Value Relevance of Historical Cost and Fair Value Accounting Information: Evidence from the European Real Estate Industry.

Transcription

1 Value Relevance of Historical Cost and Fair Value Accounting Information: Evidence from the European Real Estate Industry Fan Yang School of Accounting, University of New South Wales November 018 Abstract: Extant evidence on the value relevance of fair values for tangible asset varies depending on the type of asset being measured, sample firms being examined, as well as research design being employed. This study re-assesses the value relevance of historical cost versus fair value accounting information in the context of the European real estate industry around mandatory IFRS adoption. I use this sample because of its unique operating characteristics and the variation in the sample firms accounting treatment method for their primary operating assets, i.e., investment properties, around IFRS adoption. I find that after controlling for the effect of systematic changes surrounding IFRS adoption, switching from the cost model to the fair value model when accounting for investment properties at the adoption of IFRS (or IAS 40) improves the value relevance of both balance sheet and net income information. I further find that fair value accounting information are of higher value relevance in countries with the custom of asset revaluation. Keywords: Fair Value Accounting, Value Relevance; IFRS; IAS 40; Real Estate Industry; Investment Property 1

2 1 Introduction This study examines the effect of accounting model for investment properties on the value relevance of financial reporting information in the European real estate industry. There is no consensus on how investment properties should be reported on a firm s financial statement subsequent to initial recognition. Current accounting practices include both the cost model and the fair value model. For example, IAS 40 Investment Property [003] permits firms domiciled in countries adopting IFRS to choose between the cost model and the fair value model, whereas U.S. GAAP requires firms to apply the cost model only. Prior literature (e.g., Dietrich et al. 000; Danbolt and Rees, 008) identify reservations about mandating the fair value model when accounting for investment properties. One key issue is whether the fair value model provides more relevant information to financial statement users without a loss of reliability. Since highly liquid and transparent markets for investment properties do not generally exist, fair values for property assets are typically obtained based on estimation either by external appraisers or by the firm itself. As a result, fair values for investment properties are likely to be subject to measurement error or managerial discretion. Prior studies examining the value relevance of fair values for tangible assets provides mixed results (e.g., Beaver et al., 198; Bublitz et al., 1985; Easton et al., 1993; Barth and Clinch, 1998). Among these studies, whether evidence on the value relevance of fair value information can be documented depends on the type of asset being measured, the characteristic of the sample firm, as well as the sample period being selected. Therefore, whether investment properties reported at fair value under the IFRS regime are of higher value relevance than those reported at historical cost is unclear a priori. The mandatory adoption of IFRS, together with the application of IAS 40, creates a unique setting to investigate questions relating to fair value accounting. Prior to the adoption of IFRS, domestic GAAP around the world varied considerably with respect to the accounting treatment for investment properties, with some countries requiring or permitting the revaluation model, while others requiring only the cost

3 model. At the adoption of IFRS, IAS 40 allowed firms to choose between the cost and the fair value model and required firms selecting the cost model to disclose fair value in the notes to the accounts. The variation in firms accounting models across countries as well as around the adoption of IFRS provides the opportunity to evaluate the value relevance of historical cost versus fair value accounting information. The sample for this study consists of European listed real estate firms that adopted IFRS in 005. These firms have large amounts of investment properties in their operating assets, thus have been significantly impacted by the adoption of IAS 40. Because the balance sheet and the income statement reflect different aspects of financial reporting information, in investigating the value relevance of historical cost versus fair value accounting information, I examine the effect of accounting model on the value relevance of balance sheet information and income statement information separately. The value relevance of balance sheet information is examined based on a balance sheet model, and the value relevance of earnings and earnings components is evaluated based on a returns model. In the primary tests, I use firms domiciled in countries whose domestic GAAP requires the cost model as the sample. Depending on their choices of the accounting model for investment properties on the adoption of IFRS, these firms can be categorized into two groups: firms applying the cost model both pre-ifrs and post-ifrs and firms applying the cost model pre-ifrs and switching to the fair value model post-ifrs. These two groups of firms form a quasi-natural experiment setting which enables me to conduct difference-in-differences analyses (i.e., between-periods tests) to investigate the impacts of measurement base on the value relevance of accounting information. After controlling for the effects of systematic changes around IFRS adoption, results show that switching from the cost model to the fair value model at the adoption of IAS 40 exerts positive impact on the value relevance of both balance sheet and net income information. However, under the IFRS regime, for firms selecting the fair 3

