SPECULATIVE TRADING IN REITS

Transcription

1 SPECULATIVE TRADING IN REITS Benjamin M. Blau a and Ryan J. Whitby b Abstract: The role of speculative trading in markets is often debated. The recent extremes in the real estate economic cycle has created an ideal setting to investigate the role of speculative trading in the marketplace. Specifically, we focus on speculative trading in REITs during the recent boom and bust in real estate. While we find a strong relationship between speculative trading in REITs and the economic cycle, we do not find evidence that speculative trading is related to future returns. Increased speculative trading is apparent in REITs during the boom years, but the level of speculative trading in REITs is unrelated to the negative returns in REITs exhibited after the bust. a Blau is an Assistant Professor in the Jon M. Huntsman School of Business at Utah State University, Logan, Utah Phone: Fax: ben.blau@usu.edu. b Whitby is an Assistant Professor in the Jon M. Huntsman School of Business at Utah State University, Logan, Utah Phone: Fax: ryan.whitby@usu.edu. 1 Electronic copy available at:

2 1. INTRODUCTION In both the popular press and academic research, much has been said about speculative trading. With regards to the energy markets, the Wall Street Journal reported that Commodity Futures Trading Commission Chairman Gary Gensler indicated that every option must be on the table to curb excessive speculation. 1 On the other end of the spectrum, Buyuksahin and Harris (2010) state that speculators provide immediacy and facilitate the needs of hedgers by mitigating price risk, while adding to overall trading volume, which contributes to more liquid and well-functioning markets. A similar conclusion is reached in Friedman (1953). Other research describes both the positive and negative externalities associated with speculation in markets (Stein, 1987; Wang, 2010). Regardless of which side of the debate has more merit, a better understanding of speculative trading is critical to the design and improvement of financial markets. In this paper, we focus on the role of speculative trading in REITs during the latest real estate economic cycle. The recent boom and bust in real estate has created an ideal setting to investigate the role of speculative trading in the marketplace. The U.S. Commodity Futures Trading Commission defines a speculator as "a trader who does not hedge, but who trades with the objective of achieving profits through the successful anticipation of price movements." 2 With respect to real estate, we can think of speculation in several ways. One avenue for speculation in real estate is to buy and sell actual properties to try and profit from changing prices. In recent years, extreme examples were reported of speculators overbuilding and property flipping in cities across the United States. Speculators were betting that rapidly rising prices would translate into large profits. Another natural avenue to speculate on real estate is through the equity markets and the 1 The Politics of Speculation The Wall Street Journal, July 29, "CFTC Glossary: A guide to the language of the futures industry". cftc.gov. Commodity Futures Trading Commission Electronic copy available at:

3 purchase of REITs. REITs are an attractive asset to speculators for several reasons. First, REITs trade in liquid markets, which allows positions to be opened and closed daily if desired. Second, the transaction costs associated with purchasing REITs are orders of magnitude smaller when compared to purchasing real property. Third, while exposure to real estate through REITs is more diversified, the numerous REITs available for purchase allow speculators to focus on specific property types or regions without facing the varying levels of information asymmetry that would be encountered with more direct investments. While we focus our analysis on speculative trading in REITs, we acknowledge that the role of speculators in direct real estate markets could be quite different. Although speculative trading in the assets held by REITs would eventually show up in prices and could be related to the speculative trading of REIT securities, the illiquidity, uniqueness, and high transaction costs associated with the underlying assets allows for potentially large deviations between REIT prices and the underlying Net Asset Values (NAVs). We focus our analysis on two primary questions. First, what is the relationship between speculative trading and REITs during the recent real estate economic cycle? Both the anecdotal evidence of speculation in real property in recent years and the ease and convenience of trading equity REITs suggest an increase in the speculative trading of REITs. We find a strong relationship between our proxy for speculative trading and the economic cycle in real estate. While speculative trading in REITs is indistinguishable from speculative trading in non-reits over our entire sample ( ), mean difference of with a t-statistic of -0.55, the mean difference between REITs and Non-REITs during the real estate boom years 3 ( ) 3 We focus on 2003 to 2007 because those are the years with the highest median home values and the highest prices of the Ziman REIT index. In unreported tests, we also examine various time windows (i.e to 2005, 2002 to 2006, 2002 to 2007, etc.) for robustness and find qualitatively similar results. 3

4 is with a t-statistic of Furthermore, our multivariate tests confirm the strong relationship between speculative trading and REITs during the boom years is robust. The second question that we examine is whether or not speculative trading in REITs is related to future returns. Stein (1987) develops a model that shows that although increased speculation can benefit markets through greater risk sharing, it is also possible for increased speculation to change the information content of prices enough to have a destabilizing influence that outweighs the benefits. If speculative trading adversely affects market prices by driving them too high or too low, then speculative trading should have some predictive power with respect to future returns. In other words, if more speculation causes prices to deviate from fundamentals, then REITs with more speculative trading should have larger reversals. Conversely, if speculative trading is unrelated to future returns, then the relation between speculative trading and prices in REITs is less clear. Results in this study do not show that REITs with more speculative trading have larger losses during the bust period than REITs with less speculative trading. Differences between the top and bottom quartiles of firms ranked by speculative trading during the boom period indicate that firms with the most speculative trading actually had higher average returns than firms with the least speculative trading after the market crash. While average returns for the top and bottom quartiles are not significantly different, they are consistently positive, which does not support the notion that speculators attenuated the dramatic price reversal in REITs after While our results confirm that speculative trading in REITs increased during the recent real estate boom and decreased after the bust, we do not find evidence that supports the idea that the corresponding increase in speculative trading during the boom period harmed the well functioning REIT market or adversely affected market participants, with respect to return. Our 4

