REIT Dividend Changes and Stock Price Information Content

Transcription

1 RET Dividend Changes and Stock Price nformation Content Gow-Cheng Huang Department of Accounting, Economics & inance Tuskegee University Tuskegee, AL Kartono Liano Department of inance and Economics Mississippi State University Mississippi State, MS Ming-Shiun Pan Department of inance and Supply Chain Management Shippensburg University Shippensburg, PA

2 RET Dividend Changes and Stock Price nformation Content Abstract This study examines the extent of firm-specific information conveyed in dividend change announcements by real estate investment trusts, RETs. Using stock return to measure the information impounded into stock price from dividend announcements, we decompose stock return into three return components: firmspecific, industry-level, and market-wide components. This decomposition allows us to examine the relative importance of the fundamental components in the stock returns around announcements of dividend changes. We find that the market reaction to RET dividend changes is mainly driven by the firm-specific return component, especially for dividend decreases. We also find that firm-specific stock return variation (a proxy for stock price informativeness) affects a RET s decision to change dividend payments. Moreover, the managerial dividend signal can explain the announcement abnormal return, suggesting that dividend changes contain managerial private information. Keywords: Dividend Changes, RETs, Stock Price nformativeness, Managerial Signal JEL Classification: G14, G35

3 1. ntroduction The finance literature documents significant market reactions to changes in dividends in the same direction. Also, the market reactions to dividend changes are asymmetric. Dividend decreases are associated with a large negative abnormal return, while dividend increases result in a small positive abnormal return. One prominent explanation for this finding is the information-signaling hypothesis (e.g., John and Williams (1985) and Miller and Rock (1985)). The information-signaling model postulates that under the notion of asymmetric information, firms change their dividends to convey private information about future earnings. Specifically, firms have a tendency to smooth dividends. irms are reluctant to cut dividends and will increase dividends gradually only if they anticipate an increase in permanent earnings. The information-signaling model suggests that dividend changes contain firm-specific information. Prior studies (e.g., irth (1996)) also document that dividend changes elicit stock price reactions for other firms in the same industry. A firm s positive (negative) change in dividends induces stock price increases (decreases) among non-announcing firms in the same industry. The findings from these studies suggest that there is an intra-industry information transfer in dividend announcements. Dividend changes appear to convey information not only about the announcing firms but also about other firms in the same industry. When examining the information content of dividend changes, prior research typically measures information content based on abnormal stock returns around dividend announcements. Significant abnormal returns suggest that dividend announcements have information content. n this study, we take a different approach to investigate the extent of firm-specific information conveyed in dividend changes. We consider that a firm s stock price reflects firm-specific, industry-level, and market-level information. Accordingly, we use stock return as a measure of information signal and decompose the stock return into firm-specific, industry-level, and market-wide components. f dividend changes contain mainly firmspecific information, then stock price reactions are expected to be driven mainly by firm-specific factor. However, the presence of a contagion intra-industry effect suggests that the information conveyed in an announcement contains an industry-level component. While stock prices are sensitive to both firm- 1

4 specific and industry news, which of the two factors dominates the market reaction is an empirical question. We analyze the three return components to examine the extent of firm-specific information reflected in stock prices around announcements of dividend changes. We focus on real estate investment trusts (RETs) in this study because the RET industry has several unique attributes that are of interest when studying the information content of dividend changes. RETs are a homogenous group in that they all own real estate properties and/or mortgages. ocusing on a homogenous industry allows us to have a cleaner estimate of industry-wide common factors. ocusing on a homogenous industry also prevents the results from being contaminated by cross-industry effects. Moreover, since RET returns are closely related to private real estate returns (e.g., Boudry, Coulson, Kallberg, and Liu (2012)), investors can easily interpret the information signal revealed in a dividend announcement. n addition, RETs are required to pay out 90% of taxable earnings as dividends to shareholders in order to avoid corporate taxes. This tax provision renders a RET s discretion in setting dividend policy. The high dividend payout requirement also forces RETs to rely more on external capital than industrial firms. Consequently, RETs receive more capital market monitoring than their industrial counterparts. inally, RETs have large institutional ownership. nstitutional investors tend to be better-informed and hence can facilitate the price formation process (Chiang (2010)). n short, compared to industrial firms, RETs are more transparent and have less asymmetric information effect. Consequently, RET dividend announcements might not carry much private information about the announcing RET s operating performance and hence should yield little market reaction to dividend changes. Conversely, prior studies show that the market reacts significantly in the same direction as the RET dividend changes in the announcements (e.g., Wang, Erickson, and Gau (1993), Bradley, Capozza, and Seguin (1998), and Hayunga and Stephens (2009)). Also, the dividend announcement effect is asymmetric and is more pronounced for dividend-decreasing announcements. urthermore, while RETs have less discretion in their dividend payout, most RETs pay out more dividends than their taxable earnings (Kallberg, Liu, and Srinivasan (2003) and Ghosh and Sirmans (2006)). t appears that, due to 2

