equity consists of stock, which are shares in the REIT portion, and operating partnership (OP) 1 There are several other criteria a company must satisfy before it is allowed to be treated as a REIT, including a requirement that 50 percent or more of the REIT s stock cannot be owned by five or fewer people, but they are not germane to this research. 2 The other type of REIT is a mortgage REIT. A mortgage REIT is a publicly traded company that makes mortgages rather than purchasing and operating properties. 4

8 units, which are shares in the partnership. UPREIT shareholders do not own the underlying real estate directly, but indirectly through the REIT s ownership of the operating partnership. Other equity investors may only own OP units directly. Typically, one OP unit is convertible into one share of common stock and receives the same dividend payments. In order to cash out, a unit holder simply converts her OP units to common stock and sells the shares. The creation of the first UPREIT by the Taubman Realty Group in December 1992 ushered in a wave of such firms. Figure 4 shows how many public real estate corporations were formed each year and whether they were traditional REITs or UPREITs. By December 1992, 40 traditional equity REITs existed. In 1993 and 1994, 66 new UPREITs held their initial public offerings while only 14 new traditional REITs did so. The numbers of new UPREITs declined substantially in 1995 and now are being created at roughly the same rate as traditional REITs. 3 The capital gains tax benefits of the UPREIT structure are particularly noteworthy for the purposes of this paper. Transferring buildings to a traditional REIT is a taxable event, with the seller receiving stock and/or cash from the firm. Transferring buildings to an UPREIT is not a taxable event as long as the seller receives operating partnership units, not cash or stock. In this case, the IRS treats the deal as a tax-free exchange and the building seller defers her capital gains tax liability until one of the follow two events occurs: (a) she converts her OP units into stock and sells the shares; 4 or (b) the UPREIT sells the contributed properties. 3 Ten companies switched from being traditional REITs to UPREITs after 1992 and are excluded from the chart. Many of those companies created what has been labeled a downreit, creating an operating partnership to purchase specific properties while the REIT continues to own the original properties. Unlike the UPREIT form, the downreit structure does not have the explicit blessing of the IRS. 4 If the seller dies before converting the units, the tax basis is stepped-up. 5

9 II Analyzing the value of the UPREIT structure and the role of capital gains taxes Consider a building owner who is deciding to which type of firm to sell a property. Sale to a traditional REIT results in the seller receiving cash or liquid stock for today s market value of the building, V 0. A capital gains tax at rate J C must be paid on any taxable gain, today s value minus the basis, B. 5 The total after-tax proceeds from sale to a REIT, in today s dollars, would be: S REIT 'V 0 &J C (V 0 &B) (1) If the owner were to sell to an UPREIT, capital gains taxes could be deferred until the units were converted into shares and sold, but the seller would be locked in to holding the UPREIT s OP units. The value of this option in today s dollars is the discounted value of today s sale price minus the discounted value of the capital gains tax liability plus the discounted after-tax return from investing in the UPREIT by holding OP units. Defining H as the seller s expected holding period of the OP units, r REIT as the annual pre-tax return to an UPREIT OP unit, * as the proportion of that return paid out as dividends, J DIV as the tax rate on dividend income, and r as the equivalent-risk after-tax opportunity return, the total after-tax proceeds from a sale to an UPREIT would be: S UPREIT ' V H &J C (V H &B) (1%r) H H % j t'1 V 0 (1%(1&*)r REIT ) REIT (1&J DIV ) (1%r) t (2) The first term (in brackets) reflects that both the cash flow and the capital gains tax payment are 5 The Taxpayer Relief Act of 1997 instituted a two-part capital gains tax on real properties. Gains due to appreciation, V 0 minus the purchase price, PP, are taxed at one rate while gains due to depreciation, PP-B, are taxed at a higher rate. We abstract from this detail in this section. 6

