Recent changes to accounting standards due to the Sarbanes-Oxley Act have

Transcription

1 1. Introduction Recent changes to accounting standards due to the Sarbanes-Oxley Act have limited the reporting of non-gaap information in financial statements, but industry advocates claim that this may reduce the quality of financial reporting since many industries have been providing supplemental non-gaap information that are deemed to be useful to financial statement users. In the Real Estate Investment Trust (REIT) industry in the U.S., firms regularly report a non-gaap summary performance measure known as Funds from Operations (FFO) as an alternative to net income in measuring firm profitability. FFO is generally calculated by adding back to net income the amount of depreciation and amortization related to real estate properties, gains and losses on the sale of real estate assets and certain other unusual and infrequent specific accrual items (e.g., extraordinary items, early extinguishments of debt). Since the introduction of the FFO concept by the National Association of Real Estate Investment Trust (NAREIT) 1 in 1991, industry participants have been advocating the adoption of FFO as they perceive it to be more useful (in terms of value relevance and the ability to predict future operating performance) than net income. The widespread and consistent adoption of FFO among REIT firms has provided academic researchers an ideal platform to compare the usefulness of GAAP vs. non-gaap measures. However, there is presently no consensus among academics and no concrete empirical evidence on which of these summary measures is more useful. Previous academic efforts present mixed evidence on the relative usefulness of net income and FFO (e.g., Fields et al. (1998), Gore and Stott (1998), Vincent (1999), Graham and Knight (2000)) as all previous studies seem to show 1 NAREIT is the representative organization for U.S. REITs as well as for real estate companies worldwide. Their main tasks include providing representation before national and state policymakers affecting the REIT industry and providing a network among member firms. 1

2 that both measures are, in general, value relevant but have different explanatory power and different predictive ability of future operating performance. Moreover, no study has been conducted since NAREIT standardized the FFO definition in 2000, which supposedly should have made FFO a better measure. The purposes of this study are to provide new evidence on the relative quality of FFO vs. net income and to determine whether investors are aware of the quality differences of the two measures in the valuation of REIT securities. In this study, I provide new evidence of the relative usefulness of FFO and net income by comparing the quality of accruals in the two measures. Dechow (1994) shows that the beneficial role of accruals is to shift and adjust cash flows over time to mitigate timing and matching problems in realized cash flows such that the adjusted numbers better measure firm performance. 2 Hence, accruals should theoretically be highly correlated with a firm s prior, contemporaneous, and future cash flow streams and with firm s fundamentals such as revenue. Nonetheless, the measurement of accruals requires managers to make estimates. Estimation errors, both intentional and unintentional, could make the performance measure less reliable and map poorly into realized cash flows and firm s fundamentals. I define accruals quality as the ability of accruals to map into realized cash flows and firm s fundamentals. It is empirically evaluated from regressions relating accruals to past, present, and future cash flows and to changes in contemporaneous revenue. Large residuals from the regressions indicate poor mappings of accruals into realized cash flows and firm s fundamentals and imply low quality of accruals. By comparing the quality of accruals for FFO and net income, I offer an 2 The notion that accruals improve the ability of performance measurement is also supported by the FASB (e.g., see Statement of Accounting Concept No.1, FASB 1978, paragraph 44). 2

3 alternative perspective on the usefulness of the two measures that is not in terms of value relevance, information content, or predictive ability of future operating performance. Given that FFO and net income differ predominantly in the inclusion or exclusion of some specific accrual items, an analysis of accruals quality evaluates whether the accrual adjustments in FFO improve the mapping of FFO into cash flows and help the matching of revenue with expenses. My results show that the accruals component of FFO is of higher quality than the accruals component of net income, lending support to the industry s claim that FFO provides incremental information benefits by undoing accruals that distort the operating performance measure. The second objective of this study is to examine whether investors consider the accruals component and the quality of accruals in the pricing of REIT securities. I conduct market-based tests to examine the associations of the accruals component as well as the quality of accruals with stock prices. The analysis allows us to revisit findings in previous research and to investigate if the conflicting results in prior studies were driven by the difference in the quality of accruals inherent in the two measures. By decomposing the summary performance measures into a cash component, a component of accruals included in FFO and a component of accruals excluded in FFO but included in net income, I show, in general, all three components are valued by investors. However, on the association of stock prices and accruals quality, I find that while the quality of accruals in FFO is positively related to security prices, the quality of accruals in net income is not significant; implying accruals quality of a non-gaap performance measure serves as a good proxy of financial reporting quality that constitutes a priced factor in security valuation. 3

