The Information Content of Analysts Value Estimates. Ryan G. Chacon. Dan W. French. Kuntara Pukthanthong. University of Missouri

Transcription

1 The Information Content of Analysts Value Estimates Ryan G. Chacon Dan W. French Kuntara Pukthanthong University of Missouri Contact author: Dan French Department of Finance Trulaske College of Business University of Missouri Columbia, MO USA

2 ABSTRACT The Information Content of Analysts Value Estimates This study examines a large sample of REIT abnormal returns as an indicator of reaction to information around changes in analysts consensus REIT NAV estimate changes. Results show that there is a significant and positive relation between REIT abnormal return on the NAV change announcement day and the magnitude of the NAV change. This implies that there is significant new information impounded in analysts forecasts and value estimates. In addition, consistent with prior work on earnings forecasts, the price reaction appears to continue to develop over a number of days following the initial announcement.

3 The Information Content of Analysts Value Estimates 1. Introduction The information content of forecasts and estimates provided by professional investment analysts has been the topic of significant academic research. Prior research, and a substantial amount of it exists, concentrates on the adjustment of stock prices to the release of each of two types of forecast: (1) analysts forecasts of earnings and (2) analysts buy/sell recommendations. The goal of our research is to examine the information content of a third type of analyst estimate, firm value, estimates for which there has been a lack of research to date on any information that they may convey to the market. Value estimates are necessarily based on forecasts of earnings and other variables. An analysis of the information content of analysts value estimates would therefore be useful in measuring analysts ability to use those forecasts to produce useful value estimates. The purpose of this paper therefore is to explore the information impounded in analysts value estimates by analyzing the market reaction to revisions of analyst consensus estimates of firm value. When analyzing a stock for purposes of suggesting the purchase or sale of a stock, analysts typically other use not only earnings forecasts but a variety of methods to reach a conclusion regarding the recommendation. While an analyst may base a recommendation solely on a simple variable such as a forecasted change in the company s earnings, one of the most widely used methods is to estimate a fundamental value for the firm using forecasted cash flows (earnings and/or dividends). The analyst then compares the fundamental value to the market price and recommends that clients buy stocks that are sufficiently below their fundamental value (undervalued stocks) and sell those that have achieved or trade above their value (overvalued stocks). As Schipper (1991, p. 113) notes, analysts investment recommendations result from a decision process, of which forecasting earnings is just one component.

4 2 However, analysts seldom reveal their estimate of firm value. Instead they issue their recommendations as action categories such as strong buy, buy, hold, sell, and strong sell. Sometimes analysts will issue a target future price range for the stock, but this is not the norm nor does the target price range necessarily represent the analysts point estimate of the present value of the stock. Given this lack of information about analysts specific stock value estimates, there has been a notable absence of research on the information content of analysts value estimates. From an analyst s perspective, value estimates provide much more of a challenge than earnings forecasts or buy sell recommendations. To estimate value, an analyst needs to forecast not only the nearterm cash flows but also cash flows for a number of periods in the future. In addition, the analyst needs to forecast the appropriate rate at which to discount those future cash flows. Value estimates, therefore, have the potential to reveal more information about analysts financial forecasting ability than do other types of forecasts such as earnings, dividends, and stock recommendations. Specific value estimates are widely available for one special type of firm, real estate investment trusts (REITs). These valuations, provided by analysts and called the net asset value (NAV), estimate the value of the REITs underlying asset portfolio net of liabilities. While the relationship varies by individual REIT and over time, REIT prices on average tend to approximate NAVs and therefore represent good estimates of the market value of REITs shares (for example, Capozza and Lee (1995), Patel, Pereira, and Zavodov (2009), and French and Price (2016)). This paper examines changes in the market prices of REITs around days on which analysts changed their estimate of the REIT s NAV. Although restricting our sample to REITs has its drawbacks, these are the only consistently available point estimates of firm value over time that