10 deferred for H years. Since the original property owner is selling the units farther in the future, a higher price, V H 'V 0 (1%(1&*)r REIT ) H, that includes the expected capital gain in the OP units is received. The capital gains tax (at unchanged rate J C ) must be paid on the entire capital gain, V H!B, rather than V 0!B, but it does not need to be paid for H years. The second term starting with the summation sign values the discounted cash flow from the stream of dividend payments. The REIT generates a return of r REIT every year on the current equity value of the company, * percent of which is paid out as dividends and taxed at the rate J DIV. The remainder of the return, (1!*)r REIT, is retained by the company and creates capital appreciation. The stream of dividend payments increases over time, with the dividend yield remaining constant by assumption and the equity value of the company increasing due to capital gains. All lump sums and flows are discounted at the after-tax opportunity cost rate, r. The benefit created by deferring the capital gains tax can be seen more clearly by decomposing equation (2) as follows. First, denote the summation term for the present discounted value of the dividend stream as K (since it does not vary with J C ), and subtract S REIT (equation (1)) to obtain the additional value from taking UPREIT units: S DIFF 'S UPREIT &S REIT 'J C (V 0 &B)& J C (V 0 &B) (1%r) H % V 0 (1%r) H&V 0 % (1&J C )(V H &V 0 ) (1%r) H %K (3) The first term in brackets now reflects the lowered tax cost achieved by deferring the capital gains tax liability: the building seller is able to invest the tax for H years instead of paying it to the government today. Offsetting that benefit is the second term which measures the loss in value from not receiving the cash value of the building for H years as well. However, that value, V 0, is invested in the UPREIT and yields a discounted after-tax return that is accounted for in the last 7

11 two terms: the after-tax value of the capital appreciation in the UPREIT units plus the discounted value of the after-tax dividend payments. Although nearly all of the variables in equation (3) are easily observed or estimated, the optimal after-tax return the building seller would receive if she did not have to take UPREIT units, r, is unknown. We do know that the seller would never make an equivalent-risk investment that yielded less than the UPREIT units. The most favorable scenario for UPREITs then, is when the building seller would have invested her proceeds in UPREIT units whether or not she sold to the UPREIT. In that case, first term in brackets. V 0 '[(1&J C )(V H &V 0 )%V 0 ]/(1%r) H %K, and equation (3) reduces to the In sum, if the UPREIT receives the full benefit of the subsidy, it should pay S DIFF less for the property than would a REIT. If it pays more than S UPREIT but still less than S REIT, it does not bear the full incidence of the subsidy. III. Taxpayer Relief Act of 1997 The Taxpayer Relief Act of 1997 (TRA97) lowered the top tax rate on capital gains due to appreciation from 28 to 20 percent for assets held more than 18 months. For assets held longer than five years, the top tax rate fell to 18 percent but that rate cannot be applied until For capital gains due to depreciation, the tax rate only fell to 25 percent. There were no other changes in the legislation that should have affected UPREITs differently than REITs. Ideally, there would be some identifiable day or week where Congress surprised everyone by creating and passing the tax rate change. In Cutler s (1988) work, he identified two crucial and unexpected Congressional votes that resolved a number of the changes in the Tax Reform Act 8

12 of By looking at stock excess returns around those days, Cutler tried to identify the asset price effect of TRA86. One of his identifying assumptions was that since the time period under study was so short, few alternative hypotheses could explain any pattern of excess returns. Unfortunately, the capital gains tax provisions of TRA97 were the subject of speculation and debate for several months before the final passage of the legislation. 6 Table 1 lists the series of events leading up to the enactment of the bill. Even in early February, newspapers were reporting that there was a possibility of a capital gains tax cut. However, throughout February, March, and into April, speculation remained as the Clinton Administration and Congressional Republicans locked horns over whether Clinton would agree to cuts in the top capital gains rate to 20 percent or lower. In mid- to late-april, the negotiations moved towards reducing the capital gains tax rate. On April 30, the Congressional Budget Office reduced its estimate of the forthcoming budget deficit, reportedly the last event that enabled the Administration to reach a compromise with the Republicans. On May 3, a capital gains tax cut was announced, although the size of the cuts was still to be negotiated. On May 7, it was announced that whatever capital gains tax cuts were agreed to would take effect on that day. Until early- to mid-june, Congress debated how much to cut various taxes. On June 14, the House Ways and Means committee endorsed a plan to cut the top capital gains tax rate to 20 percent and, except for a Clinton last stand at the end of June, the next six weeks were spent nailing down the details of the bill that was finally passed on August 1. 6 Lang and Shackelford (1998) argue that the May 3, 1997 tentative budget agreement between the Clinton Administration and Congressional Republicans constituted the bulk of the news about the capital gains provisions in TRA97. 9