5 non-cash charges such as depreciation and amortization, cash flow from operations that RETs generate tends to be greater than their taxable earnings. Thus, the empirical literature suggests that RET managers might change dividends to convey private information about their firm s prospects. n the current study, we explore the information content of changes in dividend payments by RETs in detail. We examine a sample of 183 quarterly dividend decreases and 1,143 quarterly dividend increases by RETs over the period Consistent with the literature, we find that RET stock prices in the 3-day window of dividend announcements move in the same direction as dividend changes. We also find asymmetric market reactions to dividend changes. Dividend-decreasing RETs yield a larger abnormal stock return (in absolute term) than that for dividend-increasing RETs. Our stock return decomposition analysis shows that the 3-day market reaction to dividend changes is largely attributable to the firmspecific return component. That is, the magnitude of the firm-specific return component is much larger than that of the industry-level component or the market component. Moreover, the dominant firmspecific return component in the market reaction is more pronounced for the dividend-decreasing RETs. Another issue we examine in this article is whether information in stock prices an important factor in affecting a RET s decision to change dividend payments. Prior studies (e.g., Dow and Gorton (1997) and Subrahmanyam and Titman (1999)) suggest that stock prices may contain useful information about a firm s fundamentals that managers do not possess. Empirical literature (e.g., Chen, Goldstein, and Jiang (2007) and De Cesari and Huang-Meier (2015)) provides evidence showing that managers use the information in stock prices in making corporate decisions. To measure the extent of firm-specific information incorporated into stock prices (i.e., stock price informativeness), we follow Chen et al. (2007) and De Cesari and Huang-Meier (2015) and use firm-specific stock return variation. Our logistic regression analysis shows that firm-specific stock return variation has a negative impact on a RET s decision to change dividends. Specifically, RETs with a higher firm-specific stock return variation (i.e., higher stock price informativeness) are less likely to change dividend payments. 3

6 We further examine whether managerial dividend signal can explain the market s reaction to dividend changes. We extract managerial dividend signal from a probit model that controls for stock price informativeness and other factors. Our results show that the managerial information signal can explain both the abnormal return and the firm-specific abnormal return but not the industry-wide abnormal return, particularly for RETs that cut dividends. n short, our results show that the market reaction to RET dividend changes is mainly affected by firm-specific information, suggesting that RET dividend changes convey private information. The private information in stock prices can explain the likelihood of changing dividends. urthermore, managerial private signal can explain firm-specific return. The rest of the paper is organized as follows. Section 2 contains a brief review of related studies and motivation. Section 3 describes the data and empirical methodologies. Section 4 presents the empirical results. Section 5 concludes the paper. 2. Related Studies and Motivation The literature suggests that dividend announcements have information content. Lintner (1956) shows that firms will raise dividends only if they project a permanent increase in earnings. That is, dividends are sticky, and firms tend to avoid making a change in dividends if it will have to be reversed later. Since dividend changes may reflect managers permanent earnings expectations, dividend increases are interpreted as forecasts of permanently higher earnings and dividend deductions signal lower future earnings. n the presence of information asymmetry between managers and investors, the informationsignal model suggests that management of a firm uses announcements of dividend changes to convey private information (e.g., John and Williams (1985) and Miller and Rock (1985)). The information content of a dividend change announcement causes investors to react to the announcement. Consequently, the market reacts to dividend decrease (increase) announcements negatively (positively). However, the empirical literature shows mixed results on whether dividend changes convey information on a firm s future performance. Aharony and Sway (1980) find that dividend decrease 4

7 (increase) announcements are associated with negative (positive) abnormal stock returns surrounding the announcements. They also find that dividend change announcements convey useful information beyond that from earnings announcements. Yoon and Starks (1995) find that dividend decreases (increases) are associated with significant decreases (increases) in capital spending in the three years following the announcements. They also find that dividend change announcements are related to financial analysts revision of firms current earnings. Yoon and Starks argue that their findings are consistent with the information-signaling hypothesis. n contrast, DeAngelo, DeAngelo, and Skinner (1996) and Benartzi, Michaely, and Thaler (1997) suggest that dividend changes are not reliable signals about future earnings prospects of firms. They find little evidence of a positive relation between dividend changes and future earnings changes. Moreover, they find that many managers even increase their firms dividends when earnings prospects fade. n a survey study, Brav et al. (2005) document that while executives view that dividend policy conveys their confidence about the future, the information conveyance is more about reducing uncertainty in future earnings rather than signaling future earnings. The literature also suggests that the information released in one firm s dividend announcement can be transferred to other firms in the same industry. Prior studies (e.g., irth (1996)) find that dividend announcing firms stock returns are positively related to non-announcing firms returns at the announcement date. Laux, Starks, and Yoon (1998) suggest that the transfer of information in dividend announcements is related to the industry commonality. or industries that share common factors, they tend to exhibit an intra-industry contagion effect. Wang, Erickson, and Gau (1993) examine stock price reaction of RETs to dividend change announcements. Their results show significant positive abnormal returns on the announcement date for dividend increases and negative abnormal returns for dividend decreases. Wang et al. postulate that their results support the hypothesis that dividend changes of RETs convey private information. Bradley, Capozza, and Seguin (1998) find that RET stock prices move in the same direction as dividend changes over a 3-day window surrounding the announcement date. Bradley et al. also find a negative relation between dividend payout and future cash flow uncertainty. They suggest that this finding is consistent 5