4 The present study contributes to existing literature in the following ways. First, this study complements real estate research on the pricing of REITs and provides support to real estate industry professionals choice to report FFO as supplemental information in the financial statements. Previous real estate studies (e.g., Graham and Knight (2000), Gentry et al. (2003)) find FFO a relevant pricing factor that helps explain returns of REIT and claim that FFO provides incremental information without specifying the information benefits FFO provide. Given that accruals quality is used as proxy of information risks (Francis et al. (2005)), I show in this study that one of the benefits provided by FFO is an alternative measure that reduces information risks by promoting better mapping of cash flows and better matching of revenue and expenses. Second, by comparing FFO and net income in terms of accruals quality, I circumvent problems encountered by some previous studies that run a horse-race to compare the two measures and provide new evidence on the relative usefulness of FFO vs. net income. Some previous studies evaluate the usefulness of the two measures by comparing their correlations with future operating measures such as one-period-ahead cash flows. However, these studies disregard the fact that the two measures differ mainly by their accruals components and that accruals can be related to prior, contemporaneous, and future cash flows. Hence, a test of cash flow predictability only reflects how the differences in their accruals components affect the measures correlations with future cash flows and ignores the associations of these components with prior and contemporaneous cash flows. In contrast, an accruals quality model relates accruals to a series of cash flows over time and helps evaluate the ability of the performance measures 4

5 to predict future cash flows as well as to record prior and contemporaneous periods cash transactions. 3 Third, the paper contributes to the accounting quality literature by extending the methodology for estimating accruals quality in two directions. First, previous studies predominantly calculate accrual/ earnings quality using the Jones (1991) model, the modified Jones model by Dechow et al. (1995), the Dechow and Dichev (2002) model or a combination of those models (see Rees et al. (1996) and Francis et al. (2005)). However, Richardson et al. (2004) show that all these models are mis-specified since accruals are correlated with a firm s growth and future performance. Responding to this criticism, this study incorporates a growth/performance-controlled accruals quality measure. Second, REIT firms are characterized by the frequent occurrence of industryspecific accrual transactions that do not directly affect current cash flows. 4 Without adjusting for these transactions, the use of a general accruals quality model to link accruals with cash flows seems inappropriate. Hence, the accruals quality model in this study also includes industry-specific factors to control for the effects of specific real estate accrual transactions. The present study outlines an improved model for estimating accruals quality that can be easily adopted in other performance measurement studies. Fourth, the market-based tests conducted in this study to examine stock price associations with accruals extend existing literature on the usefulness of accruals and the 3 One criticism on using accruals quality as a benchmark to compare the two measures is that the comparison may be biased towards claiming FFO as the superior measure since some of the accruals excluded from FFO such as depreciation should have lower relation to cash flows than the included items such as accrued revenue. However, theoretically accounting depreciation should match revenue and expense to a reasonable extent; nonetheless FFO is also a voluntary measure which may more likely be manipulated by managers. Hence, given these tradeoffs it is an empirical question whether FFO accruals would still map better into a firm s cash flow stream than net income accruals. 4 For example, REIT firms are characterized by the frequent occurrence of asset write-downs, which is usually large-magnitude accrual adjustments that substantially affect the size of total accruals and the calculation of accrual quality. 5

6 quality of accruals in asset pricing. By decomposing the overall performance measures into their cash and accruals components and by including the quality of accruals as a risk factor in an asset pricing equation, I modify prior research design on the pricing of accruals and accruals quality by showing that in the presence of alternative summary performance measures, it is possible that different measures and different components have different explanatory powers on stock price. Moreover, I provide novel evidence on investors awareness of the accruals quality difference of the alternative summary measures in the pricing of securities. Finally, while all previously mentioned studies utilize sample observations before 2000, I hand-collect my sample for the period of The year 2000 is important for studies on FFO since the NAREIT then redefined and standardized the definition. It is likely that better guidance on the reporting of FFO by the NAREIT since 2000 may have a significant impact on improving the quality of the FFO measure. The present study provides more timely and relevant empirical findings that take into consideration this recent refinement of FFO. The remainder of the paper is organized as follows. The next section discusses related literature on the usefulness of FFO as well as on other non-gaap summary performance measures. Section three provides background on the REIT industry and the development of the FFO concept. In section four I outline the research design and present my hypotheses. The fifth section explains the sample selection process and contains a description of the data. Section six presents the empirical results while the last section concludes. 6

7 2. Review of Related Literature Surprisingly, there are only a few studies in the real estate literature on the reporting of FFO. One notable study by Graham and Knight (2000) examines the incremental information content of FFO. Using three alternative models (a price model, a returns model, and a price-changes model), they find that FFO has higher information content than net income and that it contains incremental information beyond that contained in net income. In another study, Gentry et al. (2003) perform an empirical examination of the relationship between dividend taxes and REIT share prices. They find that the inclusion of FFO as an explanatory variable improves the performance of their valuation model and that FFO provides value relevant information incremental to GAAP and dividend information. The accounting literature on REITs focuses predominantly on the horse-race comparison of FFO and net income as a summary performance measure. Fields et al. (1998) compare the predictability of these two measures on one-year-ahead FFO, cash flows from operations (CFO), net income and stock prices. Their results show that FFO dominates net income in predicting one-year-ahead FFO or CFO, but net income dominates in predicting stock prices and one-year-ahead net income. Vincent (1999) conducts a similar study but she examines the relative as well as the incremental information content of FFO as compared to earnings per share (EPS), CFO, and earnings before interest, tax, depreciation and amortization (EBITDA). Her empirical results show that all four measures provide information content and that their statistical significance is highly dependent on the econometric specifications. Gore and Stott (1998) on the one hand contrast with Fields et al. s (1998) study and show that FFO is more closely 7