5 3 are available. In defense of the REIT sample, analysts, in estimating their NAVs, undertake cash flow forecasting and discounting methods that that closely follow the valuation process of analysts in any industry. 2. Related literature 2.1 Earnings forecasts and stock recommendations A large body of financial research exists that investigates the efficacy of forecasts by financial analysts. Beginning in the 1970s, financial research investigating the information content of analysts forecasts of company earnings began to appear, and the amount of evidence that has accumulated is substantial. 1 In general, studies find that there is new information in analysts forecasts and the market reacts accordingly to bring stock prices to reflect the forecasts. However, most studies show that the time that the market takes to fully reflect the new forecast information can often be sufficiently long to allow market participants to profit. In other words, there appears to be a drift to the adjustment of post-forecast-revision prices. Gleason and Lee (2003) examine the factors contributing to this post-revision drift and reach several conclusions. First, they find that the market does not distinguish between forecast revisions that imply new differential information about the stock and those that simply move towards the consensus. They also conclude that the price adjustment time period tends to be shorter with forecasts issued by celebrity analysts and forecasts issued on stocks with a larger analyst following. Finally, they show that there is a tendency for a significant portion of the drift 1 See, for example, work by Griffin (1975) who examines stock returns around analysts earnings forecasts; Givoly and Lakonishok (1979) who find evidence of information impounded in analysts earnings forecast revisions; Elton, Gruber, and Gultekin (1981) showing that stock prices quickly reflect information in analyst forecast estimates; Imhoff and Lobo (1984) who show that positive returns accompany upward earnings revisions by analysts and negative returns follow downward revisions; Cornell and Landsman (1989) examining earnings forecast revisions and forecast errors, Stickel (1991); Peterson and Peterson (1998) demonstrating evidence of information impounded in earnings forecasts related to Value Line stock ranking changes; and Beaver, Cornell, Landsman, and Stubben (2008) who identify significant information content of the error component of earnings forecasts.

6 4 to occur on dates that coincide with subsequent forecast revisions by other analysts and actual earnings announcements. This implies that the market waits for supporting evidence from other analysts or for corroborating evidence from the firm s financial statements before completing the price adjustment process. A second line of research that followed the findings of new information implied in earnings forecasts examines the information that analysts buy/sell recommendations might convey to the market. 2 These studies show a post-recommendation pattern that is similar to that of the earnings forecasts. Abnormal stock returns tend to follow analysts recommendations. Furthermore, the time over which the price adjustments occur is sufficiently long to allow investors to realize abnormal returns based on the recommendations. 2.2 REIT prices and net asset values A REIT s NAV is NAV MVA MVL, (1) where MVA is the REITs market value of assets and MVL is the market value of liabilities. While REITs may compute their NAVs for internal purposes, they no not disseminate that information publicly. Because the MVA is not a value observable from transactions, computation of NAV is by analysts who estimate of MVA by forecasting the net cash flows that a REIT s property portfolio should generate and adjusting that number by other cash flows such as general expenses attributable to the REIT and net income from a taxable REIT subsidiary (if applicable), but not interest expense, and discounting to present value at an estimate of the appropriate rate (the cap 2 See, for example, studies by Holloway (1981) finding positive returns on a portfolio of Value Line Rank-1 stocks; Elton, Gruber, and Grossman (1986) showing that stock prices changes respond to recommendations by brokerage firm analysts in the expected direction; Womack (1996) who shows that large price adjustments follow extreme changes in recommendations by analysts; Jegadeesh, Kim, Krische, and Lee, C. (2004) who examine conditions under which analyst recommendations are more predictive of returns; and Howe, Unlu, and Yan (2009) concluding that abnormal returns follow analyst recommendations even after controlling for macroeconomic variables.