8 with the information-signaling hypothesis. Lee, Lin, Chiang, and Kuo (2012) examine the intra-industry effects of RET dividend announcements. Lee et al. document significant contagion effects in RET dividend announcements, particularly for dividend-decreasing announcements. Hardin and Hill (2008) examine the determinants of dividends in excess of the mandatory component required by the regulatory tax status for RETs. They find that excess dividend payments are related to strong operating performance, ability to access short-term bank loan, and reduced agency costs. Unlike Hardin and Hill, Boudry (2011) finds that RETs pay discretionary dividends to smooth dividends instead of to signal operating performance. Boudry attributes the differing results to the different way of measuring discretionary dividends. Similarly, Hayunga and Stephens (2009) find that equity RETs tend to smooth their quarterly dividends. Case, Hardin, and Wu (2012) find that during the liquidity crisis, RETs that cut dividends have insignificant abnormal stock returns around the announcement date, but positive abnormal returns after the announcement. They suggest that their finding implies private information is revealed when RETs cut dividends during the liquidity crisis. f dividend change announcements convey private information, the market reaction to the announcements should primarily reflect firm-specific information. Unlike prior studies that measure the information content of dividend changes based on the market reaction to the announcements, we examine the extent to which firm-specific information is reflected in stock prices around the announcements. To this end, we decompose stock return into three components: firm-specific, industry, and market components. f dividend change announcements contain relevant pricing information to the announcing RETs, we expect that the market reaction is mainly driven by firm-specific information. n contrast, the intra-industry information transfer effect suggests that the announcement returns are mainly due to industry-level information. n addition, we examine whether private information affects dividend change decisions. We measure private information using firm-specific stock return variation, which is calculated from the R-squared statistic of a two-factor regression. Our study will shed light on the information content of RET dividend change announcements. 6

9 3. Data and Empirical Methods 3.1. Data We focus on equity RETs in this study. We identify equity RETs using the SNL database. n order to avoid any misclassifications of real estate companies as RETs, only real estate companies that have elected to be taxed as RETs for corporate income tax purposes are included in the study. The CUSPs from these RETs are used to retrieve quarterly dividend announcements from CRSP. n order to remain in the final sample, daily stock returns for sample RETs must be available from the CRSP database. We also exclude observations with missing values for financial variables used in the study. We end up with 6,693 firm-quarter observations over the period Table 1 reports the sample distribution of dividend changes over time. Dividend change is the change in quarterly cash dividends scaled by the quarterly cash dividend in the previous quarter. The sample contains 183 dividend decreases, 5,367 zero dividend changes, and 1,143 dividend increases. We focus on non-zero dividend changes. Dividend cuts cluster in years 2008 and 2009 during the liquidity crisis. Dividend increases are more common in late 1990 s and also in recent years. The number of dividend increases is greater than that of dividend decreases, except 1991 and Also, most absolute magnitudes of dividend decreases are larger than those of dividend increases. The mean dividend change is for the dividend-decreasing sample and is for the dividend-increasing sample. The findings suggest that RETs are reluctant to cut dividend payments, but when they do, the cuts tend to be large Decomposition of stock return into firm-specific, industry-level, and market-wide components Same as Roll (1988), Morck, Yeung, and Yu (2000), and Liu (2011), we expect that a firm s stock price is affected by firm-specific, industry-level, and market-wide information. f dividend changes 1 n response to the 2008 financial crisis, the RS issued Revenue Procedure in December 2008 that allows RETs to issue elective stock dividends in lieu of cash dividends to satisfy the taxable income distribution. Devos, Spieler, and Tsang (2014) find that only 17 RETs issued elective stock dividends over the period They also find that the RETs that issue elective stock dividends do not face immediate cash flow problems. 7

10 convey management s private information, then the market reaction to the announcement should be mainly caused by the firm-specific information. To examine this proposition, we follow Liu (2011) and decompose stock returns around dividend announcements into market, industry, and firm-specific components. Similar to Liu s framework, we consider that the value of a firm will change when new information becomes available. Specifically, the value of RET j s stock, v j, is influenced by market (M), industry (), and firm-specific ( j) factors as follows: v j = v j + β j M M + β j + j, (1) where v j is the expected value of v j, β M j is the market beta of RET j, β j is the industry beta of RET j, and M,, and j are independent of each other. The market will update the value of RET j s stock when new information from a signal about RET j becomes available (e.g., a dividend announcement). We consider the market reaction to a dividend change announcement reflects the information signal and decompose the market reaction into market, industry, and firm-specific components. Specifically, we run the following regression to estimate the market and industry betas of a RET, β M j and β j : R j,t = α j + β j M R t M + β j (R t β j M R t M ) + e j,t, (2) M where R j,t is the dividend announcing RET j s stock return on day t (t = 0 is the announcement date), R t is the return on the CRSP value-weighted market index on day t, and R t is the return on a non-announcing equity RET industry portfolio on day t. 2 β M j is the market beta of the non-announcing equity RET industry portfolio and is estimated from the following regression: R t = α j + β j M R t M + e t. (3) 2 The non-announcing equity RET industry portfolio is a value-weighted equity RET portfolio that excludes the announcing RET j. The weight of each equity RET is calculated daily. Consequently, the composition of the equity RET industry portfolio is different for different announcing RET j. 8