8 associated with stock returns than net income. On the other hand, by regressing ex-post future one-to-three-year-ahead dividends on present year FFO and net income, they find that dividend forecast ability tends to be higher for net income than for FFO. Hence, it remains unclear which measure is more value relevant. The present study on FFO is also related to the broader literature that compares GAAP to other non-gaap measures. Bradshaw and Sloan (2002) are among the first to examine the use of non-gaap performance measures such as pro-forma or Street earnings. They show that there has been a dramatic increase in the disparity between proforma and net income and that the market participants value pro-forma more than net income as the primary determinant of stock prices. Bhattacharya et al. (2003) show that pro-forma earnings are more informative and more persistent than net income. Brown and Sivakumar (2001) compare pro-forma earnings with GAAP earnings in terms of their future earnings predictability, value relevance, and information content and find that proforma earnings are of higher quality. Lougee and Marquardt (2002) confirm Brown and Sivakumar s findings by showing that pro-forma earnings have greater information content when GAAP earnings are of lower quality. Recently, studies by Yi (2006), Heflin and Hsu (2005), Marques (2006) and Kolev et al. (2007) examine the effects of intensified scrutiny over pro-forma earnings by the SEC as a consequent of Sarbanes- Oxley Act of Finally, this paper relates to recent studies on the pricing of accruals quality. Measuring accruals quality as the standard deviation of residuals from regressions relating current accruals to cash flows, Francis et al. (2005) show that poorer quality is associated with larger costs of debt and equity. Cohen (2003) documents a positive 8

9 association of investor s demand of firm-specific information and a negative association of proprietary costs with financial reporting quality. 3. Background on REIT and the Concept of FFO Real Estate Investment Trusts (REIT) were first created in the U.S. by Congress in 1960 for the purpose of allowing individual investors to invest in real properties. It is essentially a closed-end fund where investors invest in a company that manages a pool of investment assets consisting predominantly of real properties. The REIT industry enjoyed tremendous growth over the past few decades. By 2000 there were a total of approximately 193 publicly-traded REIT firms, of which 156 of them are equity REIT, 24 are mortgage REIT and 13 are considered hybrid. 5 The major difference between equity REIT and mortgage REIT is that an equity trust manages real properties while a mortgage trust usually manages a pool of mortgage obligations. To increase the attractiveness of investing in REIT companies, the Congress elected to waive the corporate level income tax on REIT. At the same time, the legislation also placed tight restrictions on the REIT industry with a unique set of rules governing the operation, organizational structure, and dividend and financial policy of REIT companies. To qualify as a REIT, a company must meet the following requirements: Earn no less than 75% of its income from real estate related investments. Distribute a minimum of 90% of its annual taxable income as dividends. 5 Source: Wang, K., S.H. Chan and J. Erickson, 2003, Real Estate Investment Trusts: Structure, Performance, and Investment Opportunities. 9

10 At least 75% of assets in REIT must be invested in real estate assets, mortgages, REIT shares, or government securities and cash. No more than 30% of the entity s gross income can be derived from sale or disposition of stock held for less than six months or real property held for less than four years other than property involuntarily converted or foreclosed on. REIT has to be an investor in real estate and cannot be a real estate broker. Regarding ownership structure, REIT needs to have at least 100 shareholders and the five largest shareholders cannot control more than 50% of the shares of the REIT firm. In financial reporting, net income has been recognized as the key summary metric of a company s financial performance. 6 However, the REIT industry argues that net income, computed according to GAAP, does not accurately reflect the profitability of real estate assets. The industry believes that traditional GAAP-governed financial measures are ineffective in capturing value relevant information for REITs and urges NAREIT to develop a non-gaap industry measure supplemental to GAAP financial information. In 1991, NAREIT defined a summary performance measure called Funds from Operations (FFO) and the term soon became widely used by the industry. NAREIT is convinced that the industry benefits from having this alternative measure and is encouraged by the publications of the American Institute of Certified Public Accountant (AICPA), the 6 Though the REIT industry has traditionally been recognized as a cash flow business and hence cash flows from operation (CFO) is historically considered the measure for firm performance, net income and FFO have received more attention in recent years as the accounting profession advocates that accrual performance measures provide incremental information beyond CFO as they differ from CFO by (i) the changes in accounts receivable, accounts payable and inventory, (ii) changes in working capital and (iii) other accruals such as equity income, stock compensation and straight-line rents. 10