5 evidence is not consistent with the idea that increases in speculative trading harms markets, at least not with respect to the market for REITs. 2. RELATED LITERATURE Speculative trading has been examined from many different perspectives. Stein (1987) develops a model that considers the role of increased speculation in financial markets. He notes that speculation can benefit the marketplace by lowering the aggregate risk aversion or by changing the information content of prices. In some cases, speculators can negatively impact the market by making prices noisier, or changing the information content in a way that inflicts a negative externality on those already in the market. Wang (2010) considers the case where speculative traders add noise to the market and finds that speculative noise trading increases liquidity, but also results in less efficient prices. While many papers have addressed speculative trading with theoretical models, few papers have examined the question empirically. Llorente, Michaely, Saar, and Wang (2002) examine the relation between return and volume for individual stocks and develop a model where returns generated by speculating tend to continue and returns generated by hedging tend to reverse. We utilize their measure as our proxy for speculative trading. They describe the rationale behind their measure as follows: [W]hen a subset of investors sell a stock for hedging reasons, the stock s price must decrease to attract other investors to buy. Since the expectation of future stock payoff remains the same, the decrease in the price causes a low return in the current period and a high expected return for the next period. However, when a subset of investors sells a stock for speculative reasons, its price decreases, reflecting the negative private information about its future payoff. Since this information is usually only partially impounded into the price, the low return in the current period will be followed by a low return in the next period, when the negative private information is further reflected in the price. This example shows that hedging trades generate negatively autocorrelated returns and speculative trades generate positively autocorrelated returns (pg. 1005). 5

6 A detailed description of how speculative trading is calculated can be found in the Data section of the paper. Grishchenko, Litov, Mei (2006) utilize the measure of Llorente et al (2002) in the examination of stocks from emerging markets. Han and Kumar (2012) examine the role of speculative trading by retail investors and find that stocks that are dominated by speculative retail trading tend to be overpriced with significantly negative alphas. Buyuksahin and Harris (2010) examine the role of speculative trading in the crude oil futures market and find little evidence that speculators Granger-cause price changes. The recent volatility in the real estate markets and the associated financial crisis have also been the focus of many studies. Devos, Ong, Spieler, and Tsang (2012) examine the role of institutional investors in REITs during the financial crisis and find that institutional ownership increased prior to the crisis, but declined significantly during the crisis. Huang (2011) investigates the role expectations played in the recent housing boom and bust through a volatility feedback model. She finds a strong positive relationship between housing market volatility and expected housing returns. Anderson, Brooks, and Tsolacos (2011) test for periodic, partially collapsing speculative bubbles in US REITs. Driessen and Van Hemert (2012) study the pricing of commercial real estate derivatives during the financial crisis and find little systematic mispricing relative to REITs. While there are several papers that look at speculative trading in real estate, we are the first to closely examine the role of speculators in REITs during the recent boom and bust. Tegene and Kuchler (1993) examine speculative trading in farmland and find little evidence that speculative trading affects prices. Bjorklund and Soderberg (1999) look at speculative trading of real estate in Sweden and find that speculation partly explains the real estate bubble during the 1980s. Malpezzi and Wachter (2005) develop a model that examines land speculation. They 6

7 find that land speculation only impacts prices when supply is inelastic. Case, Cotter, and Gabriel (2011) develop a multifactor asset-pricing model for housing and conclude that speculative forces are an important determinant for U.S. housing returns. Focusing more on real estate after the financial crisis, Zhou and Anderson (2011) examine the role of herding behavior in the US REIT market and find that investors are more likely to herd in REITs when market conditions are turbulent. They also find that circumstances that lead to herding have evolved since the recent financial crisis. This paper contributes to the literature in several ways. First, we find clear evidence that speculators utilized REITs during the housing boom. Understanding the role of speculators in different markets and time periods allows for both better market design and regulation. Second, examining speculative trading during both the boom and bust periods allow us to examine whether speculative trading exacerbated the pain felt during the bust. While we find significant increases in speculative trading, we do not find that more speculative trading resulted in bigger losses or more volatility. Third, while many theoretical models have examined speculative trading, fewer papers have documented the role of speculation empirically. While our analysis is focused on the REIT market, our results have broader implications and support the model developed by Llorente et al. (2002). 3. DATA DESCRIPTION The data used in the analysis is obtained from a variety of sources. From the Center for Research on Security Prices (CRSP), we obtain daily returns, volume, prices, shares outstanding, etc. From Ziman, we obtain the REIT type (i.e. equity, mortgage, or hybrid) and property type (i.e. diversified, residential, retail, etc.). Our sample time period extends from 1993 to 2011 and we obtain the universe of REITs that are available on CRSP/Ziman Real Estate. We begin our 7