7 5 rate in real estate terminology). MVL is also an estimate but usually approximated very well with the book value of liabilities adjusted for market value of debt disclosed in the financial statement footnotes. REIT analysts therefore have to forecast the two principal inputs of the NAV, MVA and the discount rate. In estimating the discount rate, analysts consider observed market transactions and quotes of properties similar to those in the REIT s portfolio (by property type, geographic location, property quality, etc.). In doing so, their estimates incorporate market-wide factors affecting the entire real estate market. If price discovery takes place in the REIT market so that REIT price changes anticipate changes transaction prices in the physical real estate market, then REIT prices changes should lead changes in analysts NAV estimates. Applied research verifies this anticipatory relationship. NAVs (physical real estate) not only closely match REIT prices (as mentioned in the previous section) and correlate closely with REITs returns, 3 but NAV changes tend to lag changes in REIT prices. Yavas and Yildirim (2009) show that REIT price changes Granger-cause changes in NAV estimates. Yavas and Yildirim s conclusion that the REIT market anticipates analyst NAV changes implies at first glance that there might be no information implied in analyst NAV estimates. This however would imply that analysts are unable to impart new information via either their cash flow forecasting process for specific REITs or through their ability to forecast property-explicit discount rates to individual REIT property portfolios. 3 For example, Toluca, Myer, and Webb (2000) show that indices of both REITs and physical real estate are cointegrated; Liow (2003) studies NAVs and REIT prices and concludes that they tend not to drift apart; Liow and Lee (2006) find a long-run cointegrating relationship between publicly traded and physical real estate in eight Asia-Pacific markets; Hoesli and Oikarinen (2012) conclude that REITs and direct real estate investment are good substitutes in the long run.

8 6 3. Sample SNL Financial provides daily NAV estimates that equal the mean of all new NAV estimates for a REIT issued by analysts on each day. On days when analysts issue no new NAV estimates for a REIT, the database retains the previous day s consensus estimate so that changes in the NAV signify a new consensus estimate. The measure of NAV information flow is the day s percent change in the NAV ( NAVt), NAV t NAVt (2) NAVt 1 where NAV is the REIT s NAV and t and t-1 signify the current day and the preceding day respectively. NAVt is zero for all days on which no analyst issues a NAV revision or estimate. Analysts can issue a NAV or a revision in their previous NAV estimate at any time, so for any NAV t the prior NAV revision could have occurred on any day before day t. For example, suppose that there a new NAV consensus estimate arrives on day 0 and the nearest change occurred 8 days before that. We compute the abnormal return (ARt) for each REIT as P I, (3) t t ARt P t 1 I t 1 where P is the closing price of the REIT, I is the value of a market index, and t and t-1 represent observations for the current day and the prior day respectively. We calculate three different varieties of ARt, one for each of the following market indexes: the CRSP Value Weighted index, CRSP/Ziman REIT Data Series Value Weighted index, and the CRSP/Ziman REIT Equal Weighted Index.

9 7 The SNL database also provides all financial statement data for individual REITs. In addition, We identify and record dates on which REITs filed either a 10k, 10Q, or 8K report with the SEC. Our sample period spans the 2001 to 2015 period is the earliest year for SNL contains consistent daily NAV analyst estimates. We include in our sample all equity REITs within the SNL database with non-missing NAV and financial data. We exclude from the sample all REITs with a negative NAV (NAV < MVA MVL) and additionally all REITs that cannot liquidate and pay the full NAV to shareholders (NAV < MVA MVL accumulated depreciation, see French and Price (2016)). The final sample contains 235 publicly traded equity REITs representing all real estate subsectors yielding a total number of firm-day observations of 392,529. Table 1 summarizes the sample by REIT sector. Table 2 contains a summary of the sample descriptive statistics. Panel A provides statistics for the key variables of interest and Panel B displays basic firm characteristics. Of special note is the mean number of days between a NAV revision of days. This means that on average a REIT s NAV receives a revision by analysts about every 15 days. However, revisions tend to be grouped together, so the typical pattern may see a REIT going 45 days with no revision and then a 10-day period in which 6 revisions occur. There are a mean of 8.15 analysts (median of 7.00) covering the average REIT in the sample. For days on which revisions occur, the mean NAV is 0.29% and the mean absolute value of the NAV revision is 1.66%. This suggests that while the overall average NAV adjustment is close to zero, analysts tend to revise NAV estimates up slightly more frequently than down. Analyst estimate revisions tend to occur more frequently in clusters around major information releases such as the filing of a 10-Q or 10-K report. The frequency of an analyst revision on all trading days within the sample is 9.36%. While the Price/NAV premium is not the