11 Association of Investment Management Research (AIMR), 7 and the Security and Exchange Commission (SEC) s Accounting Series Release 142 which all advocate the development of industry standard accounting terms. Under the rules set forth in Item 10(e) of Regulation S-K by the SEC, REIT companies are free to report their non-gaap financial measures as long as they provide reasons as to why management believes such measures benefit investors. Managers generally claim that by excluding depreciation, amortization, and some non-recurring accrual expenses, FFO provides useful information about firms operating performance. They believe that accrual adjustments to derive net income, including depreciation and non-recurring accruals, distort the true profitability of REIT for three reasons. First, the conventions made for financial statement depreciation allocation usually assume that the value of real estate assets diminishes predictably over time. This assumption also tends to underestimate the period of time over which depreciation occurs. Since real estate values have historically risen or fallen with market conditions, managers consider the use of historical depreciation inappropriate for REIT firms that invest heavily in long-lived real assets. This problem may become more severe under volatile or inflationary market environments (such that the values of real estate assets exhibit larger fluctuations and deviate more from their book values that decrease predictably over time). Second, it is difficult to categorize some assets/ capital improvements owned by a REIT into the right asset classes such that the correct depreciation schedule can be applied. As a result, the accrual adjustments for depreciation could contain huge measurement error. Third, managers also believe that another advantage of FFO over net income is that it excludes unusual and extraordinary items that 7 AICPA, 1994, Improved Business Reporting: A Customer Focus, and AIMR, 1993, Financial Reporting in the 1990 s and Beyond. 11

12 are deemed to have little relation with the firm s future operating performance. However, given that there is no standard method to calculate FFO and that FFO is not audited (hence the possibility that managers can cherry-pick accounting items to include in or exclude from FFO), the SEC and accounting standard setters doubt the usefulness and reliability of the FFO measure. Responding to the controversies among REIT companies and standard setters on the reliability of the FFO measure, NAREIT clarified its definition of FFO in 1995 and 1999 and published a White Paper on Funds from Operation in April 2002 to recognize the FFO definition created in 1999 (effective January 1, 2000). The standard 8 definition is as follows: Funds from Operations means net income (computed in accordance with GAAP), excluding gains (or losses) on sales of property, plus depreciation and amortization, and after adjustments for unconsolidated partnerships and joint ventures. NAREIT s White Paper reinforces the idea that the industry benefits from the reporting of FFO as a supplemental measure to net income. In a recent news release, 9 NAREIT also indicated that the creation of Regulation G (on the issue of non-gaap financial information) to implement Section 401(b) of the Sarbanes-Oxley Act by the SEC allows the continued use of FFO in SEC filings as well as all other public disclosures as supplementary information to the GAAP financial statements. 8 REIT firms are encouraged to follow the standard definition of FFO as defined by NAREIT. However, managers are not obligated to strictly follow the standard definition and they have the flexibility to make further adjustments on items to include or exclude that they deem appropriate. 9 REIT Alert, Use of FFO in SEC Filings after Recent Rulemaking on Non-GAAP Financial Information, March 4,

13 4. Hypothesis Development/ Research Design 4.1 Comparing the accruals quality of FFO and net income Proponents of FFO claim that FFO is a better performance measure due to the reversal of those accruals that do not accurately reflect profitability of REITs. Given that FFO provides management with more discretion than net income on which items to include or exclude, proponents of FFO believe that managers have the potential to adjust FFO such that it can mitigate estimation problems in depreciation and better match cash flows over time. If this proposition is true, FFO should possess higher accruals quality than net income. Opponents of FFO, on the other hand, believe that flexibility of the measure does not necessarily ensure that FFO has higher quality. By undoing some particular accrual transactions, FFO merely increases net income by the amount of the accrual adjustments. However, the accrual items excluded (such as depreciation) may still be significantly related to contemporaneous and future revenues and cash flows of the firm. Moreover, it is possible that REIT managers are more likely to manipulate FFO given that it is not governed by GAAP. The first hypothesis investigates which of the two measures possesses higher accruals quality: H1: The accruals component of FFO is of higher quality than the accruals component of net income. Previous literature has been using both indirect and direct methods to calculate total accruals. 10 Collins and Hribar (2002) argue that the direct method provides a more accurate measurement of total accruals. Hence, this study uses the direct method and defines total accruals in net income and in FFO respectively as: 10 The indirect method calculates total accruals using information from both the balance sheet and income statement. The direct method calculates total accruals directly from the statement of cash flows. 13