13 IV. Investigating the Incidence Question with Property Transactions Data If UPREITs bore some incidence of the tax subsidy, the capital gains tax rate reduction should have increased the expected prices UPREITs would have to pay for future acquisitions relative to REITs. Data on property acquisitions by REITs since the beginning of 1997 are examined to see whether this in fact occurred. Unfortunately, the available property transactions data are not ideal for our purposes. We observe only the prices reported by the companies that purchased the properties, not how much of the purchase was in cash or OP units. The only building quality controls are size (square footage) and age (year built), and they are not always reported. Table 2 outlines how the sample was constructed. Equity REITs purchased 4,442 properties between January 1, 1997 and May 31, 1998 the beginning and end of our sample period. Of that total, 2,430 observations in the SNL data were missing one of our three primary characteristics: acquisition price, square footage, or date built. 7 Table 3 reports the means and standard deviations of the transactions-level data. On observable characteristics, REITs and UPREITs tend to purchase similar-looking properties, with UPREITs buying a higher proportion of residential buildings and traditional REITs more likely to invest in retail. UPREITs also purchase much more property than traditional REITS: nearly 90 percent of the acquisitions in our sample were made by REITs. The specification estimated regresses building i s transaction price at time t (P it ) on an intercept term, an indicator variable denoting UPREIT status, time dummies corresponding to periods before and after TRA97, interactions of UPREIT status with the period dummies, and a 7 As far as we can determine, the failure to report is random. We also eliminate 95 transactions in property sectors other than retail, office, or residential. The results do not change when we include them but it makes for a clearer exposition to limit the number of property types. 10

14 series of covariates, Z. 8 Below, we use January 1, 1997-May 7, before the capital gains cut was decided -- as our baseline, denote May 7, 1997-December 31, 1997 as Period 2, and call January 1, 1998-May 31, 1998 Period 3. May 31, 1998 is the end of the sample period. The precise specification is given by equation (4), with Table 4 reporting the results. ln(p it )'(%* i %( t %( t % ( PERIOD2 it %( PERIOD3 it %( 6 Z it %µ it (4) In this specification, ( 4 provides an estimate of the difference between percent change in prices paid by UPREITs between periods 1 and 2 and the percent change in prices paid by REITs over the same time period. The estimated coefficient, ( 5, measures the relative difference between periods 1 and 3. The first column of Table 4 reports results using weighted OLS and acquisition price as the left-hand-side variable. The estimates of ( 4 and ( 5 are of most interest and are in bold-face. They indicate that prices paid by UPREITs rose 41 percentage points more than for REITs in the latter half of In the first half of 1998, UPREITs were paying 32 percentage points more than they were before TRA97, relative to REIT prices. While these results indicate that UPREIT shareholders themselves bore the incidence of the capital gains cut, the point estimates themselves strike us as too large to be believed. If taken literally, they imply that the incidence on UPREITs 8 Due to the lack of good data, we use the following: square feet, square feet2, age, age2, indicator variables for two of the three property sectors, the interaction of the property type dummies with square feet, the interaction of the property type dummies with age, the interaction of the property type dummies with period 2, and the interaction of the property type dummies with period 3. The results are not sensitive to the inclusion of these covariates. Including indicator variables for company, state, or metropolitan area, or state or metropolitan area interacted with period has virtually no effect on the results. 11