8 analysis in 1993 to coincide with the modern REIT era. The total number of unique REITs is The total number of REIT-year observations is 3,814. Similarly, from CRSP we gather data on the universe of non-reits. In the subsample of non-reits, we have 17,795 unique securities and 139,031 non-reit-year observations. From CRSP/Ziman, we obtain the following property types for our sample of REITs: Diversified, Retail, Residential, Industrial/Office, Self Storage, Hotel/Lodging, and Healthcare. Of the 500 (3,814) REITs (REIT-year observations), CRSP/Ziman does not provide a property focus for 9 REITs (15 REIT-year observations). Further, CRSP/Ziman classifies the property focus of 4 REITs (10 REIT-year observations) as unknown and 30 REITS (207 REIT-year observations) as unclassified. In the case that CRSP/Ziman does not provide a property type, or provides a property type of unknown or unclassified, we classify these REITs with a property type of Other. To provide an estimate of speculative trading, we follow Llorente, Michaely, Saar, and Wang (2002), who examine the dynamic relation between returns and volume of individual securities. They argue that after controlling for volume, hedging trades will generate negatively autocorrelated returns while speculative trades will generate positively autocorrelated returns. We closely follow the empirical methods of Llorente et al. (2002) when estimating speculative trading. For instance, we estimate daily turnover, which is equal to the ratio of daily volume to shares outstanding. Lo and Wang (2000) indicate that the daily time series of turnover is nonstationary, so Llorente et al. (2002) detrend the time series and take the log of turnover. On days when volume is zero, they add a small constant ( ), which has been shown to maximize the likelihood of normally distributed trading volume at the daily level (Richardson, 4 We also analyze only equity REITs identified by Feng, Price, and Sirmans (2011) and find even stronger results. Results for the full sample are reported so that differences between REIT type can be analyzed in our multivariate analysis. 8

9 Sefcik, and Thompson, 1986; Ajinkya and Jain, 1989; Cready and Ramanan, 1991). We calculate our measures of turnover in the following way. logturnover i,t = log(turnover i,t ) (1) V i,t = 1 logturnover i, t logturnoveri, t (2) s= 200 V i,t is our measure of trading activity that is used to estimate speculative trading and is obtained by taking the difference between the log of turnover and mean of the log of turnover from day t-1 to t-200, where day t is the current trading day. Using CRSP daily returns and V i,t we then estimate the following time series equation for the universe of securities in our sample. R i,t+1 = β 0 + β 1 R i,t + β 2 R i,t V i,t + ε i,t+1 (3) Equation (3) shows a simple autoregressive formula where daily returns for each stock on day t+1 are regressed on daily returns for each stock on day t. Llorente et al. (2002) include the interaction between R i,t and V i,t in equation (3) to obtain the estimate for speculative trading. The idea is that the larger (and more positive) the estimate for β 2, the more likely that trading activity increases the return autocorrelation. Therefore, the estimate for β 2 is our estimate for speculative trading according the model in Llorente et al. (2002). While the argument by Llorente et al. (2002) that positively autocorrelated returns proxy for speculative trading is fairly straightforward, the idea of hedging trades for an individual security is somewhat awkward. They also describe hedging trades as allocational shocks, which we think is more representative. In other words, trades are being made for allocation or liquidity reasons, not because of any information or expectation related to future prices. Regardless of how a negative estimate of β 2 is described, we think that Llorente et al. have developed a useful measure that empirically identifies speculative trading for individual securities. We recognize, however, that the relation between speculative trading and hedging might not be as simple as the linear model proposed by 9

10 Llorente et al. (2002). And that using the estimate of β 2 as a proxy for the level of speculative trading assumes a linear relation that, if misspecified, could result in incorrect inferences. Therefore, in unreported tests, we conduct a variety of robustness tests in which we censor the estimate of β 2 to only be positive. That is, we approximate speculative trading using the estimate of β 2 when β 2 is positive zero otherwise. We also use a censored estimate of β 2 to equal this estimate when the estimate is in the 75 th percentile (in a particular year), zero otherwise. We also use a simple binary specification in which speculative trading is equal to 1 if β 2 is in the 75 th percentile, zero otherwise. We replicate the analysis below using these censored estimates in various Tobit regressions and find results consistent with the reported findings. We choose to focus our attention on results that use the continuous estimate of β 2 because it is easier to interpret and directly comparable to other papers that use Llorente et al. (2002). Table 1 reports statistics that summarize our sample. Panel A reports the summary statistics for our sample of REITs while Panel B shows the statistics for our sample of non-reits. The variables we analyze in the table include Spec, which is our measure of speculative trading or our estimate of β 2 from equation (3) for each stock in each year during our sample time period. We also include the market capitalization (Size), the price of each security (Price), and the daily turnover (Turn). We estimate a daily Capital Asset Pricing Model during each year for each stock and obtain an estimate for systematic risk (Beta). Using the residual returns from the daily CAPM model, we calculate the standard deviation of these residuals in order to calculate idiosyncratic volatility (IdioVolt). Panel A shows that the average REIT has a Spec of , an average market capitalization of $1.13 billion, and an average price is $ The average REIT has a turnover of , suggesting that approximately 115% of shares outstanding are traded during a particular year. 10