10 8 main focus of this study, it is worth noting that during the sample the premium centers very close to 1 (0.9957). Panel C provides more detail regarding the Price/NAV premium by year and shows the significant variation that is well documented in prior literature. Panel C describes the variables of interest by year. The number of observations, analyst following, and analyst revisions increase almost monotonically by year. This is due to SNL increasing its coverage from 2001 to 2015 as well as the growth of the REIT industry as a whole. The average absolute value NAV decreases by year. As the number of analysts following REITs increases, revisions occur more frequently and thus in lower increments. For example, the percentage change in NAV estimate is likely to be much larger for a REIT that is covered by one analyst who updates his revisions quarterly than it is for a REIT that is covered by twenty different analysts. The average Price/NAV premium is equal to for the entire sample period but exhibits considerable variation by year. As would be expected, REITs trade at the largest discount to NAV in the crisis years from 2007 to There are two primary obstacles in executing an examination of the information content around NAV estimate revisions. Some NAV revisions occur in clusters on or near major information releases. This is especially notable around 10-K and 10-Q releases. Analyst revisions likely reflect the information being released by the REIT. Therefore, it is difficult to disentangle whether returns reflect analyst estimates of NAV or they simply reflect new information to the market through the REIT s reports. The second obstacle is the clustering of analyst estimates several days in a row. For example, if 3 analysts are following REIT i and they submit NAV estimates in three consecutive days, this results in a consensus NAV estimate revision for 3 consecutive days. The most direct and clean approach to mitigate these drawbacks is to simply identify NAV estimate revisions which do not occur in clusters or near filing dates. This approach has costs and

11 9 benefits. The obvious cost is the lack of power in our tests by dropping observations. The benefit is that the observations we do observe are most likely to reflect a clean examination of our research question. In what we call our isolated sample, we drop observations whose NAV estimate revisions that are within a [12,+12] trading day window of 10-K and 10-Q filing date. We also restrict our sample to NAV estimate revisions which are isolated to a 5 day window (meaning we only examine NAV estimate revisions with no other revisions from [-2,+2] trading days). This subsample results in a total of 12,122 NAV estimate revisions. These revisions exhibit qualitatively similar characteristics to the full sample. The absolute value of the average ΔNAV is 1.78% (comparable to 1.66% for the full sample). The annual average patterns are consistent to the full sample as well. The number of revisions increases almost monotonically by year and the average absolute value of the ΔNAV decreases almost monotonically by year. Overall, the subsample appears qualitatively similar to the full sample, alleviating concerns of a biased subsample. 4. Empirical Methods and Results: 4.1 Average CAR around ΔNAV In our first test, we examine the average short window CARs around NAV revisions in our isolated sample. If analyst revisions have information content, we would expect to see CARs that are significantly different from zero around NAV revisions. However, because NAV revisions can be both positive and negative, a simple average of all revisions would result in a canceling out effect. Therefore, we adjust our test in two different ways. First, we split the sample into quintiles sorted on ΔNAV values. The first quintile contains the largest negative ΔNAV values and the fifth quartile contains the largest positive ΔNAV values.

12 10 We then calculate the average CAR for 5 and 3 day windows around each NAV revision use both the CRSP Ziman REIT Value Weighted (REIT_VW) and CRSP Ziman REIT Equal Weighted (REIT_EW) indices as the baseline index. We expect to find negative CARs in the lower quartiles (Q1 and Q2) and positive CARs in the higher quartiles (Q4 and Q5). Q3 will contain a mix of positive and negative revisions, so we expect a CAR close to zero. Panel A of Table III presents results for this test. For both the 5 day and 3 day windows, and across both indices, results are consistent with information content in NAV revisions. CARs increase monotonically from the first quintile to the fifth. The average CAR for the first (fifth) quintile is negative (positive) and statistically significant at the 1% level for all variations. In economic terms, the first (fifth) quintile average CAR for the 5 day window is -21 (35) basis points. The effect appears to be somewhat stronger for the 5 day window than the 3 day window, which implies day -2 and +2 together are relevant. These results are consistent with significant information content of NAV revisions. To learn more about each day within the 5 and 3 day widows, we examine mean ARs by day. To avoid the canceling out effect, we multiply CARs for downward revisions by (-1). This treatment allows us to test all revisions within the Isolated sample in the same test. If the revision is driving the CARs, we would expect to see the strongest results on day 0. The results are tabulated in Panel B of Table III. The full window CARs are consistently positive and statistically significant at the 1% level across all variations. The average CAR for the 5 day window using the REIT_VW index as the base index is 17 basis points. Interestingly, all days appear to have significant CARs except for day +2 in the 5 day window analysis. It appears the market tends to impound all of the information during the day -2 to day 1 window. As expected, the strongest result is on day 0.