15 was three to four times the tax subsidy. 9 The large standard errors suggest that poor data quality and our inability to control completely for property level characteristics are a problem. Repeating the estimation using median regression instead of OLS yields the results reported in column two. The point estimates imply an incidence on UPREITs of between 50 and 80 percent, but the standard errors are so large that the coefficients are indistinguishable from zero or one. These results are statistically different from the point estimates in specification one, suggesting that outliers may be responsible for the large OLS estimates. The basic results are unchanged if we use price per square foot as the dependent variable (column three) or if we elaborately control for all of the property-level characteristics through a system of interacted dummy variables (in results not reported here.) If we compare before and after the passage of TRA97, the estimated incidence falls to 200 percent on the UPREITs (column four), but the standard error is so large that the result again is not distinguishable from zero or one. In sum, we interpret Table 4 as being consistent with at least some, and possibly all, of the incidence being borne by UPREITs. Unfortunately, the underlying data are not good enough to provide truly convincing evidence on this matter. Hence, we proceed to a detailed analysis of relative share price movements that are implied by UPREITs having to pay more for properties following the reduction in the capital gains rate. V. Investigating the Incidence Question with Firm Data 9 We do not mean to imply that it is not possible for the true incidence to be greater than one. For example, if the capital gains change were to reduce building sellers desire to hold UPREITs OP units (beyond the decrease in value), the incidence could be greater than one. However, other factors probably are at work to reduce the measured price response and, hence, the implied incidence. We only observe transactions that occurred before June of Long lead times on deals may have locked in pre-tra97 pricing on transactions that appear to have been consummated after the legislation was finalized. In addition, if UPREITs do not purchase properties on which the price has risen too much, the estimated coefficient again would be lower. 12

16 a. Sample and Variable Construction To investigate whether the capital gains changes in TRA97 were capitalized into firm prices, we begin with a list of all 202 existing REITs as of the first quarter of 1998 that was obtained from SNL Datasource, a computerized database that reports a wide variety of information culled from SEC filings of publicly-traded real estate companies. Table 5 shows how this sample ultimately is reduced to 129 companies. 10 We also obtained a variety of corporate characteristics from SNL such as whether the company used the UPREIT form, dividends declared each year, property focus, the date of the initial public offering, several measures of company size, and measures of institutional and insider ownership. This firm information is then matched to daily stock price data for 1996 and 1997 from the Center for Research in Securities Prices (CRSP) database. The price variable used is the average of the closing bid-ask prices, with the daily price averaged over a month before constructing the percent growth in share price. Thus the left-hand-side variable in the specifications reported below is %)P'lnP where 0 and 1 refer to calendar months &lnp 0 Taking the monthly average captures large, fundamental movements in share prices while minimizing sensitivity to the start 10 Forty-six companies were created during or after our sample period and had to be discarded. Two other firms were discarded because no stock price information could be obtained for them. Of the remaining 154 companies, 12 were mortgage or hybrid REITs. Since those types of REITs perform quite differently than equity REITs, they are not an appropriate comparison for UPREITs and were dropped from the sample. Finally, eleven REITs adopted either an UPREIT structure or a close substitute, the downreit structure, during our sample period. Because our estimation strategy requires that the UPREIT form be exogenous during the period of analysis, we dropped those firms that switched structure in 1996 or Removing the switchers may create a bias in that REITs may endogenously decide not to switch to the UPREIT form. The most plausible scenario would be that high growth REITs switch to UPREIT status. Since we delete the switchers, the remaining pool of REITs would be laggards and would have lower and lower growth over time. This bias would work against us finding any capitalization. 11 Dividend flows are so stable that using total return on the left-hand-side, which we compute as %RET'(P t &P s %(annualdividend)@(t&s)/12)/p s, where t and s correspond to specific months, makes no difference in the results. 13

17 date by reducing contamination from daily price fluctuations. Regressions are weighted by market capitalization at the beginning of the sample period, January 2, 1996, computed by multiplying the shares outstanding by the share price as reported by CRSP. 12 b. Empirical Approach and Basic Specification Due to the long gestation period of TRA97, share prices before the legislation was discussed are compared to prices after the bill was passed. This ensures that we measure the change in price between a period when the tax cut was completely unexpected and when it was known with certainty. Even if one could identify an event that was a surprise, it is impossible to know how much information was actually revealed during that period. Although comparing returns over a multiple-month time frame increases the possibility of spurious correlation with some other event, we are confident that comparing UPREITs to REITs will control for any factors that affect the REIT industry. More specifically, excess UPREIT share price growth (defined as above REIT share price growth) around the time of the tax change is compared to the excess UPREIT share price growth in the previous year. Assuming that UPREIT price growth relative to REIT price growth would otherwise be constant, any relative decline in UPREIT price growth around the time of the tax change must be due to the change in the value of the capital gains tax subsidy. Figure 5 plots the weighted average share price for REITs and UPREITs in 1996 and Comparing the first seven months of each year (i.e., the period around the passage of TRA97), UPREITs grew faster than REITs in 1996 but failed to do so beginning around February In addition, UPREIT share prices appear to change relative to REIT prices at certain key 12 Unweighted regressions yield slightly higher elasticities. 14