11 We then decompose the cumulative return from day -1 to day 1, CR j,( 1,1), into a market M component, an industry component, and a firm-specific component. The market component, CR j,( 1,1), is calculated as M CR j,( 1,1) = β M j CR M 1,1. The industry component, CR j,( 1,1), is calculated as CR j,( 1,1) = β j (CR 1,1 β M j CR M 1,1 ), M where CR 1,1 is the cumulative return on the CRSP value-weighted market index from day -1 to day 1; CR 1,1 is the cumulative return on the equity RET industry portfolio from day -1 to day 1; and β M j, β j, and β M j are estimated using the data from day -60 to day -6. The firm-specific component, CR j,( 1,1), is calculated as CR j,( 1,1) = CR j,( 1,1) β M j CR M 1,1 β j (CR 1,1 β M j CR M 1,1 ) Measure of private information in stock prices n the finance empirical literature, firm-specific return variation (idiosyncratic stock return volatility) is commonly used to measure private information in stock prices (e.g., De Cesari and Huang- Meier (2015) and references cited therein). diosyncratic return variation reflects the return variation of a stock that cannot be explained by market and industry returns (Roll (1988) and Morck et al. (2000)). n this study, we measure firm-specific return variation as NO = ln((1 R 2 )/R 2 ), where R 2 is the R 2 from regression (2). We estimate regression (2) over the period from day -60 to day -6 before the dividend announcement, Durnev, Morck, Yeung, and Zarowin (2003) examine the relation between stock return synchronicity (R 2 ) and stock price informativeness. They find that firm-specific return variability is positively associated with the amount of information on future earnings contained in current stock returns. 9

12 Their result suggests that the larger the firm-specific return variation, the higher the information content about future earnings. Durnev, Morck, and Yeung (2004) show that firms with higher firm-specific price variation tend to make corporate investment decisions more efficiently. Piotroski and Roulstone (2004) find that more analyst coverage results in smaller firm-specific return variation. Barberis, Shleifer, and Wurgler (2005) report that inclusion in (deletion from) the S&P 500 index decreases (increases) a stock s firm-specific return variation. Liu (2011) finds a positive relation between firm-specific return volatility and firm-specific return components around analysts recommendation changes. n short, prior research shows that for firms with higher firm-specific price variability, the stock prices track the fundamentals closely Private information in stock prices and dividend changes Prior studies (e.g., Dow and Gorton (1997) and Subrahmanyam and Titman (1999)) suggest that stock prices may contain useful information that managers do not have. Thus, market prices can help managers in making corporate decisions (e.g., Chen et al. (2007) and De Cesari and Huang-Meier (2015)). f managers incorporate the information in stock price in corporate decisions, the amount of private information in stock price should be an important factor in affecting a RET s dividend-change decision. To examine the sensitivity of dividend changes to private information in stock price, we run the following regression as a logistic model: DVDEND CHANGE j = α 0 + α 1 NO j, 1 + α 2 EXO j, 1 + α 3 EXO j, 1 + α 4 MV j, 1 + α 5 LDEBT j, 1 + α 6 SDEBT j, 1 + α 7 MB j, 1 + α 8 ROA j, 1 + α 9 TYPE j, 1 + ψ 5. (4) The dependent variable DVDEND CHANGE is a latent variable that is one for non-zero dividend changes and zero for zero dividend changes. We estimate the logistic models separately for dividenddecreasing and dividend-increasing RETs. NO is a measure of private information in stock price prior to the dividend announcement as described above. f managers of a RET use private information in stock price in their dividend change announcements, the private information variable should be significantly correlated with the dividend change. 10

13 n addition to private information in stock price, the finance literature suggests several factors that affect dividend policy. To control for other possible factors, the regression model includes excess funds from operations (EXO), change in excess funds from operation (ΔEXO), market value of equity (MV), long-term debt (LDEBT), short-term debt (SDEBT), market-to-book (MB), return on assets (ROA), and equity RET property type (TYPE). EXO is funds from operations (O, SNL key field ) minus mandatory dividend payments scaled by total assets (SNL key field ). 3 EXO is included to control for the free cash flow or excess capital hypothesis of Jensen (1986). We expect that RETs with high (low) EXO are more likely to increase (decrease) dividends. ΔEXO is change in excess O. Hardin and Hill (2008) find that change in excess O is a significant determinant of RET excess dividends. MV is the market value of equity of a RET (SNL key field ) and is used to proxy for size. The size of a RET is generally positively related to the diversification of the RET s property holdings. Larger RETs tend to face less volatile operating cash flows and can pay out more dividends. We also include the size of a RET to capture the asymmetric information effect because financial analysts tend to provide less coverage for smaller firms and hence smaller firms are less transparent than larger firms. We use the natural logarithm of the market value in the models because size might have a nonlinear impact on the dividend change decision. LDEBT is non-current long-term debt (SNL key field ) scaled by total assets. We include LDEBT to control for the leverage effect. The high payout requirement forces RETs to raise funds from the capital market more often than industrial firms. Thus, RETs receive more monitoring. Prior studies (e.g., Bradley, Capozza, and Seguin (1998) and Hardin and Hill (2008)) show that leverage is inversely related to dividend payout. SDEBT is short-term and current long-term debt (SNL key field ) scaled by total assets. We include short-term and current long-term debt to control for a RET s liquidity position. RETs with more debt maturing in the short term will face a 3 Mandatory dividends are 95% of before-tax net income prior to 2001 and are 90% thereafter. 11