13 11 This is true both in terms of magnitude and statistical significance. Overall, results from Panel A and Panel B from Table III are consistent with information content of analyst revisions of NAV. 4.2 Regression analysis: ΔNAV on CAR for isolated sample The previous test allows us to identify a general relationship between ΔNAV and CAR. In our next test, we utilize univariate and multivariate regressions to quantify a more direct relationship between ΔNAV and CAR. Using the same Isolated sample used in Table III, we estimate the following model: CAR a a NAV a MC a VOL a Analysts b Year c Firm e t 0 1 t 2 t 3 t 4 t i t i t t i 1 i 1 n n where CARt is equal to the 5 and 3 day cumulative market adjusted return for a REIT day t. The vector of controls consists market capitalization for REIT i on day t (MC), shares traded for REIT i on day t (VOL), and the number of analysts following REIT i on day t. Year fixed effects and firm fixed effects are used to control for any potential time or firm related omitted variables that may affect the relationship between ΔNAV and CAR. If there is information content in NAV analyst revisions, we would expect to find a positive and statistically significant coefficient on ΔNAV in all specifications. The predictions of the control variables are not as clear, so we do not make them a priori. The results of this regression are contained in Table IV. Columns (1) to (4) present results for the regressions with the 5 day CAR and columns (5) to (8) present results using the 3 day CAR. All CARs in Table IV are calculated using the REIT_VW index as the baseline index. The coefficient on ΔNAV is positive and statistically significant at the 1% level for all specifications. These regressions were run using the REIT_EW index and results were similar (specifically, the coefficient on ΔNAV was similar in magnitude and significant at 1% level across all specifications). Results for columns (3), (4), (7), and (8)

14 12 include both year and firm fixed effects. Untabulated results are robust to using either firm or year fixed effects individually. Results are virtually unchanged when including the vector of controls or fixed effects. The coefficient on ΔNAV ranges from to for the regressions on 5 day CARs and to for the regressions on 3 day CARs. None of the vector of controls appear to have a significant effect on CAR, all else equal. In terms of economic magnitude, a 1% increase in the consensus analyst NAV estimate corresponds to a % increase in 5 day CAR. Results from the regression analysis is consistent with the market responding to information contained in analyst NAV revisions. 4.3 Regression Analysis: ΔNAV on CAR for Clustered Sample In previous tests, we have used the Isolated sample to examine the information content of analyst NAV revisions. Next, we use a new model specification to test for information content to account for the clustering of revisions on days close together. The model is defined as follows: w 0, w 0 j j t j t 2 t 3 t 4 t j 0 j 0 n n i t i t t i 1 i 1 w CAR a b D NAV c D NAV a MC a VOL a Analysts b Year c Firm e also going to need the individual part of equasion in next paragraph as we define it. Where CAR0-N is the CAR for days 0 to +n (n takes the values of 2, 5, and 8 in our test). changenav0 is equal to the first NAV revision in n previous daysfor the [0,+2] test, the first revision is defined as any NAV revision that occurs for REIT i when there are no other revisions for REIT i during the previous 2 days. For the [0,+5] days, the first revision one in which there are no other revisions during the previous 5 days, and for the [0,+8], there are no other revisions during the previous 8 days. The number of observations decreases with the size of the window. As with the Isolated sample, all revisions within [ ] of a 10-K or 10-Q filing