11 The average REIT has a Beta of and an average idiosyncratic volatility of 2.23%. Panel B shows that the average non-reit has Spec of , Size of $1.93 billion, a Price of $30.04, a Turn of , a Beta of , and an IdioVolt of Panel C reports the difference between the means reported in Panels A and B. Comparing Spec in Panels A and B, we see that Spec is more than eight times larger for our sample of non- REITs. However, while the difference is , the difference is not statistically different from zero (t-statistic = -0.55). In column [2], we find that the difference in Size between Panels A and B is -$805,204,614 and is statistically different from zero (t-statistic = -4.56) indicating that REITs are generally smaller, in terms of market capitalization, than non-reits. Column [3] shows that the difference in Price is statistically close to zero (difference = -8.74, t-statistic = ). In column [4], the difference in Turn is and is statistically significant (t-statistic = -3.66) indicating that non-reits generally have more trading activity than REITs. Finally, columns [5] and [6] show that the differences in systematic risk (Beta) and idiosyncratic risk (IdioVolt) between Panels A and B are both statistically significant (differences = , ; with respective t-statistics = -4.72, ), indicating that non-reits have more systematic and idiosyncratic risk than REITs. The findings from Table 1 indicate that while various characteristics differ between our samples of REITs and non-reits, the measure of speculative trading is similar between samples during our time period of 1993 to EMPIRICAL TESTS Next, we begin to address our research question examining speculative trading in REITs during the period when real estate prices were the highest. We begin by plotting median real estate prices according to S&P/Case Shiller. The top panel of Figure 1 shows median real estate prices as well as the normalized price of the Ziman REIT index from 1993 to As seen in 11

12 the figure, there is generally an upward trend in median home prices from 1993 to However, in 2003, both median home prices and prices of the Ziman index were higher thant they had been in the previous 10 years. Prices seem to peak in 2005 and 2006 and then subsequently decline. In the second panel of Figure 1, we focus on the percentage change in median real estate prices and 3-year cumulative returns of the Ziman REIT index. We find that in the second panel, boom period in REITs and in median home prices appears to occur between 2004 to Beginning in 2002, and continuing until 2005, the percentage change in median home prices was above 10%. In 2004 and 2005, the percentage change in prices was above 14.5%. The 3-year cumulative return of the Ziman REIT index is highest in 2005 at nearly 70%. The bottom panel of the figure shows speculative trading for our sample of REITs and non- REITs. While the results in Table 1 show that the mean estimates of speculative trading between samples are statistically similar when examining the entire time period, we find some variation in speculative trading across time. In particular, we find that speculative trading in REITs becomes observationally higher than speculative trading in non-reits from 2002 to 2008 (with the exception in 2005). We recognize that other security-specific factors might be influencing this variation so we are cautious when inferring anything from the figure. 3.1 Speculative Trading in REITs and Non-REITs During Periods With Increasing Real Estate Prices Univariate Tests Next, we test for statistical differences in speculative trading between samples across time. Table 2 reports our estimates for speculative trading for various time windows. Panel A reports the results for relatively equal time periods. The panel shows that Spec for REITs is during the five year period of 1993 to 1997 while Spec for non-reits is The difference is but statistically close to zero (t-statistic = -1.05). During the five-year period from 12

13 1998 to 2002, the mean estimate for speculative trading in REITs is while the mean estimate for speculative trading in non-reits is The difference in this second time window is also statistically close to zero (difference = , t-statistic = -1.20). During the period 2003 to 2006, we find that Spec for REITs is while Spec for non-reits is The difference is and is statistically different from zero (t-statistic = 2.92). In economic terms, Spec for REITs is nearly 2.4 times larger than Spec for non-reits during the period. We also note that this period corresponds with a period when real estate prices were highest (see Figure 1). Finally, in the last five-year period from 2007 to 2011, the mean estimate for speculative trading for REITs is while the mean estimate for speculative trading for non- REITs is The difference in means is again statistically close to zero (difference = , t-statistic = -1.57). Panel B reports time windows that directly correspond to Figure 1. First, the time period 2002 to 2005 is the period when the percentage change in median real estate prices was the highest and the 3-year cumulative return of the Ziman REIT index. The average percentage change during this four-year period was more than 12.6%. Second, the time period from 2003 to 2007 represents a period when real estate prices and the prices of Ziman REIT index were highest. Third, the period from 2008 to 2011 represented a period of sharp decline in real estate prices. Table 2 Panel B shows that the mean estimate for Spec for REITs was during the first time window. The mean estimate for Spec for non-reits was The difference is both economically and statistically different from zero (difference = , t-statistic = 3.88). In particular, Spec for REITs during this period was approximately 2.4 times larger than Spec for non-reits. In our second time period (2003 to 2007), we also find that the mean estimate for Spec was significantly larger than the mean estimate for Spec for our sample of non-reits 13