14 greater liquidity shortfall and hence tend to cut dividends. 4 The free cash flow hypothesis also suggests that firms would distribute cash flow in excess of investment opportunities. We, therefore, include MB, which is the sum of the market value of equity, total debt (SDEBT plus LDEBT), and preferred equity (SNL key field ) divided by total assets, as a proxy for the investment opportunities of a RET. RETs with better investment opportunities tend to pay lower dividends as they retain cash to finance investments. ROA is net income (SNL key field ) scaled by total assets. RETs with better operating performance might retain funds for reinvesting. Thus, we expect a negative relation between ROA and dividend changes. inally, since RETs with different property types might have different cash operating cycles, we include property dummies, TYPE (SNL key field ) to proxy for equity RET types (i.e., hotel, industrial, multifamily, office, retail, storage, etc.). All control variables for firm characteristics are measured at the end of the year prior to the quarterly dividend announcement for computing the DVDEND CHANGE variable. 4. Empirical results 4.1. Stock market reaction and the return components Table 2 reports cumulative stock returns (CR) and abnormal return (AR) around the dividend announcement date. ollowing the literature, we measure the market reaction to dividend changes based on a 3-day window from day -1 to day 1. CR j,( 1,1) is cumulative stock return from day -1 to day 1, defined as CR j,( 1,1) = (1 + R j, 1 )(1 + R j,0 )(1 + R j,1 ) 1, where R j,t is RET j s stock return on day t (t = 0 is the dividend announcement date). We calculate abnormal return as the difference between RET j s cumulative stock return and that of the CRSP value- 4 Devos, Spieler, and Tsang (2014) find that cash flow problems do not affect a RET s decision to issue elective stock dividends. nstead, short-term debt is the primary determinant of paying elective stock dividends. 12

15 weighted market index. That is, AR j,( 1,1) = CR j,( 1,1) CR M M 1,1, where CR 1,1 is the cumulative return over the 3-day window around the dividend announcement for the CRSP value-weighted index. The results show that the market reacts unfavorably (favorably) to dividend decreases (increases). RETs with a dividend cut yield a mean 3-day cumulative return of %, which is significantly less than the 0.250% for the CRSP value-weighted index. The dividend-increasing RETs exhibit a mean 3- day cumulative return of 0.632%, which is significantly greater than the 0.186% for the CRSP valueweighted index. Table 2 also reports the return decomposition results for the 3-day announcement returns. The information-signaling theory suggests that the market response to dividend announcements is caused by firm-specific information. Accordingly, a large proportion of the market reaction should be from the firm-specific return component. The results show that the market reaction to the dividend decrease announcements is mainly due to the firm-specific information. The mean firm-specific return component (i.e., CR j,( 1,1) ) for the dividend-decreasing RETs is %. The firm-specific return component is negative and is even more negative than the raw cumulative return of %. The mean industry return (i.e., CR j,( 1,1) M ) and market return (i.e., CR j,( 1,1) ) components are % and 0.336%, respectively. The firm-specific return component accounts for about 117% of the market reaction to the dividend cuts, while the industry and market return components account for 20% and -37%, respectively. f return is mainly driven by firm-specific information, then abnormal return calculated using the firm-specific return component (i.e., AR j,( 1,1) = CR j,( 1,1) using the industry-level return component (i.e., AR j,( 1,1) M CR j,( 1,1) ) should be significant but not the one calculated = CR j,( 1,1) M CR j,( 1,1) ). or the dividend- decreasing RETs, the mean firm-specific abnormal return, AR j,( 1,1), is %, which is statistically significant at the 5% level. The mean industry-level abnormal return, AR j,( 1,1), is %, which is insignificant. Consistent with the information-signaling hypothesis, our results indicate that the announcement effect for the dividend-decreasing RETs is mainly caused by the firm-specific factor. 13

16 or the dividend-increasing RETs, the return decomposition results are similar to those of the dividend-decreasing RETs. The mean firm-specific return component is 0.465%, which is much larger than the industry return component of 0.039% or the market return component of 0.128%. The proportions of the firm-specific, industry, and market return components are 74%, 6%, and 20%, respectively. The firm-specific abnormal return (AR j,( 1,1) ) is positively significant at 0.279%, while the industry-level abnormal return (AR j,( 1,1) ) is negatively significant at %. Thus, like the dividenddecreasing RETs, the market reaction to the dividend-increase announcements is also mainly due to the firm-specific return component. Overall, our return decomposition results suggest that the announcement effect of RET dividend changes is mostly caused by the firm-specific information, especially for RETs with dividend cuts. This finding is consistent with the dividend information-signaling hypothesis Dividend changes and stock price informativeness Table 3 reports key operating and financial characteristics of the control variables in the year prior to the dividend announcement. Dividend decreases are preceded by less information in stock prices than dividend increases. unds from operations in excess of the mandatory dividend are positive for both dividend-decreasing and dividend-increasing RETs, indicating that both sample RETs generate more funds from operations than the mandatory dividend amount. nterestingly, the dividend-decreasing RETs have a positive change in EXO, while the dividend-increasing RETs exhibit a negative change in EXO. Compared to the dividend-increasing RETs, the dividend-decreasing RETs also appear to be smaller in size, have larger long-term debt, face more investment opportunities, and generate a higher return on assets. Table 4 reports the empirical estimates of logistic regression model (4). We find that the firmspecific stock return variation NO has a statistically significant negative impact on both dividend decreases and dividend increases. RETs are more (less) likely to change dividends to signal managerial 14