14 TACC NI = NI CFO TACC FFO = FFO CFO Dechow (1994) claims that the purpose of accruals is to shift the recognition of a firm s cash flows across different periods to mitigate timing and matching problems such that net income better reflects profitability of the reporting period. Dechow and Dichev (2002) argue that the sums of accounting income and the sums of cash flows over time should be equal theoretically unless accruals suffer from serious measurement errors. Hence, they hypothesize that accruals are of higher quality (subjected to less measurement errors) if the accruals have higher correlation with prior year, contemporaneous, and one-year-ahead cash flows. Although the Dechow and Dichev s model focuses on current accruals, their intuition can be extended to non-current accruals as well. In Appendix A, I outline the Dechow and Dichev s accruals quality model and illustrate how the model is able to capture estimation errors for both current and noncurrent accruals. However, I also show in the Appendix that one shortcoming of their model is that the coefficients for the explanatory variables are biased when the impact of accruals extends beyond one period. A better specified model should explain accruals with a longer time series of the firm s cash flows and should include accruals of adjacent periods as explanatory variables. Unfortunately, such a model imposes significant data constraints and is difficult to implement empirically in most settings. Hence, I follow Dechow and Dichev (2002) and include only the one-period-lagged and lead, and present cash flows as explanatory variables in my main econometric specification. 11 However, to mitigate the short time horizon problems in the model of Dechow and Dichev, I follow 11 Nonetheless, as robustness check I also examine an alternative specification with the addition of twoperiod-lagged and lead cash flows and one-period-lagged and ahead accruals on a smaller sub-sample. 14

15 alternative suggestions by McNichols (2002) and Francis et al. (2005). I augment the Dechow and Dichev accruals quality model by including changes in revenue ( REV) and the gross value of property, plant, and equipment (PPE) as explanatory variables in order to capture the long-term impact of non-current accruals. However, Richardson et al. (2004) show that these accruals quality models still suffer from mis-specification errors as total accruals are highly correlated with firm performance. Moreover, a general accruals quality model may not work well in the context of the REIT industry. Hence, I further modify the accruals quality models from prior studies and evaluate accruals quality of net income and FFO using the following regression framework: TACC NIj,t = α 1 + β 11 CFO j,t-1 + β 21 CFO j,t + β 31 CFO j,t+1 + β 41 REV j,t + β 51 PPE j,t + 2 γ 11 IMP j,t + γ 21 DEBT j,t + γ 31 ARENT j,t + κ 11 GROWTH j,t + κ 21 ROA adj j,t + ε NIj,t (1) TACC FFOj,t = α 2 + β 12 CFO j,t-1 + β 22 CFO j,t + β 32 CFO j,t+1 + β 42 REV j,t + β 52 PPE j,t + 2 γ 12 IMP j,t + γ 22 DEBT j,t + γ 32 ARENT j,t + κ 12 GROWTH j,t + κ 22 ROA adj j,t + ε FFOj,t (2) All financial statement variables are scaled by average total assets to control for heteroskedasticity. CFO, REV, and PPE 12 are cash flows from operations, changes in contemporaneous revenue, and the gross values of plant, property and equipment, respectively, and are the variables used in the Francis et al. (2005) model. IMP, DEBT, and ARENT are the industry factors I include in my accruals regression models. 13 Rees et al. (1996) suggest that asset write-downs with large accrual adjustments signify a 12 Alternatively, I use NAV (Net Asset Values) instead of PPE and run a sub-sample regression for firms with NAV data available from Green Street Advisors. The results are weaker due to a smaller sub-sample but are qualitatively the same. I thank Qintao Fan for sharing the NAV data. 13 One caveat of including these industry factors in the model is that the empirical results may bias towards favoring net income since expenses such as asset write-down or debt restructuring are included in net income accruals but excluded in FFO accruals. Hence in alternative specifications I compare the quality of accruals in net income and FFO without these industry control factors. Results have not changed after the exclusion of these variables. 15

16 permanent shift in accruals in the write-down years. Given that REIT predominantly invests in real properties, ignoring the issue of asset write-downs leads to serious omitted variable bias in the regressions. Therefore, I include the variable IMP, defined as the amount of asset write-down, to capture this accruals effect. Investments in real estate assets by REIT usually involve a high degree of leverage. In adverse interest rate environments, it is not uncommon for REIT to renegotiate its loans with the bank and undergo debt restructuring. I include DEBT, the amount of debt restructuring expense, into my accruals regression model to signify debt restructuring. Under current accounting rules (FASB Statement No.13), REIT is required to report rental revenue with straightlined rents for net income under GAAP but has an option to adjust out the straight-line rents in the calculation of FFO. 14 I control for the impacts of straight-line rents using the measure ARENT, which represents the amount of adjustments with straight-lining of rents and expect a positive relation of ARENT and FFO accruals. To control for the impact of growth and performance on the size of total accruals, I include the variables GROWTH and ROA 2 adj. Since firms with larger growth opportunities may report higher accruals all else being equal, the inclusion of GROWTH is intended to control for possible correlations between firm growth and the magnitude of accrual adjustments. I measure GROWTH as the average annual growth rate of PPE because properties represent the key productive asset for REITs. 15 Kothari et al. (2005) introduce a performance-matched accruals measure (by industry and year) to address the non-linear 14 Rental contracts of real properties (especially office buildings) usually have an escalation clause where contracted rents rise in the later years of the contract term. The straight-lining of rents is an aggressive revenue recognition method that allows REIT to estimate the sum of the present values of the rental streams and to recognize rental revenue equally over the life of the contract. 15 Alternatively, I measure GROWTH as the average annual growth rate of the revenue of the firms. However, it does not change the empirical results. 16