18 points in 1997: the end of April, when a tentative budget agreement was reached; the end of June when Clinton held out for a higher top marginal capital gains tax rate; and the end of July when the bill finally was passed. It is impossible to tell from the chart, however, whether or not those changes are due to noise or whether other factors might be affecting the share prices. It is also difficult to measure elasticities from a chart. Thus, we proceed with a more formal analysis. The effect of a capital gains tax change on UPREIT share prices can be written as follows: %)P'(%)P UPREIT,97 &%)P UPREIT,96 )&(%)P REIT,97 &%)P REIT,96 ) (5) where %)P is the average percent change in price for each company over the time period examined, UPREIT and REIT indicate the organizational form, and 96 and 97 refer to the year. This specification compares the percent difference between the mean percent growth in UPREIT prices and the mean percent growth in REIT prices in 1997 to To estimate equation (5), we rewrite it in deviations form and run the following regression: %)P t1&t0,it '"%$ i %$ t %$ Y1997 it %$ 4 X it %, it (6) where the left-hand-side variable is the percent growth in the average share price of company i from period t0 to period t1 in year t. UPREIT is an indicator variable for whether the company is an UPREIT and Y1997 takes a value of one if the year is X is a vector of covariates that may affect the relative growth rate of UPREITs and REITs over time. With this specification, fixed effects that affect the level of a company s share price are controlled for by looking at company-specific share price growth. If the real estate sector performed differently in 1997 than in 1996, the Y1997 indicator variable controls for that effect. If UPREITs grow faster than REITs on average, the UPREIT indicator variable nets that effect 15

19 out. Thus, the estimated coefficient, $ 1, will measure the additional return to the UPREIT structure relative to REITs in 1996, $ 2 corresponds to the average additional return to REITs in 1997 relative to 1996, and $ 3 is an estimate of the percent change in UPREIT share price growth relative to REITs between 1996 and 1997 described in equation (5). If the lower capital gains tax rate led to a decline in UPREIT share prices, the estimated $ 3 should be negative. The identification strategy employed here is well-known in the labor and public economics literatures (for examples, see Gruber (1994), Gruber and Poterba (1994), and Eissa (1995)). We think our application is especially appropriate because, unlike some other studies, we do not have to artificially divide a continuous distribution into discrete groups. Instead, we naturally have three binary categories: a company is either a REIT or an UPREIT, the year is either 1996 or 1997, and the time period within a year is either before or after the tax change. Further, we avoid a problem endemic to the event study literature -- having to estimate a market model to calculate excess returns -- by comparing UPREIT returns to those of REITs. c. Summary Statistics and Results Table 6 presents summary statistics on key variables used in the analysis. Table 7 then reports the results from estimating equation (6) over a number of different time periods. The dependent variable is the percent change between the average share price in the first month and the average share price in the second month listed in each column. In the first column, January is used as the start date because that was before any speculation of possible capital gains tax cuts. September is employed as the end month since the final legislation was passed on August 1 and we wanted to estimated a model that allowed sufficient time for the market to incorporate the information about the tax change into its 16