17 private information when stock returns convey less (more) firm-specific information. The control variable coefficient estimates show that the likelihood of cutting dividends increases for RETs with small market value and more investment opportunities. n contrast, RETs with large market value, low longterm debt, and low operating profits are more likely to increase dividend Abnormal stock return and managerial dividend signal According to the dividend information-signaling hypothesis, dividend changes contain firmspecific information content. f dividend changes convey managerial private information, there should be significant relations between managerial dividend signal and the market s reaction to changes in dividend payments. To extract the managerial information of dividend changes, we use Heckman s (1979) twostage method. n the first stage, a RET s decision to change dividend is described as a standard probit model as equation (4). Using the probit estimates from equation (4), we then calculate the inverse Mills ratio for dividend-changing RETs that can be seen as a dividend signal not known to the market when a dividend change is made, and use it as a proxy for the managerial dividend signal. 5 To investigate the information content of the managerial dividend signal, we regress abnormal returns on management s private information, measured by the inverse Mills ratio, and dividend changes. The regression model is: AR j,( 1,1) = β 0 + β 1 PRVATE j + β 2 ΔDV j + e j. (5) The dependent variable is abnormal return calculated from raw return (AR j,( 1,1) ), from firm-specific return component (AR j,( 1,1) ), or from industry return component (AR j,( 1,1) ) as reported above. PRVATE j is the inverse Mills ratio and is to proxy for managerial dividend signal. f dividend changes convey managerial private information, we expect a significant and positive coefficient for PRVATE j when the dependent variable is AR j,( 1,1) or AR j,( 1,1). However, when the dependent variable is 5 The inverse Mills ratio is the ratio of the standard normal probability density function over the standard normal cumulative density function using the predicted values from regression (4) as a probit model. The inverse Mills ratio calculated in this way is to approximate the probability that the explanatory variables fail to predict the dividend change decision (i.e., the failure rate) and is used to proxy for the dividend signal. 15

18 AR j,( 1,1), we do not expect the coefficient on PRVATE j to be significant because AR j,( 1,1) contains no firm-relevant information. ΔDV j is change in quarterly dividends scaled by the quarterly dividend in the previous quarter. ΔDV j is included because dividend changes are positively associated with stock returns surrounding dividend change announcement dates. Table 5, Panel A reports the regression results using the abnormal return (AR j,( 1,1) ) as the dependent variable. The results show that the variable PRVATE is significantly positive for the dividenddecreasing RETs, suggesting that unexpected information revealed by the announcement of a dividend decrease can explain stock returns. When the analysis is conducted on firm-specific abnormal return (see Panel B), the results are similar to those from abnormal return. The coefficient on PRVATE remains significantly positive for the dividend-decreasing RETs. As expected, the results in Panel C show that the managerial dividend signal cannot explain the industry abnormal return. This finding, coupled with the results reported in Panels A and B, suggests that the managerial signal from changing dividend payments contains firm-specific but not industry information, which is consistent with the informationsignaling hypothesis. 5. Conclusion The literature documents significant market reactions to dividend changes and abnormal stock returns are positively associated with the direction of dividend changes. One explanation for the positive relation between dividend changes and abnormal stock returns is that dividend announcements convey managerial private information. n the presence of asymmetric information, a dividend increase (decrease) is a positive (negative) signal of future corporate earnings. The information-signal explanation implies that the market reaction to dividend change announcements is mainly driven by firm-specific information. The literature also documents a contagion effect of intra-industry information transfers. News about one firm s dividend announcement transfers to other firms in the same industry. n this study, we examine whether the information contained in RET dividend changes is firm-specific or 16

19 industry-wide. To this end, we propose that stock return reflects informational signal and decompose stock return into the firm-specific, industry, and market return components. Using a sample of 183 quarterly dividend decreases and 1,143 quarterly dividend increases by RETs from 1990 to 2016, we find that the market reacts negatively to dividend decrease announcements and positively to dividend increase announcements. Our results show that the market reaction to dividend change announcements is mainly due to firm-specific return component, especially for dividend cuts. Using firm-specific return variation as a measure of stock price informativeness, we find that the degree of informativeness of stock prices is negatively related to a RET s decision to change dividends. RETs are more likely to change dividends when stock prices are less informative. Moreover, we find that the managerial signal from changing dividends can explain the firm-specific abnormal return but not the industry-wide abnormal return, especially for RETs that cut dividends. Overall, our results provide support for the information-signaling hypothesis in the sense that dividend changes contain firm-specific information. 17