14 (difference = , t-statistic = 2.68). In economic terms, the Spec for REITs was nearly 3.5 times larger than Spec for our sample of non-reits. Finally, we find that, during our last time period (2008 to 2011), the mean estimate for speculative trading for our REIT sample was while the mean estimate of speculative trading for our non-reit sample was The difference is statistically different from zero and suggests that, according to the model in Llorente et al. (2002), more hedge trading occurred in REITs than in non-reits during this last time period. The results from Table 2 indicate that while our estimates for speculative trading are similar across samples when examining the entire time period, sample differences in speculative trading occur during the most recent time period. Further, the univariate tests in Table 2 seem to indicate that speculative trading in REITs was more prevalent during periods when real estate prices were highest and were growing at the fastest rate. We continue our tests by estimating univariate correlations between our estimate for speculative trading and other security-specific characteristics that we use in our multivariate analysis below. Table 3 reports Spearman correlations between Spec, and several variables that are defined in Table 1 (i.e. Size, Price, Turn, Beta, and IdioVolt). After pooling our sample of REITs and non-reits together, we also include an indicator variable REIT, which equals one if a particular security is a REIT zero otherwise. Panel A reports the correlations using all years in our sample time period. Focusing on the first row of Panel A, we observe that speculative trading is negatively related to Size, Price, Turn, and Beta, and positively to IdioVolt. We also find that Spec is unrelated to the dummy variable REIT (correlation = , p-value = 0.264). These results tend to support our findings in Table 1 that show that speculative trading is similar between our two samples when examining the entire time period. 14

15 We also note that there exists a strong cross-correlation between Size, Price, Turn, Beta, and IdioVolt. In column [7] we see that the REIT indicator variable is directly related to Size and Price. However, we find that the variable REIT is inversely related to Turn, Beta, and IdioVolt. For brevity, we report the results for the time period 2003 to 2007 in panel B also similar results are found when we examine the time period 2002 to In the first row of Panel B, we find that speculative trading is negatively related to Size, Price, Turn, Beta, and IdioVolt. 5 However, we find that Spec is also positively related to the indicator variable REIT supporting our findings in Table Speculative Trading in REITs and Non-REITs During Periods With Increasing Real Estate Prices Multivariate Tests Thus far, Tables 2 and 3 have shown that our estimate of speculative trading for REITs is similar in magnitude to the estimate for speculative trading for non-reits when examining the period prior to However, beginning in 2003 and continuing to 2007, the level of speculative trading in REITs is markedly higher than the level of speculative trading in non- REITs. This observed increase in speculative trading in our REIT sample corresponds with a period when real estate prices and Ziman REIT index prices were highest. We recognize, however, the need to control for other factors that might influence the level of speculative trading. Therefore, we estimate the following equation using pooled data that includes both the REIT-year observations and the non-reit-year observations. Spec i,t = β 0 + β 1 ln(size i,t ) + β 2 ln(price i,t ) + β 3 Turn i,t + β 4 Beta i,t + β 5 IdioVolt i,t + β 6 REIT i + ε i,t (4) The dependent variable is Spec, which is the Llorente et al. (2002) measure of speculative trading. We include as independent variables the natural log of market capitalization (ln(size i,t )), 5 We note that the univariate correlation between size and speculative trading is only marginally significant in Panel B (p-value = 0.115). 15

16 the natural log of price (ln(price i,t )), the share turnover (Turn i,t ), the level of systematic risk (Beta i,t ), and the idiosyncratic volatility (IdioVolt i,t ). The variable of interest is the indicator variable REIT, which equals one if the security is a REIT zero otherwise. We report t-statistics that control for two dimensional clustering although similar results are found when we use White (1980) standard errors that control for conditional heteroskedasticity. Furthermore, we include Year Fixed Effects in some of the econometric specifications. 6 Given the level of cross correlation that exists between the independent variables (see Table 2), we estimate variance inflation factors to determine whether our results suffer from multicollinearity bias. Variance inflation factors are relatively small and are each under three indicating that our findings do not suffer from bias casued by multicollinearity. However, in Table 4, we report various specifications of equation (4) by including different combinations of independent variables to show that our results hold regardless of the control variables we include. Table 4 Panel A reports the results for the entire sample time period. Columns [1] through [3] show the results without controls for Year Fixed Effects. Column [1] shows that Spec is negatively related to both the natural logs of Size and Price. However, the indicator variable REIT produces an estimate that is statistically close to zero (estimate = , t-statistic = ). Column [2] provides some evidence that idiosyncratic volatility is directly related to speculative trading activity. However, Turn and Beta do not provide estimates that are statistically different from zero. The variable of interest, REIT, again produces an estimate that is both economically and statistically insignificant (estimate = , t-statistic = -0.32). In the full model (without Year Fixed Effects), column [3] shows that Spec is negatively related to the natural logs of Size and Price and is unrelated to Turn, Beta, and IdioVolt. Further, the estimate 6 Because we include the variable REIT, we cannot include security fixed effects because the variable REIT does not vary across the time series and therefore, cross-sectional fixed-effects estimates will be inconsistent. 16