15 13 date are dropped. ChangeNAVt is equal to the percentage change in consensus NAV estimate if there is a revision on day t. If there is no revision on day t, then ChangeNAVt is equal to 0. The SUM(Dt*changeeAR) variable defined as the summation of the interaction of a vector of indicator variables ( Dt ) and a vector of abnormal returns ( ARt ). In Table V, this variable is labeled SumAR. The indicator variable Dt is equal to 1 when ChangeNAVt is equal to 0, and is equal to 0 otherwise. The vector of controls and fixed effects are the same as defined in Table IV. This model is best explained using a rather in depth example using the [0,+5] sample. Imagine REIT i has no NAV revisions for some number of days greater than 5, then has a revision on two days within the [0,+5] day window (day 0 and day 4), and no revisions on days 1, 2, 3, and 5. ChangeNAV0 and ChangeNAV4 will be equal to the percentage changes in the NAV estimate on days 0 and 4, respectively. The ChangeNAV variable for days 1,2,3 and 5 will all equal 0. Additionally, the abnormal return for days 1,2,3 and 5 will be summed and added to the right side of the regression. Day 0 and day 4 abnormal returns will be omitted from the right side of the regression. The model will then estimate the effect of the NAV revisions on days 0 and 4 on the CAR of days 0 and 4. This model is similar to regressing each revision in the sample on day 0 abnormal return. However, the unique advantage is we can test to see if the relationship between ΔNAV and CAR is different for clusters of revisions vs isolated revisions. We expect bt to be different from 0 if the NAV revisions on each day contain incremental information content. More precisely, we expect bt to be greater than 0 if the market reacts in the same direction as the revision. We expect co to be close to 1 since it regresses the sum of abnormal returns on non-revision days on the CAR that includes these days. Similar to table IV, we make no predictions regarding the vector of control variables.

16 14 Results for this model are presented in Table V.As predicted, all ChangeNAV0 coefficients load positive and statistically significant at the 1% level for all three event windows. Additionally, all of the coefficients on ChangeNAVt are positive and almost all of them are statistically significant. This suggests that each revision within a cluster of revisions contain incremental information content to the market. We find an interesting result when we ask whether the relationship between ΔNAV and CAR is different for clusters of revisions than with isolated ones. While almost all coefficients are positive and statistically significant, the coefficients are notably larger for subsequent NAV revisions than the first NAV revision. This is especially true when we compare NAV0 with NAV1 The SumAR coefficient is consistently close to 1 and highly statistically significant. The vector of controls continue to be insignificant with the exception of volume for the 5 and 8 day tests. Though volume loads statistically significant, the coefficient is very small and likely economically insignificant. The regressions tabulated all contain both year and firm fixed effects. Tests were run without fixed effects and with year and firm fixed effects individually and all tests produce qualitatively similar results. The results shown are using the REIT_VW index as the baseline index for the CAR calculations. Untabulated analysis shows the results are robust to using the REIT_EW index for the CAR calculations. The R-Squared values are notably high which is a result of model construction. The greatest impact on R-Squared is coming from the SumAR variable which regresses returns of some but not all days included in the summation of the returns in the dependent variable. Results from Tables III, IV, and V are all consistent with analyst NAV revisions containing information content. 5. Conclusion

17 15 This study has examined a large sample of REIT abnormal returns as an indicator of reaction to information around changes in analysts consensus REIT NAV estimate changes. Results show that there is a significant and positive relation between REIT abnormal return on the NAV change announcement day and the magnitude of the NAV change. This implies that there is significant new information impounded in analysts forecasts and value estimates. In addition, consistent with prior work on earnings forecasts, the price reaction appears to continue to develop over a number of days following the initial announcement.

18 16 References Baik, B., Billings, B., Morton, R Reliability and transparency of non-gaap disclosures by real estate investment trusts (REITs). Accounting Review 83, Beaver, W., Cornell, B., Landsman, W.R., Stubben, S.R., The impact of analysts' forecast errors and forecast revisions on stock prices. Journal of Business Finance and Accounting 35, Capozza, D. and Lee, S., Property type, size, and REIT value. Journal of Real Estate Research 10, Cornell, B., Landsman, W., Security price response to quarterly earnings announcements and analysts' forecast revisions, Accounting Review 64, Elton, E., Gruber, M., Grossman, S., Discrete expectational data and portfolio performance. Journal of Finance 41, Elton, E., Gruber, M., Gultekin, M., Expectations and share prices. Management Science 27, Givoly, D., Lakonishok, J., The information content of financial analysts forecasts of earnings: Some evidence on semi-strong inefficiency. Journal of Accounting and Economics 1, Gleason, C. and C. Lee (2003). Analyst Forecast Revision and Market Price Discovery. The Accounting Review 78, Gore, R., Stott, D., Toward a more informative measure of operating performance in the REIT industry: Net income vs. funds from operations. Accounting Horizons 14, Griffin, P., Competitive information in the stock market: an empirical study of earnings, dividends. Journal of Finance 31, Hawkins, E., Chamberlin, S., Daniel, W., Earnings expectations and security prices. Financial Analysts Journal 40, 24-38, 74 Holloway, C., A note on testing an aggressive investment strategy using value line ranks. Journal of Finance 36, Hoesli, M., Oikarinen, E., Are REITs real estate: evidence from international sector level data. Journal of International Money and Finance. 21, Howe, J., Unlu, E., Yan, X., The predictive content of aggregate analyst recommendations. Journal of Accounting Research 47, Imhoff, E., Lobo, G., Information content of analysts' composite forecast revisions. Journal of Accounting Research 22,