20 References Aharony, J. and. Swary, 1980, Quarterly Dividend and Earnings Announcements and Stockholder s Returns: An Empirical Analysis, Journal of inance 35, Barberis, N., A. Shleifer, and J. Wurgler, 2005, Comovement, Journal of inancial Economics 75, Benartzi, S., R. Michaely, and R. Thaler, 1997, Do Changes in Dividends Signal the uture or the Past? Journal of inance 52, Boudry, W., 2011, An Examination of RET Dividend Payout Policy, Real Estate Economics 39, Boudry, W., N. Coulson, J. Kallberg, C. Liu, 2012, On the Hybrid Nature of RETs, Journal of Real Estate inance and Economics 44, Bradley, M., D. Capozza, and P. J. Seguin, 1998, Dividend Policy and Cash-low Uncertainty, Real Estate Economics 26, Brav, A., J. Graham, C. Harvey, and R. Michaely, 2005, Payout Policy in the 21st Century, Journal of inancial Economics 77, Case, B., W. Hardin, and Z. Wu, 2012, RET Dividend Policies and Dividend Announcement Effects during the Liquidity Crisis, Real Estate Economics 40, Chen, Q.,. Goldstein, and W. Jiang, 2007, Price nformativeness and nvestment Sensitivity to Stock Price, Review of inancial Studies 20, Chiang, K. C., 2010, On the Comovement of RET Prices, Journal of Real Estate Research 32, DeAngelo, H., L. DeAngelo, and D. Skinner, 1996, Reversal of ortune: Dividend Signaling and the Disappearance of Sustained Earnings Growth, Journal of inancial Economics 40, De Cesari, A. and W. Huang-Meier, 2015, Dividend Changes and Stock Price nformativeness, Journal of Corporate inance 35, Devos, E., A. Spieler, D. Tsang, 2014, Elective Stock Dividends and RETs: Evidence from the inancial Crisis, Real Estate Economics 42,

21 Dow, J. and G. Gorton, 1997, Stock Market Efficiency and Economic Efficiency: s There a Connection? Journal of inance 52, Durnev, A., R. Morck, and B. Yeung, 2004, Value-Enhancing Capital Budgeting and irm-specific Stock Return Variation, Journal of inance 59, Durnev, A., R. Morck, B. Yeung, and P. Zarowin, 2003, Does Greater irm-specific Return Variation Mean More or Less nformed Stock Pricing? Journal of Accounting Research 41, irth, M., 1996, Dividend Changes, Abnormal Returns, and ntra-ndustry irm Valuations, Journal of inancial and Quantitative Analysis 31, Ghosh, C. and C.. Sirmans, 2006, Do Managerial Motives mpact Dividend Decisions in RETs? Journal of Real Estate inance and Economics 32, Hardin, W. and M. Hill, 2008, RET Dividend Determinants: Excess Dividends and Capital Markets, Real Estate Economics 36, Hayunga, D. and C. Stephens, 2009, Dividend Behaviour of US Equity RETs, Journal of Property Research 26, Heckman, J., 1979, Sample Selection Bias as a Specification Error, Econometrica 47, Jensen, M., 1986, Agency Costs of ree Cash low, Corporate inance, and Takeovers, American Economic Review 76, John, K. and J. Williams, 1985, Dividends, dilutions, and Taxes: A Signaling Equilibrium, Journal of inance 40, Kallberg, J., C. Liu, and A. Srinivasan, 2003, Dividend Pricing Models and RETs, Real Estate Economics 31, Laux, P., L. Starks, and P. Yoon, 1998, The Relative mportance of Competition and Contagion in intra- ndustry nformation Transfers: An nvestigation of Dividend Announcements, inancial Management 27, Lee, M.-L., C.-W. Lin, K. Chiang, S.-H. Kuo, 2012, The ntra-ndustry Effects of RET Dividend Announcements, Pacific Rim Property Research Journal 18,

22 Lintner, J., 1956, Distributions of ncomes of Corporations among Dividends, Retained Earnings, and Taxes, American Economic Review 46, Liu, M., 2011, Analysts ncentives to Produce ndustry-level versus irm-specific nformation, Journal of inancial and Quantitative Analysis 46, Miller, M. and K. Rock, 1985, Dividend Policy under Asymmetric nformation, Journal of inance 40, Morck, R., B. Yeung, and W. Yu, 2000, The nformation Content of Stock Markets: Why Do Emerging Markets Have Synchronous Stock Price Movements? Journal of inancial Economics 58, Piotroski, J. and D. Roulstone, 2004, The nfluence of Analysts, nstitutional nvestors, and nsiders on the ncorporation of Market, ndustry, and irm-specific nformation into Stock Prices, Accounting Review 79, Roll, R., 1988, R 2, Journal of inance 43, Subrahmanyam, A. and S. Titman, 1999, The Going-Public Decision and the Development of inancial Markets, Journal of inance 54, Wang, K., J. Erickson, and G. W. Gau, 1993, Dividend Policies and Dividend Announcement Effects for Real Estate nvestment Trusts, Real Estate Economics 21, Yoon, P. and L. Starks, 1995, Signaling, nvestment Opportunities, and Dividend Announcements, Review of inancial Studies 8,

23 Table 1 This table reports the frequency of RET dividend changes over the period from 1990 to Dividend change is the change in quarterly dividends scaled by the quarterly dividend in the previous quarter. Dividend Decreases Dividend ncreases No. of Announcements Dividend Change No. of Announcements Dividend Change Total ,