17 relation between accruals and firm performance. They show that accruals for firms with better or extreme performances are expected to be higher. I control for this non-linear relation of accruals and firm performance by using ROA 2 adj. ROA adj is calculated as the difference between the firm s ROA and the industry average ROA by year. ROA 2 adj, the square of ROA adj, is intended to control for the positive non-linear relation between accruals and extreme firm performance. To compare the accruals quality of FFO and net income, I first examine the adjusted R 2 of the regressions. The adjusted R 2 measures provide an assessment on how well the accruals components of FFO and net income map into cash flows and firm s fundamentals. If accruals of FFO are of higher quality than net income, the regression fit of the model should be better with lower residuals and higher adjusted R This implies that by excluding certain accrual adjustments, the accruals component of FFO has smaller measurement errors than the accruals component of net income. Alternatively, I also compare statistically the parameter estimates in (1) and (2) using the Chow test. Since (1) and (2) are nested models of each other, the Chow test is appropriate in this case to evaluate whether the two groups of observations (i.e., net income accruals and FFO accruals) are significantly different from each other. 16 Since FFO differs from net income mainly by excluding depreciation and other non-current accruals, one may argue the finding that FFO has higher adjusted R 2 is mechanical because current accruals should have less measurement errors than non-current accruals that extend for more than one period. However, given that measurement errors can be due to intentional manipulations or unintentional mistakes, a priori there is no reason to believe that the non-current accrual adjustments excluded in FFO but included in net income will lower the adjusted R 2 for regression specification (1). Even though non-current accruals are expected to contain more measurement errors because their impacts extend for more than one period and are harder to predict than current accruals, Teoh et al. (1998b) also show that current accruals are subjected to serious managerial manipulation and managers do not seem to manage non-current accruals. Hence, measurement errors in current accruals can be higher or lower than measurement errors in non-current accruals and whether FFO or net income has higher accrual quality remains an empirical question. 17

18 4.2 The pricing of the quality of FFO and net income accruals Having established that accruals quality differs between FFO and net income, I next investigate whether investors are aware of these differences when they use the summary performance measures in the pricing of REIT securities. Previous research efforts (Fields et al. (1998), Vincent (1999)) have established that both net income and FFO are value relevant to investors. Other more recent empirical work (Francis et al. (2005)) shows the direct association between accruals quality (as a proxy of information risk) and cost of capital. I combine and extend these two streams of research by examining the impact of accruals quality on the pricing of alternative summary measures. Though net income and FFO as well as their cash and accruals components are expected to be associated with stock prices, no study has compared the value relevance of the quality of alternative performance measures. In the presence of two alternative measures, it is possible that the quality of one measure may represent a better proxy of information risks and may be valued more heavily by investors. If investors recognize that given the voluntary nature of FFO, the quality of FFO can better reflect the integrity of management and the risks of the information environment, then investors may place more emphasis on the quality of FFO than net income. If investors recognize the GAAPgoverned net income as the most objective evaluation of a firm s information environment, they may place more emphasis on the quality of net income. I hypothesize that the accruals quality of both measures are positively related to prices as poor accruals quality of either measures signifies higher information risk for the firm and thus higher costs of capital and lower share price, all else being equal. However, the quality of one measure may have higher stock price association than the other. 18

19 H2: Investors price both the accruals quality of FFO and of net income. H3: Investors place more emphasis on the accruals quality of FFO than on the accruals quality of net income. To test H2 and H3, I examine the association of stock prices and accruals quality of the performance measures using the following empirical specification: P j,t = α + β 1 CFO j,t +β 2 TACC FFOj,t + β 3 (TACC NIj,t -TACC FFOj,t )+ β 4 AQ FFOj,t + β 5 AQ NIj,t +β 6 CONTROLS j,t + ξ j, t (3) The dependent variable, P, is the end-of-year stock price of the REIT securities. I scale all financial statement variables by the outstanding number of shares to control for heteroscedasticity. CFO is included because it represents the cash component of the performance measures and is expected to have a positive relation with stock price. TACC FFO is the accruals component in the FFO measure. (TACC NI -TACC FFO ) is the accruals component included in net income but excluded in FFO. 17 The main coefficients of interest are β 4 and β 5, which their significances indicate if investors price the quality of the accruals components of the summary performance measures. AQ NI and AQ FFO are measured respectively as the inverse of the squared residuals from the accruals regressions in (1) and (2), since the accruals component that better matches cash flows and revenues should produce lower residuals in the accruals regressions. I include two alternative sets of variables to control for omitted variable biases. First, I based my research design loosely on the valuation model developed by Ohlson (1995) and assume price is a function of book values, net dividends, net income and other information not 17 I include (TACC NI -TACC FFO ) instead of TACC NI in the regression as TACC NI and TACC FFO are correlated. However, as alternative specification I also substitute TACC NI as an explanatory variable. Empirical results do not change with TACC NI as an explanatory variable. 19