19 17 Jegadeesh, N., Kim, J., Krische, S., and Lee, C., Analyzing the analysts: when do recommendations add value? Journal of Finance 59, Kang, S., Zhao, Y., The information content and value relevance of depreciation: A crossindustry analysis. The Accounting Review 85, Latané, H., Jones, C., Standardized unexpected earnings a progress report. Journal of Finance 32, Liow, K., Property company stock price and net asset value: a mean-reversion perspective. Journal of Real Estate Finance and Economics 27, Liow, K., Li, Y., New asset value discounts for Asian-Pacific real estate companies: longrun relationships and short-term dynamics. Journal of Real Estate finance and Economics 33, Patel, K., Pereira, R.,Zavodov, K., Mean-reversion in REITs discount to NAV and risk premium. Journal of Real Estate Finance and Economics, 39, Schipper, K., Analysts forecasts. Accounting Horizons 5, Stickel, S.E., Common stock returns surrounding earnings forecast revisions: more puzzling evidence. Accounting Review 66, Toluca,W., Myer, F., and Webb, J., Dynamics of private and publish real estate markets. Journal of Real Estate Finance and Economics 21, Vincent, L., The information content of funds from operations (FFO) for real estate investment trusts (REITs). Journal of Accounting and Economics 26, Womack, K., Do brokerage analysts recommendations have investment value? Journal of Finance

20 Table I Sample Selection Number of REITs by Sector REITs % of Total Office % Retail % Multifamily % Lodging/Resorts % Health Care % Diversified % Other % Residential % Industrial % Self-Storage % Total REITs % Table 1 describes the type of Equity REITs that are within the sample by sector. The total number of REITs in the sample is

21 19 Table II Descriptive Statistics N=392,509 Panel A: Variables of Interest Mean Median Abs Val ΔNAV 1.66% 0.83% Raw ΔNAV 0.29% 0.18% Number of days between NAV revision Price/NAV Premium Number of Analysts Following REIT Analyst NAV Estimate per Share Panel B:Firm Characteristics Mean Median Total Assets 3,883,695 2,335,140 Volume Market Cap 2,917,047 1,495,886 Accumulated Depreciation 675, ,613 Total Liabilities 2,197,394 1,312,950 Total Debt 1,998,425 1,206,916 Leverage Market Value of Assets 5,060,412 2,917,603 Net Income 20,832 9,605 Panel C: Sample by Year Year # Observations # of Revisions Mean. ΔNAV Mean Price/NAV Mean number of analysts , % ,553 1, % ,577 2, % ,873 2, % ,205 2, % ,728 2, % ,126 2, % ,496 2, % ,822 2, % ,630 2, % ,901 3, % ,242 3, % ,472 3, % ,197 3, % ,198 3, % Total/Average 392,509 36, % Panel A of this table provides details of the descriptive statistics of the variables of interest. Each variable is defined in the appendix. Absolute Value of ΔNAV is taken to eliminate the canceling out effect of positive and negative estimate revisions. The mean reflects only values of ΔNAV that are non-zero and provide an average size of estimate revision. Panel A examines the primary variables of interest. Panel B

22 provides detail of the firm characteristics (all variables in Panel B in 000's). Panel C presents panel A statistics by year. 20