17 for REIT is and is statistically close to zero (t-statistic = -0.34). We are able to draw similar conclusions when examining the results in columns [4] through [6] that control for Year Fixed Effects. A few results are noteworthy. First, the natural log of Size produces an insignificant estimate in columns [4] and [6]. Second, the estimate for the indicator variable REIT is statistically close to zero in all three columns. These findings support the results in columns [1] through [3] and further provide evidence that, when examining the entire sample time period, the level of speculative trading in REITs is similar to the level of speculative trading in non-reits. Panel B presents the results when examining the period when real estate prices and Ziman index prices were highest (years 2003 to 2007). We only discuss the findings in column [6] for brevity. During this period, we find that the natural log of Size produces a positive and significant estimate (estimate = , t-statistic = 2.14). Turn, on the other hand, produces an estimate that is negative and marginally significant (estimate = , t-statistic = -1.69). Again, the indicator variable REIT produces an estimate that is both positive and significant (estimate = , t-statistic = 2.29). Relative to the mean estimate for speculative trading for REITs during the entire time period, the estimate is more than 17 times greater than the mean. As before, the estimate for REIT is positive and significant in each of the columns in Panel B of Table 4 and suggests that the level of speculative trading in REITs was greater than the level of speculative trading in non-reits during the period when real estate prices were highest. Similar results are found when we use various time windows to capture the real estate boom years. We also note that the regression results are robust when we redefine the dependent variable as the estimate of β 2 in equation (3) if the estimate is in the 75 th percentile in a particular year zero otherwise. The Tobit regression results are qualitatively similar to those reported in Table 4. 17

18 3.3 Speculative Trading in REITs During the Boom and Bust Periods Univariate Tests Next, we compare the level of speculative trading in REITs during the periods when real estate prices were highest and the growth rate in real estate prices was the highest. Table 5 reports the level of speculative trading in REITs for the period when real estate prices and Ziman index prices were highest and the rest of the sample time period. For exposition, we denote the period 2003 to 2007 as the boom period and the years 1993 to 2001, and years 2006 to 2011 as the non-boom period. In the first row of Panel A, we find that the mean level of speculative trading is during the non-boom period. However, we find that the mean level of Spec is during the boom period. The difference is (t-statistic = -3.82). In economic terms, the level of speculative trading in REITs is nearly four times greater during the boom period than during the non-boom period. In rows 2 through 4, we partition the REITs into equity REITs, mortgage REITs, and hybrid REITs to determine which type of REIT drives the observed increase in the level of speculative trading. We find that the level of speculative trading in equity REITs during the boom period is significantly greater than the level of speculative trading in equity REITs during the non-boom period (difference = , t-statistic = -4.09). Further, we do not find that speculative trading during the boom period is statistically different than speculative trading during the non-boom period for either mortgage REITs or hybrid REITs. These results indicate that equity REITs drive the observed increase in speculative trading during the boom period. Next, we examine whether the property focus of REITs play a role in the level of speculative trading during the boom period. Using the property-type focuses from CRSP/Ziman, we partition the level of speculative trading during the boom period and non-boom period for each 18

19 of the eight property focuses. 7 Focusing primarily on column [3], we find that REITs with a property focus of Diversified, Residential, Industrial/Office, and Hotel/Lodging have higher levels of speculative trading during the boom period than during the non-boom period. 3.4 Speculative Trading in REITs During the Boom and Bust Periods Multivariate Tests We recognize the need to control for other factors that influence the level of speculative trading in REITs in a multivariate framework. Therefore, we estimate the following equation using pooled REIT-year observations. Spec i,t = β 0 + β 1 ln(size i,t ) + β 2 ln(price i,t ) + β 3 Turn i,t + β 4 Beta i,t + β 5 IdioVolt i,t + β 6 DYEAR t +ε i,t (5) Once again, the dependent variable is Spec, which is the Llorente et al. (2002) measure of speculative trading. As independent variables, we include the natural log of market capitalization (ln(size i,t )), the natural log of price (ln(price i,t )), the share turnover (Turn i,t ), the level of systematic risk (Beta i,t ), and the idiosyncratic volatility (IdioVolt i,t ). The variable of interest is the indicator variables DYEAR. DYEAR is defined as an indicator variable (D03-07) that equals one during the years 2003 through 2007 zero otherwise. We report t-statistics that control for two dimensional clustering although similar results are found when we use White (1980) standard errors that control for conditional heteroskedasticity. In order to test for the presence of multicollinearity, we again estimate variance inflation factors. We find that inflation factors are each below 3.30 indicating that our results are not subject to multicollinearity bias. When including DYEAR, we do not include Year Fixed Effects in order to meet the full rank condition required for consistent estimates. Like previous tables, we estimate different versions of equation (5) to show that the main inferences that we can draw from our results are robust to 7 We note that CRSP/Ziman also includes the property focus mortgage, but nearly all of these REITs are considered mortgage REITS, so we did not include mortgage as a property type as the results are almost identical to the REIT-type mortgage. 19