4 value model to report investment properties, current-year changes in investment property fair values have no explanatory power to stock returns. I then use firms applying the cost model and disclosing fair values under the IFRS regime as the sample to corroborate the findings revealed from primary tests. I. This sample enables me to conduct within-firm tests to compare the value relevance of historical cost accounting amounts (recognized) and fair value accounting amounts (disclosed). Balance sheet model-based results confirm that levels of investment property amounts prepared at fair value are relatively more value relevant than those prepared at historical cost. Returns model-based tests marginally support that fair value-based earnings are more value relevant than historical cost-based earnings. However, because fair value-based earnings are obtained from selfconstruction rather than the firms recognition, results regarding earnings can be biased because of estimation error. Moreover, consistent with primary test results, unrealized gains and losses (analogous) are not incrementally value relevant to stock returns. To address the lack of value relevance for unrealized gains and losses documented in both the primary tests and in the within-firm tests, in additional tests, I examine whether the country-level difference in the custom of asset revaluation impacts the value relevant of post- IFRS fair value information. Results show that the value relevance of unrealized gains and losses varies significantly between the two groups. For firms whose domestic GAAP does not permit revaluation, only realized earnings are helpful in explaining share returns, whereas for firms whose domestic GAAP permits revaluation, only unrealized gains and losses are helpful in explaining share returns. This finding reinforces the broad question as to whether uniform reporting policy can achieve uniform reporting consequences (Daske et al., 013). This study bears directly on the issue of whether model-based fair value estimates will lack value relevance. Both the IASB and the FASB establish a fair value hierarchy, which categorizes fair value measurements into three levels based on the inputs employed to determine fair values. Recent evidence based on financial instruments reveals that Level 3 fair values are 4

5 discounted by investors in firm valuation (Kolev, 009; Song et al.,010). Because fair values for investment properties are mostly determined based on mathematical models and therefore falling under Level 3 in the fair value hierarchy, re-assessing the value relevance of accounting information for investment properties in the context of the adoption of IAS 40 adds to the accumulation of empirical evidence in this respect. This study also contributes to the growing literature examining the capital market effects of the mandatory adoption of IFRS around the world. IFRS have been argued to be a set of high-quality standards and one reason supporting this argument is the greater use of fair value under IFRS (Ball, 006; De George et al., 016). However, in examining the implications of IFRS adoption, extant literature generally treats IFRS as a comprehensive set of new accounting standards, which makes it difficult to attribute any documented consequences to a certain characteristic of IFRS (e.g., fair value-oriented, principle-based etc.) 1. Taking advantage of the variation in accounting treatment for investment properties across countries surrounding IFRS adoption, this study provides direct evidence on the capital market effect of IFRS adoption arising from the application of fair value accounting. More importantly, by addressing potential difference in the value relevance of fair value information across countries with different institutional, economic or cultural conventions, this study also adds to the discussion that the same accounting policy may exert consequences and implications in different environments in a non-uniform manner. The remainder of this study proceeds as follows: Section first reviews related literature, then provides institutional background information of this study and develops hypotheses. Section 3 introduces research design. Section 4 describes sample selection procedure and presents sample description. Section 5 reports primary empirical results. Section 6 presents results for additional analyses and Section 7 concludes. 1 For instance, studies can investigate whether there is any change in the value relevance or the comparability of financial reporting information upon IFRS adoption (e.g., Daske et al., 008; Armstrong et al., 010). 5