20 valuations. The small estimated value of for the coefficient on the 1997 year dummy in equation (6), $ 2, indicates that the traditional REITs did not experience any higher growth over January to September 1997 than they did during the same period in UPREITs, on the other hand, had their share prices increase by 7.4 percent relative to REITs between January and September 1996, as indicated by the estimate of $ 1 reported in the second row. Around the passage of TRA97, however, UPREIT share prices relative to REITs fell by 8.6 percent relative to 1996, as indicated by the estimated for $ 3 reported in row three. Thus, UPREIT share prices declined by 1.2 percent relative to REITs overall (8.6!7.4=1.2). The tax elasticity of share prices can be computed assuming a given change in the capital gains rate. Standardizing on a dollar s worth of value, the after-tax cost of a purchase for a traditional REIT is always $1 since it does not get any capital gains subsidy. If people thought that the average top capital gains tax rate was going to decline to 21 percent from 28 percent due to TRA97, the after-tax price of the same purchase for an UPREIT prior to May 7, 1997, can be written as 1!0.28=0.72. After May 7, the after-tax price increased to 1!0.21=0.79. Under these assumptions, the after-tax cost of purchasing a dollar s worth of building increases by 9.7 percent (0.07/0.72). 13 To compute the tax elasticity reported in Table 7, we replace the UPREITx1997 interaction by log(after-tax price) and instrument with UPREITx1997. The estimated tax elasticity of!0.93 in the first specification implies that nearly all of the 13 The statutory after-tax cost of purchasing a building rose 9.7 percent. In equation (3), this is tantamount to assuming that building sellers prefer real estate shares to alternative investments, have an infinite holding period or plan to die before selling the stock, and the building is fully depreciated. Since optimizing investors following a capital gains tax-minimization strategy should defer realizing their stock with the most capital gains tax liability as long as possible, we believe that UPREIT OP unitholders should be holding their shares until death. We also suspect that many of the buildings sold to UPREITs were largely depreciated. However, to the degree that these assumptions are incorrect, the increase in the effective after-tax cost will be less than for the statutory cost. 17

21 statutory change in the capital gains tax subsidy to the UPREIT structure was capitalized into UPREIT share prices. However, the standard error of 0.39 is large enough to make it impossible to reach precise conclusions regarding incidence. Still, we can comfortably reject the null hypothesis that the incidence on UPREITs is zero. The remainder of Table 7 examines different time periods to determine the robustness of the regression to the starting and ending dates. Comparing January to October makes virtually no difference in the estimated elasticity (column two). Ending the sample period in August yields a smaller estimated elasticity. Although the difference is not statistically significant, this result suggests that the market took some time to incorporate the effect of the tax rate change into market valuations. Moving the start date later, to February or March, reduces the estimated elasticity, a change consistent with the perceived probability of a tax change increasing during that period. 14 Table 8 incorporates additional explanatory variables that might be varying over time in a way that would affect UPREIT share prices relative to traditional REITs. This table suggests that the results regarding relative UPREIT share price growth in 1997 versus 1996 are quite robust to including a variety of covariates. For example, UPREITs may have changed their dividend payout ratios over time in a way that traditional REITs did not. Since share prices reflect dividend 14 As one check, our baseline specification was re-estimated around short periods of time when we suspected new information may have been revealed. These results were mixed; during three of the five periods that we identified as being times when news was released relative share prices moved in the predicted direction. In no case was any result statistically significant at conventional levels. We did not expect large or statistically significant results since the tax bill was gradually developed over several months. However, short horizon relative share price movements in response to news consistent with our expectations provides additional evidence that our results are not due to spurious movements in UPREIT relative to traditional REIT share prices. However, it is far from clear that there was any real news released during these specific periods; the information could have leaked earlier or there may not have been any informational content at all. 18

22 payments, that could induce a differential change between UPREIT and REIT share prices. In column one, companies with higher dividend yields command higher share prices as expected. However, including it has little effect on the relative growth of UPREIT share prices the estimated coefficient on the interaction term increases slightly in magnitude to!0.094 (0.036). Since REITs have such little control over how much they pay out in the form of dividends, this result should not be surprising. High return companies have to pay out more dividends since their taxable incomes are higher on average and they also have higher share price growth. 15 Column two includes the debt/asset ratio and return on assets, variables typically used as controls in this type of analysis. Their coefficients are very small and imprecisely estimated, and their inclusion has very little impact on the UPREITx1997 coefficient. Column three reports results for a specification controlling for the fraction of shares owned by insiders and institutions, as well as the vintage of the company (measured by the years since the firm s IPO). The percent of the company owned by insiders does not appear to affect share price growth at all. The percent of the company owned by institutions seems to have a small effect but it is not statistically different from zero. 16 Older companies have lower share price growth, but the rate of decline slows with age. However, including these covariates had little effect on the estimated coefficient on the interaction term Using a measure of total return on the left-hand-side that incorporates dividend payments does not have any effect of the estimated elasticities. 16 This result also could be due to the fact that institutions are more likely to invest when companies get larger, not the other way around. 17 We also estimated a specification that included the gross value of the company s properties at the end of 1995 as a proxy for company size. Companies with more than $2 billion in properties had higher than average share price growth rates, with smaller companies below average. However, because initial property value does not vary with year, it had no effect on the UPREITx1997 interaction term. 19