20 the inclusion of different combinations of control variables. As before, our results are qualitatively similar across columns so, for brevity, we only discuss the results of the full model (column [3]). Column [3] shows that Turn produces a negative estimate (estimate = , t-statistic = ). All of the other control variables produce estimates that are statistically close to zero. We do find, however, that the variable D03-07 produces a reliably positive estimate (estimate = , t-statistic = 5.36). In terms of magnitude, the estimate for D03-07 is more than 40 times greater than the mean estimate for Spec for REITs during the entire time period (Table 1). These results are similar in sign and magnitude to those in columns [1] and [2] and indicate that the level of speculative trading in REITs was substantially higher during the period when real estate prices and REIT prices were the greatest. Next, we extend the tests from Table 6 to include the REIT type. In Table 5, we found that the higher levels of speculative trading in REITs during the period when real estate prices were highest were driven by equity REITs. To test for the effect of REIT type in a multivariate framework, we extend equation (5) in the following way. Spec i,t = β 0 + β 1 ln(size i,t ) + β 2 ln(price i,t ) + β 3 Turn i,t + β 4 Beta i,t + β 5 IdioVolt i,t + β 6 E-REIT i + β 7 DYEAR t + β 8 E-REIT i DYEAR t + ε i,t (6) As before, the dependent variable is Spec. As independent variables, we include the natural log of market capitalization (ln(size i,t )), the natural log of price (ln(price i,t )), the share turnover (Turn i,t ), the level of systematic risk (Beta i,t ), and the idiosyncratic volatility (IdioVolt i,t ). We also include the indicator variable DYEAR, which is equal to the indicator variable D D03-07 equals one when the years are between 2003 and We include an indicator variable capturing whether the REIT is an equity REIT (E-REIT). Finally, we also include an interaction between these indicator variables. As before, we report t-statistics that control for two 20

21 dimensional clustering although similar results are found when we use White (1980) standard errors that control for conditional heteroskedasticity. 8 Table 7 reports the results from estimating equation (6). Column [1] is similar to column [3] (in the previous table. The difference however, is that we include the indicator variable E-REIT. Column [1] shows that the estimate for E-REIT is statistically close to zero (estimate = , t- statistic = 0.31). However, we still find that the variable D03-07 produces a reliably positive estimate (estimate = , t-statistic = 6.01). Column [2] shows the results when we include the interaction variable. Consistent with our univariate findings in Table 5, we see that the interaction estimate is positive and significant (estimate = , t-statistic = 3.08) indicating that the increase in speculative trading during years 2002 to 2005 is driven by equity REITs as opposed to mortgage REITs or hybrid REITs. Further, we find that the estimate for D03-07 is positive but statistically close to zero (estimate = , t-statistic = 0.45). This statistically insignificant estimate indicates that the level of speculative trading in mortgage REITs and hybrid REITs does not increase during the period 2003 to We continue our multivariate tests by estimating the following equation using pooled REITyear data. Spec i,t = β 0 + β 1 ln(size i,t ) + β 2 ln(price i,t ) + β 3 Turn i,t + β 4 Beta i,t + β 5 IdioVolt i,t + β 6 DYEAR t + β 7 PropType i + β 7 DYEAR t PropType i +ε i,t (7) The dependent variable and the independent variables have been defined in the previous equation with one exception. Instead of including the indicator variable E-REIT, we include a different indicator variable PropType. We define PropType seven different ways. First, PropType is defined as Diverse, which equals one if the property focus of the REIT is classified 8 We should note that we do not include Year Fixed Effects because doing so would violate the full rank requirement for consistent estimates. In particular, the indicator variable DYEAR and Year Fixed Effects are linear functions of each other. 21

22 as Diversified according the CRSP/Ziman zero otherwise. Second, we define PropType as Retail, which is equal to one if the property focus is Retail zero otherwise. Third, we define PropType as Resident, which is equal to one if the property focus is Residential zero otherwise. Fourth, we define PropType as Office, which is equal to one if the property focus is Industrial/Office zero otherwise. Fifth, we define PropType as Storage, which is equal to one if the property focus is Self Storage zero otherwise. Sixth, we define PropType as Lodging, which is equal to one if the property focus is Hotel/Lodging zero otherwise. Finally, we define PropType as Health, which is equal to one if the property focus is Healthcare zero otherwise. The omitted category is Other which accounts for REITs that are unclassified or unknown according to the CRSP/Ziman data. As before, we include an indicator variable DYEAR, which is defined as D The variable of interest is the interaction between DYEAR and PropType. We again report t-statistics that account for two dimensional clustering and do not control for Year Fixed Effects in order to obtain full rank. In Table 5, we provide evidence that the increase in speculative trading in REITs is driven primarily by REITs with Diversified, Residential, Industrial/Office, and Hotel/Lodging property focuses. The estimation of equation (7) and the interaction estimates in particular will allow us to make inferences about which property type attracts the most speculative trading after including some control variables. Table 8 reports the results. In each of the columns, we find that the estimate for D03-07 is positive and significant indicating that regardless of which interaction variable is included, REITs generally had higher levels of speculative trading during the period when real estate prices grew the most. Column [1] shows that the interaction between D03-07 and Diverse is statistically close to zero (estimate = , t-statistic = -0.32). In fact, in each column, the interaction variables produce estimates that are statistically insignificant. 22