23 21 Table III CAR around ΔNAV Panel A: Isolated Subsample CAR (%) by Window REIT VW REIT EW REIT VW REIT EW Quintile [-2,+2] [-2,+2] [-1,+1] [-1,+1] Q *** *** ** *** Q ** *** Q Q *** ** Q *** 0.275*** 0.238*** 0.191*** Q5-Q *** 0.579*** 0.394*** 0.400*** Panel B: Isolated Subsample AR (%) by Day REIT VW REIT EW REIT VW REIT EW Day [-2,+2] [-2,+2] [-1,+1] [-1,+1] *** 0.039*** ** 0.024* 0.027** 0.024* *** 0.056*** 0.055*** 0.056*** *** 0.027* 0.034*** 0.027* Full Window 0.166*** 0.160*** 0.116*** 0.107*** Table III presents cumulative abnormal returns for 5 and 3 day windows surrounding NAV estimate revisions. Panel A shows abnormal returns for quartile portfolios sorted on ΔNAV. The largest negative revisions are in quartile 1 and the largest positive revisions are in quartile 5. Panel B presents returns by day within the 5 and 3 day windows. ARs in Panel 2 are calculated by multiplying abnormal returns around negative revisions by (-1). Each Panel is presented for abnormal returns against 2 return indices: CRSP Ziman REIT VW Index (REIT VW) and CRSP Ziman REIT EW Index (REIT EW). *,**,*** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively.

24 22 Table IV Regression Analysis: ΔNAV on CAR for Isolated Sample Return Window [-2,+2] [-1,+1] (1) (2) (3) (4) (5) (6) (7) (8) INTERCEPT (0.86) (0.77) (0.63) (0.68) ΔNAV *** *** *** *** *** *** *** *** (4.74) (4.820) (4.65) (4.62) (4.73) (4.72) (4.70) (4.66) MC 6.77E E E E-08 (1.13) (0.34) (0.80) (0.74) VOL 2.00E E E E-10 (0.80) (-0.11) (0.28) (-0.80) #Analysts -7.29E E E E-05 (-0.98) (-0.75) (-0.75) (-0.92) Observations 12,122 12,122 12,122 12,122 12,122 12,122 12,122 12,122 R-Squared Fixed Effects (Year/Firm) No No Both Both No No Both Both Table IV presents univariate and multivariate regression results for the isolated sample of NAV revisions. The isolated sample are those revisions which have no other revisions within a [-2,+2] window around them and are not within a [-12,+12] window of a 10-K or 10-Q filing date. Table IV examines a regression of ΔNAV on abnormal return for 3 and 5 day windows for the Ziman REIT VW index. Year and firm fixed effects were used for regressions which indicate "Both". Results are qualitatively similar for the Ziman REIT EW index. Vol is the number of shares traded on day t. MC is equal to the Market Cap of REIT i on day t, divided by 1,000,000. #Analysts is the number of analysts following REIT i on day t. The numbers in parenthesis are t-statistics. *,**,*** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively.

25 23 Table V Regression Analysis: ΔNAV on CAR for Clustered Sample Return Window [0,+2] [0,+5] [0,+8] (1) (2) (3) SumAR MC VOL #Analysts *** *** *** (6.66) (7.72) (6.80) *** *** *** (10.53) (7.55) (4.32) *** *** *** (6.75) (3.64) (4.35) *** *** (3.21) (3.65) *** *** (3.53) (3.92) *** (0.80) (3.28) *** (3.05) * (1.88) ** (2.13) 1.035*** *** *** (124.73) (144.13) (139.79) -3.00E E E-08 (-0.57) (-0.11) (-0.19) 0.00E E-10* -4.00E-10*** (1.27) (-1.67) (-2.85) -8.52E E E-07 (-1.43) (-1.22) (0.00) Observations 15,806 11,649 9,152 R-Squared Fixed Effects (Year/Firm) Both Both Both

26 Table V presents results for a regression on the sample of revisions which includes revisions close together. ΔNAV0 is the first revision in NAV in a potential series of changes. In column 1, the first revision is defined as having no other revisions during the previous 2 days. For column 2, the first revision is defined as having no other revisions in the past 5 days, and so on. ΔNAVt is equal to the percentage change in NAV on day t. If there is no change in NAV on day t, then ΔNAVt is equal to 0. SumAR is equal to the sum of abnormal returns during days in the series in which there is no revision. See text for more in depth explanation. Market Cap, Volume, and Number of analysts are calculated the same as in table IV. Numbers below coefficients in parenthesis are t-statistics*,**,*** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively. 24