23 Next, we further test the robustness of the specification by controlling for property type. Table 9 shows that while the UPREIT structure exists in all property types, it is more heavily concentrated in the hotel, office building, and multifamily sectors. If these property sectors were at different stages of the real estate cycle, the different composition of UPREITs could generate a different return over time versus traditional REITs. In the first column of Table 10, a baseline specification is estimated that includes indicator variables for a number of property types -- health care, hotel, industrial/warehouse, office, residential, retail, and self-storage. 18 The point estimates indicate that health care and office were relatively slowly appreciating sectors during the first halves of 1996 and In column two, the return to each property focus is allowed to vary by year. Given the small number of firms in many property sectors, this specification asks a lot of the data as identification comes from the share price appreciation of UPREITs relative to REITs in 1997 relative to 1996 within property types. That is, health care UPREITs are compared to health care REITs, office UPREITs to office REITs, and so forth. The point estimate on the UPREITx1997 interaction term declines substantially, to!0.027 (0.039) and is not statistically different from zero. It also is not statistically different from the point estimate in the previous column. Since health care, office, and residential companies appear to have the greatest fluctuations in their relative growth rates between 1996 and 1997, we re-estimate the regression lumping all sectors but those three into the omitted category. The point estimate on the UPREITx1997 interaction term increases only slightly as reported in the third column. It is possible that the high estimated tax elasticity in the absence of property focus controls 18 The OTHER property type is the omitted category. 20

24 may be reflecting different market factors. It is also possible that performance in these sectors may be affected by the composition of UPREITs and REITs and the effect of the capital gains tax change in TRA97. We think a more likely explanation is that including property sector dummy interactions simply reduces the amount of variation to the point where it is impossible to discern an effect of the tax code on share prices. That is, there just is not enough within-property type variance to allow a reasonably powerful test of the incidence question. The specifications so far treat all UPREITs as if they are equally acquisitive. However, for an UPREIT s share price to be adversely affected by the reduction in tax subsidy in TRA97 the firm had to at least be planning on purchasing new properties. 19 Equation (7) below shows that, all else equal, UPREITs that derived more of their market value from planned property acquisitions should have been more adversely affected by the capital gains cut. Writing the expected present value of the rental stream on currently owned properties as E[R] and the expected rents from future acquisitions as E[A], the stylized market value of a firm is V=E[R]+E[A]. Percent changes in V can be decomposed into the weighted average percent changes in expected revenue from current or future properties: %)V' E[R] V (7) To the degree that the incidence of the subsidy accrues to UPREITs, when the capital gains tax rate changes E[A] will change proportionally. A change in capital gains rates should not affect the expected rent stream on current properties. Thus, the effect on V should be an increasing 19 A number of real estate companies may have chosen the UPREIT form to obtain a one-time capital gains deferral for a large property owner who wanted to securitize his portfolio. A change in the capital gains tax rate would not affect the share price for these companies if they were not planning on purchasing additional properties. 21

25 function of the weighting of future acquisitions in the market value of the company, E[A]/V. This proposition is tested in Table 11. The sample of UPREITs is divided based upon how acquisitive they were in Since one cannot measure expected acquisitions directly, we replace E[A]/V with 1!E[R]/V. We proxy for expected rent with 1995 total rental revenue. The denominator V is 1995 firm value (debt plus equity). Since a company with high E[R]/V must have low E[A]/V, companies with below-median rent-to-value ratios are deemed the most acquisitive. In the first column of Table 11, the share price appreciation of acquisitive UPREITs is allowed to vary relative to less-acquisitive UPREITs in 1997, but share prices are constrained to move together in This specification captures the potentially different effect of the capital gains tax change on the two classes of UPREITs. The least acquisitive income UPREITs declined in value relative to traditional REITs between 1996 and 1997, but the 3.7 percent change is small and not significantly different from zero. The share prices of the most acquisitive growth UPREITs fell 11 percent (3.4 percent standard error) more than the other UPREITs, relative to traditional REITs. All the share price response is coming through the channel we would expect: UPREITs that appear likely to be highly valued for their expected future purchases. In column two, we free up the specification to allow for the possibility that acquisitive UPREITs always have a lower rate of price appreciation than less-acquisitive UPREITs. The results show that although share prices for growth UPREITs rose slightly less than for income UPREITs in 1996 (!0.014), they declined substantially more relative to income UPREITs in 1997 than they did in 1996 (!0.099). One possible explanation for these results could be that prices for properties rose more in 22