23 Some other noteworthy results are that the indicator variables Diverse, Resident, and Storage produce reliably positive estimates indicating that, in general, REITs with these types of the property focuses have higher levels of speculative trading. The insignificant interaction estimates, however, suggest that after controlling for other factors that might influence the level of the speculative trading, no particular property type drives the observed increase in speculative trading during the period when real estate prices were growing the most. 3.5 Speculative Trading and Future REIT Returns Univariate Tests In our last set of tests, we examine whether the level of speculative trading contributed to the substantial crash in REITs during the year In unreported results, our sample of REITs underperformed during The average CRSP raw return for our REIT sample during 2008 was -36.7%. Table 9 provides some univariate tests. In particular, the table reports the REIT returns in 2008 across four portfolios of REITs that are based on the level of speculative trading during the period 2003 to Quartile I (Q I) contains the REITs with the least speculative trading during the time period while Quartile IV (Q IV) contains the REITs with the most speculative trading during the time period. Column [5] reports the difference in 2008 REIT returns between extreme quartiles along with a corresponding t-statistic testing for significance of the difference. We report four different measures of returns. First, we include CRSP raw returns. Second, we calculate adjust returns (Adj. Returns) as the difference between a particular REIT raw return and the return of the CRSP Ziman (value-weighted) index. Third, we estimate a daily Fama-French Three-Factor model and obtain the residual returns. Therefore, FF3F Returns are the cumulative 2008 returns from daily FF3F residual returns. Similarly, Fama-French Four- Factor Returns (FF4F Returns) are the cumulative 2008 returns from daily FF4F residual returns, where the fourth factor is the momentum factor. 23

24 In the first row of Table 9, we find that 2008 CRSP raw returns are neither increasing nor decreasing across speculative trading quartiles. Column [5] shows that the difference between extreme quartiles is statistically close to zero (difference = , t-statistic = 1.13). Similar results are found when we examine Adj. Returns, FF3F Returns, and FF4F Returns. In each case, the difference between extreme quartiles is effectively zero indicating that the level of speculative trading during the period when real estate prices grew the most is orthogonal to the large price decline in REITs during Speculative Trading and Future REIT Returns Multivariate Tests We recognize the need to control for other factors that possibly influenced the level of 2008 REIT returns. Table 10 reports the results from estimating the following equation using crosssectional data. 2008Returns i = β 0 + β 1 ln(size i, ) + β 2 ln(price i ) + β 3 Turn i + β 4 Beta i + β 5 IdioVolt i, + β 6 Spec i +ε i (8) The dependent variables include our four measures of returns during 2008 (Raw Returns, Adj. Returns, FF3F Returns, and FF4F Returns). As independent variables, we include the natural log of market capitalization (ln(size i )), the natural log of price (ln(price i )), the share turnover (Turn i ), the level of systematic risk (Beta i ), and the idiosyncratic volatility (IdioVolt i ). The variables of interest are the variable Spec, which is Llorente et al. (2002) measure of speculative trading. All independent variables are measured from years 2003 to We report t-statistics that control conditional heteroskedasticity using White (1980) standard errors. The results are qualitatively similar across columns so for brevity, we will only discuss our findings in column [1]. We find that Size, Price, Turn, Beta, and IdioVolt produce estimates that are statistically close to zero. These results indicate that none of the control variables, which are measured from 2002 to 2005 in this column, explain the variation in 2008 REIT returns. 24

25 Furthermore, we find that the estimate for the variable of interest Spec is also statistically close to zero (estimate = , t-statistic = 1.34). The insignificant estimate for Spec indicates that, while REIT speculative trading levels were unusually high during 2003 to 2007, the high levels of speculative trading did not contribute the substantial price decline of REITs in In each column, the estimate for Spec is statistically close to zero. These results support our univariate tests in Table 9 and suggest that the unusual levels of speculative trading in REITs during the period when real estate prices were increasing did not affect the REIT returns in CONCLUSION To examine the role of speculative trading in REITs during the most recent boom and bust period in real estate, we utilize a measure of speculative trading developed by Llorente et al. (2002). They argue that hedging trades will generate negatively autocorrelated returns while speculative trades will generate positively autocorrelated returns. While we find no differences between the speculative trading in REITs and non-reits from 1993 to 2011, we do find differences in speculative trading for specific subsamples. Corresponding with the period when median home prices and prices of the Ziman REIT index were highest ( ), we find significant differences in speculative trading between REITs and non-reits. In unreported results, we find similar results for other various time windows around this period. Furthermore, the differences in speculative trading appear to be driven by equity REITs, but not by any particular property type, although we do find that some property types generally have more speculative trading. To better understand the impact of speculative trading on market participants, we also examine the relation between speculative trading during the boom years and returns after the bust. While many have recently criticized the role of speculators in financial markets, we do not 25