20 captured by the accounting system. 18 Hence, I include book values and net dividends as control variables. Second, Peterson and Hsieh (1997) find that REIT returns are significantly influenced by size of the firms and their book-to-market ratios. Hence, alternatively I also include Log (SIZE) and BM as control variables. 5. Sample Selection and Data Table 1 summarizes my sample selection process. I first identify an initial sample of 190 real estate firms under SIC code 6798 (Real Estate Investment Trust) for which financial data are available in the Compustat annual industrial and research files during Since my study focuses on equity REITs that predominately invest in real assets and report FFO as their common industry practice, I eliminate 31 mortgage REITs, 6 hybrid REITs, and 4 liquidating trusts from the initial sample. For the remaining 149 firms, data on FFO, straight-line rents, and property, plant and, equipment are handcollected from 10-K filings through the SEC s EDGAR database. I further exclude 13 firms that do not have 10-K filings and 33 firms that do not report FFO in their financial statements. 19 My final sample contains 103 firms and 309 firm-year observations over the period In my research design, net income is decomposed into its cash and accruals components and the information not captured by the accounting system would be the perceived quality of the summary performance measures, which proxy for information risks. 19 The 13 firms do not report 10-K filings as they are either privatized or being acquired. Given that there are 33 firms that choose not to report FFO, my final sample may be subjected to potential selection and self-reporting biases that could affect my statistical results. To address the concerns of these sampling biases, I compare descriptive statistics of firms that do and do not report FFO. However, the two sets of firms do not seem to be significantly different. Hence, I believe that sampling biases have minimal impact on my empirical studies. 20

21 Table 2A contains descriptive statistics for the 103 sample REITs. 20 The mean value of TACC FFO is higher than TACC NI as the accruals included in net income but excluded in FFO are mostly expense items. TACC NI is also slightly more volatile than TACC FFO. REV has a mean of ; implying revenue of REIT firms grows by an average of 1.5% annually. PPE has a mean of and a median of , indicating the distribution is highly skewed. I eliminate observations with extreme PPE in my empirical tests to ensure the results are not driven by outliers. DEBT and ARENT are relatively small with means of and respectively. IMP has a mean of GROWTH has a mean of , which is consistent with the fact that REIT firms expand by acquiring new properties over the sample period. The variables in price regressions provide some insights on the magnitudes of net income, FFO and CFO. NI has a mean of , which is substantially lower than the means of FFO (99.23) and CFO (106.10) because of the inclusion of expense accruals in net income. Table 2B presents the Pearson correlations for variables used in the accruals quality regression model. Consistent with previous findings by Dechow and Dichev (2002), total accruals (TACC NI ) is negatively correlated with contemporaneous CFO. Similarly, TACC FFO is also negatively correlated with contemporaneous CFO. The three CFO variables are positively and significantly correlated with each other, indicating that CFO is serially correlated and that the inclusion of the lead and lag CFO variables in the accruals regression model helps to control autocorrelation in CFO. IMP is significantly negatively correlated with TACC NI. Both REV and PPE are significantly positively 20 The descriptive statistics cover all firm-year observations without eliminating extreme observations. Concerning with the effect of outliers, the empirical tests presented in this study are conducted based on a reduced sample where 5% (or 15 outliers) of observations are eliminated. I have also performed the same tests using the total sample and obtained stronger results. 21

22 correlated with GROWTH, indicating the variables included in Francis et al. (2005) may have already controlled for the growth factor. Table 2C presents the Pearson correlations for variables used in the price regression in (3). Consistent with expectation, Price is significantly positively correlated with NI, FFO and CFO. Moreover, Price is also highly correlated with the control variables with expected signs. 6. Empirical Analysis 6.1 Comparing the accruals quality of FFO and net income In solving equations (1) and (2), I run pooled regressions across all firms and years on a common sample using OLS with the inclusion of fixed effects for year and the type of properties that the REIT invests in. 21 The t-statistics are estimated using White heteroskedasticity consistent standard errors. One concern about my findings is that the results are dependent on the accruals regression specifications I employ. To ensure the results are not model-driven, I first compare the accruals quality of FFO and net income using the Francis et al. (2005) model. The first two columns of Table 3 report the empirical results. Judging from their adjusted R 2, the accruals component of FFO has significantly higher quality than the accruals component of net income ( vs ). The Chow F-statistic equals and is significant at the 1% level. Consistent with Francis et al. (2005), the coefficient estimates all have expected signs but TACC FFO is more highly correlated with the CFO variables and with PPE. The third and fourth columns of Table 3 show similar results for model specifications (1) and (2). After 21 Given that the error terms of (1) & (2) are likely to be correlated and that the dependent variables, total accruals of FFO and net income, are possibly serially correlated across time, I also estimate (1) & (2) using Seemingly Unrelated Regressions (SUR) with Newey-West heteroskedasticiy-autocorrelation-consistent standard errors. However, the empirical results of the SUR regressions are qualitatively very similar to the OLS regressions and hence not reported. 22