26 1997 than in 1996 for some reason other than capital gains tax changes. If so, and if traditional REITs do not purchase many properties, only share prices for acquisitive UPREITs would decline. However, the results reported in column three cast doubt on this explanation. This specification compares the change in share prices in 1997 for acquisitive UPREITs relative to acquisitive REITs, all relative to less-acquisitive UPREITs and REITs and relative to Any change in property prices should affect acquisitive UPREITs and REITs equally while a change in capital gains taxes would reduce the share price of acquisitive UPREITs relative to acquisitive REITs. The capital gains explanation seems to be primary, as acquisitive UPREITs saw their share prices decline by a relative 14.6 percentage points (7.3 percent standard error). 20 We consider this evidence in column (3) to be the most convincing that UPREITs bore most of the incidence of the capital gains tax subsidy since this specification controls for every alternative explanation of the results that we can come up with. The point estimate of the incidence, however, is quite large 154 percent of the benefit of the tax subsidy accrues to the acquisitive UPREITs when we assume a 9.7 percent rise in the statutory after-tax cost of purchasing a building. Even though the standard errors are large enough that we cannot statistically distinguish this incidence from the 93 percent we measured earlier, a large share price elasticity should not be surprising. Several factors can increase the measured incidence. For one, levered firms should be more sensitive to changes in asset prices than we have assumed by using the statutory tax change as our benchmark. 21 Also, if the tax rate were to fall, building sellers 20 In addition, it appears from the rent-to-value distribution that the market expected that REITs are equally likely as UPREITs to be building purchasers ex ante. Even if REITs do purchase considerably less property ex post, the market still expected them to be acquiring. 21 A simple illustrative two-period example shows that leverage can cause share prices to fall by a greater percentage than the cost of a project increases. Assume that a firm uses the REIT industry average leverage, 35 23

27 would be likely to take a smaller proportion of units, diminishing the aggregate benefit of the UPREIT structure. The change in the statutory tax rate does not capture that effect, thus underestimating the potential change in UPREIT value with respect to changes in the capital gains tax rate. Finally, a reduction in the tax subsidy may have a second effect on asset values: not only will the price of new acquisitions rise for UPREITs, but the companies may reduce the amount of property buying they undertake in responsed to the higher price. If returns on new acquisitions are higher than on currently owner properties, firm value can fall even more than the tax change would suggest when the UPREIT s focus changes. VI. Conclusion By comparing the performance of two nearly-identical organizational forms of publiclytraded real estate companies, we are able to estimate the effect of the capital gains tax rate reduction in TRA97 on the share prices of UPREITs while holding all other industry-level and time-varying changes constant. The result gives insight into the classic question of the asset price incidence of taxes while minimizing the identification problems of previous work. In our preferred specification, we find that the capital gains tax rate changes in TRA97 led to a 8.6 percent decline in the share price of UPREITs relative to REITs, relative to the same time period in Assuming that REIT investors thought that the tax rate was going to fall to 21 percent, the estimated tax elasticity of the UPREIT share price is 0.93, indicating that the bulk of the incidence of the tax subsidy is borne by the buyer of properties, the UPREIT. percent, and that debt costs 7 percent. The equity return on a $10 million building than generates $1 million in revenue in the second period is 11.6 percent. ((1!3.5*0.07)/6.5) If the price of the building were instead $11 million but the second period revenue remained at $1 million, the equity return would decline to 10.2 percent. ((1!3.85*0.07)/7.15) That is a 12.1 percent decline in E[A] when the price of the building rose only 10 percent. 24