24 Table 2 This table reports descriptive statistics of stock returns and the three return components around dividend change M announcements. CR j,( 1,1) is the cumulative return from day -1 to day 1. CR 1,1 is the cumulative return from day - 1 to day 1 for the CRSP value-weighted index. AR j,( 1,1) is the abnormal return of a RET over the CRSP valueweighted index from day -1 to day 1. Day 0 is the dividend announcement date. CR j,( 1,1), CR j,( 1,1), and CR j,( 1,1) M are the firm-specific, industry-wide, and market components of CR j,( 1,1), respectively. AR j,( 1,1) is firm-specific abnormal return from day 1 to day 1, calculated as the difference between the firm-specific component (CR j,( 1,1) ) and the CRSP value-weighted market index. AR j,( 1,1) is industry-level abnormal return from day 1 to day 1, calculated as the difference between the industry component (CR j,( 1,1) ) and the CRSP value-weighted market index. All returns are in percentages. ***, **, and * denote significantly different from zero using a t-test at the 1%, 5%, and 10% levels, respectively. Dividend Decreases Dividend ncreases Mean Min Median Max Mean Min Median Max CR j,( 1,1) M CR 1, AR j,( 1,1) ** *** CR j,( 1,1) CR j,( 1,1) M CR j,( 1,1) AR j,( 1,1) ** *** ** AR j,( 1,1) 22

25 Table 3 This table reports descriptive statistics of financial data used in the regression models. NO = ln((1 R 2 )/R 2 ), where R 2 is from a regression of the daily stock return on daily market and industry returns. EXO is excess funds from operation in the previous quarter scaled by total assets (SNL key field ). Excess funds from operation is funds from operation (O, key filed ) minus the mandatory dividend payment. ΔEXO is EXO in the current year minus EXO in the previous year, scaled by total assets. MV (in billion) is the market value of equity (key field ). LDEBT is long-term debt (key field ) scaled by total assets. SDEBT is short-term debt (key field ) scaled by total assets. MB is the sum of the market value of equity, total debt (LDEBT + SDEBT), and preferred stocks (key field ) scaled by total assets. ROA is net income available to common shareholders (key field ) divided by total assets. Dividend Decreases Dividend ncreases Mean Min Median Max Mean Min Median Max NO EXO ΔEXO MV LDEBT (%) SDEBT (%) MB ROA (%)

26 Table 4 This table reports the logistic regression coefficients from regressing dividend change on private information in stock price (NO) and various control variables. The dependent variable is one for non-zero dividend change and zero for zero dividend change. NO = ln((1 R 2 )/R 2 ), where R 2 is from a regression of the daily stock return on daily market and industry returns. EXO is excess funds from operation in the previous quarter scaled by total assets. Excess funds from operation is funds from operation (O) minus the mandatory dividend payment. ΔEXO is EXO in the current year minus EXO in the previous year, scaled by total assets. MV is the market value of equity. LDEBT is long-term debt scaled by total assets. SDEBT is short-term debt scaled by total assets. MB is the sum of the market value of equity, total debt, and preferred stocks scaled by total assets. ROA is net income available to common shareholders divided by total assets. TYPE is a RET s property sector type. nside the parentheses are t-statistics, based on heteroscedasticity consistent standard errors. ***, **, and * denote statistically significance at the 1%, 5%, and 10% levels, respectively. Dividend Decreases Dividend ncreases ntercept ** (-0.37) (-2.13) NO *** *** (-3.75) (-3.34) EXO (-0.74) (-0.07) ΔEXO (0.94) (-0.89) MV *** * (-2.90) (1.75) LDEBT * (1.36) (-1.65) SBEDT (-0.96) (-0.59) MB ** (2.37) (-0.63) ROA ** (0.40) (-2.59) TYPE (-0.67) (-0.41) Pseudo R

27 Table 5 This table reports the OLS regression coefficients from regressing the 3-day abnormal stock return around the dividend announcement date on unexpected information revealed by dividend announcements (PRVATE) and dividend change (ΔDV). Abnormal return is the difference between a RET s return and the return of the CRSP value-weighted index. irm-specific abnormal return is the difference between the firm-specific return component and the return of the CRSP value-weighted index. ndustry-level abnormal return is the difference between the industry return component and the return of the CRSP value-weighted index. PRVATE is the inverse Mills ratio that is used to proxy for unexpected information revealed by a dividend announcement. The inverse mills ratio is calculated using parameters estimated from regression (4) as a probit model. ΔDV is the change in quarterly dividends scaled by the quarterly dividend in the previous quarter. nside the parentheses are t-statistics, based on heteroscedasticity consistent standard errors. ***, **, and * denote statistically significance at the 1%, 5%, and 10% levels, respectively. Dividend Decreases Dividend ncreases Panel A: Dependent variable is abnormal return (AR j,(1,1) ) ntercept ** (-2.20) (-0.58) PRVATE ** (2.13) (1.08) ΔDV *** (0.78) (3.52) Adjusted R Panel B: Dependent variable is firm-specific abnormal return (AR j,( 1,1) ) ntercept * (-1.75) (-0.91) PRVATE * (1.72) (1.17) ΔDV * (1.31) (1.67) Adjusted R Panel C: Dependent variable is industry-level abnormal return (AR j,( 1,1) ) ntercept (-0.59) (0.09) PRVATE (0.55) (-0.38) ΔDV ** (0.42) (2.38) Adjusted R