23 controlling for industry effects and growth and performance, the adjusted R 2 in (1) is still lower than that in (2) ( vs ) and the Chow F-statistics is significant at 1% level, indicating that the accruals component in net income has significantly lower quality than the accruals component in FFO. As for the coefficient estimates of the explanatory variables, consistent with findings of Francis et al. (2005), TACC NI is positively related to prior and future cash flows and negatively related to contemporaneous cash flows. TACC FFO has the same relationships with the cash flow variables but is shown to be more highly correlated with prior and contemporaneous cash flows. More importantly, TACC FFO is significantly related to future cash flows while TACC NI is not, which indicates that FFO has better cash flow predictability. REV is positive for both regressions but significant in (2), indicating that the accruals component of FFO can better match revenues with expenses. PPE is also significantly related to accruals in FFO but not net income. For both regressions, the coefficients for IMP and DEBT are negative as expected, indicating that large adjustments in asset write-downs and debt restructuring expense reduce total accruals. However, IMP is significantly related to only TACC NI, which is consistent with the fact that asset write-offs are included in net income but excluded from FFO. ARENT is insignificant perhaps because only a small amount of companies report this information. A positive coefficient for GROWTH shows that the size of accruals is positively correlated with growth of the companies; however, the coefficient is insignificant in both specifications. Finally, the performance-control is significant in (2), which shows that accruals are higher for firms with more extreme performance. 23

24 To ensure the robustness of my results, I next compare the accruals quality of FFO and net income using different variations of model specifications (1) and (2). First, I use reduced models of (1) and (2) that eliminate either the industry factors or the growth and performance controls. Next, given that some accruals (such as depreciation) can be related to investing cash flows, I also expand my accruals quality models (1) and (2) by including one-period-lagged, contemporaneous, and one-period-ahead investing cash flows as additional explanatory variables. To address the concern that free cash flows may be a better criterion to evaluate how accruals map into cash flows, I alternatively include one-period-lagged, contemporaneous, and one-period-ahead free cash flows, measured as CFO minus investing cash flows and dividends, as additional variable in models (1) and (2). 22 Finally, to mitigate the short time horizon problem in the Dechow and Dichev (2002) model, besides including REV and PPE as in Francis et al. (2005), I add two-period-ahead and lagged CFO and one-period-ahead and lagged accruals as additional explanatory variables. 23 The results reported in Table 4 indicate that accruals quality regressions for FFO accruals have significantly higher explanatory power than regressions for net income accruals under all specifications examined. 6.2 The pricing of accruals quality in alternative summary performance measures Table 5 compares the stock price associations of the components and the quality of FFO and net income accruals under different variations of specifications (3). I first conduct simple regressions of stock price on the NI and FFO measures to compare my findings with Fields et al. (1998). Column I and II of Table 5 show that FFO is more 22 Given that some industry practitioners believe free cash flows is the only criterion that matters to financial statement users, in unreported alternative specification I examine accruals quality using the free cash flows variables to replace the CFO variables in (1) and (2). However, I obtain similar results and find that the accruals component of FFO is of higher quality. 23 Because of data constraint I can only estimate this augmented regression model for the year

25 significantly related to stock prices than net income, which contrast with previous findings by Fields et al. (1998). The difference in findings can be due to a later sample period used in this study and the possibilities of increasing investor recognition and an improved FFO definition since Table 5, columns III to V, reports results of estimating (3) without the inclusion of the accruals quality variables. Column III shows that the cash component, the accruals component in FFO and the accruals component excluded in FFO but include in net income are all significantly related to contemporaneous stock prices. Column IV includes BV and D as control variables but the variables are insignificant. Column V includes Log (SIZE) and BM and the control variables are highly significant with expected sign. Overall, results in columns III to V show that the cash as well as the accruals components of the performance measures are significantly related to stock prices. Columns VI to VIII in Table 5 report results that include the AQ measures in the regressions. Interestingly, the AQ measure for FFO is highly positively significant (at 1% level) while the AQ measure for net income is negative and insignificant. Combining these results with the results in Table 3 (which shows FFO has higher accruals quality than net income) suggest that investors recognize the quality differences between the accruals components of FFO and net income. Francis et al. (2005) use accruals quality to proxy for the information risk associated with earnings and find that lower accruals quality is associated with higher cost of equity. My findings suggest the FFO measure, which has higher accruals quality than net income, actually serves as a better indicator of information risks. Investors may view the accruals quality of FFO as a better proxy of a firm s information risks than the accruals quality of net income because they may 25