6 Literature Review, Research Setting and Hypotheses Development.1 Prior Literature In the past decades, there have been numerous accounting studies empirically investigating the relationship between share prices (or share returns) and accounting amounts recognized or disclosed in financial reports. These studies are referred to by Holthausen and Watts (001) and Barth et al. (001) as value relevance literature. In the value relevance literature, an accounting amount is value relevant if it is priced in an equity valuation model in a predicted way. Advocates of fair value accounting propose that fair value measurements are more relevant than historical cost measurements, if fair values can be reliably estimated (Barth et al., 001; Landsman, 007). Studies exploring the value relevance of fair value information for non-financial assets provide mixed results. One stream of studies examines the value relevance of fair value disclosures required by SFAS 33 in the U.S. Early studies mostly fail to provide evidence that the disclosed current cost estimates are value relevant (e.g., Beaver and Landsman, 1983; Beaver and Tyan, 1985; Bernard and Ruland, 1987). These studies attribute the lack of significant results to measurement errors of SFAS 33 disclosures. Later studies lend various degrees of support that disclosed fair values are value relevant and attribute the difference in findings to correlated omitted variables in model specification (Bublitz et al., 1985; Murdoch, 1986) or cross-sectional variation among sample firms (Hopwood and Schaefer, 1989; Bernard and Ruland, 1987). Another stream of studies investigates the value relevance of asset revaluations. Some accounting regimes, such as Australia, New Zealand, and the U.K., have long been permitting revaluation for tangible assets. Using data from these countries as the sample, prior studies provide some evidence that revalued accounting amounts are value relevant (e.g., Easton et al., SFAS 33 requires public companies meeting certain criteria to disclose a) income from continuing operations on a current cost basis; and b) the current cost amounts of inventory and property, plant and equipment. 6

7 1993; Barth and Clinch, 1998; Lin and Peasnell, 1998; Aboody et al., 1999). Using a sample of Australian industrial firms, Easton et al. (1993) find that asset revaluation reserves are significantly associated with price-to-book ratios and net increments to asset revaluation reserves have significant explanatory power for stock returns. However, after categorizing assets into investments, property, plant and equipment, and intangibles, Barth and Clinch (1998) find that revalued amounts for properties are only positively associated with stock prices for certain financial firms, whereas revalued amounts for plant and equipment are only value relevant for mining firms. Taken together, extant evidence shows that the value relevance of fair value information for tangible assets vary depending on the model specification, the type of assets being measured, and the characteristics of the sample firms. This study extends literature on the value relevance of fair value accounting by investigating the value relevance of investment property fair values under the IFRS regime.. Research Setting Prior studies mostly investigate the implications of tangible asset fair values in settings where fair values have marginal impacts to a firm s financial statement. This study examines the value relevance of fair value accounting in the context of the European real estate industry. A major advantage of employing this setting is that real estate firms are likely to have great exposure to investment properties and the accounting model for investment properties should be highly relevant in impacting the ability of financial statements to reflect a real estate firm s economic position and financial performance. Moreover, the accounting requirement for investment properties in European countries exhibit substantial variation surrounding the adoption of IFRS, which provides an opportunity to compare the relevance of historical cost versus fair value information. As shown in Table 1, prior to IFRS adoption, accounting requirements for investment properties can be broadly 7

8 categorized into three models: (1) the cost model: investment properties should be depreciated over their estimated useful life, with depreciation expense as well as any write-off arising from impairment tests recognized in the income statement 3 (e.g., Italy, France, Germany, Spain, Switzerland, and Poland); () the revaluation model: investment properties should be revalued to fair value on the balance sheet, with upward or downward fair value adjustments accumulated to equity (e.g., U.K., Denmark, and the Netherlands); and (3) the mixed model: investment properties are carried on the balance sheet at an amount which equals to the revaluation date fair value less any subsequent depreciation (e.g., Belgium, Greece, Finland, and Sweden). Figure 1 European Firms Accounting Model for Investment Properties Surrounding IFRS Adoption Note: This figure depicts the possible changes in firms accounting treatment methods for investment properties at the adoption of IFRS. HC represents historical cost; FV represents fair value. Explanation on the characteristics and accounting requirements for each model can be found in Section. Following the mandatory adoption of IFRS in 005, investment properties are required to be accounted for in accordance with IAS 40 Investment Property [003]. IAS 40 permits firms to choose between the fair value model and the cost model. The cost model prescribed under IAS 40 is similar to that required under some countries domestic GAAP. The difference is that firms selecting the cost model must disclose investment property fair values in the notes 3 Under the cost model, some firms may voluntarily disclose the fair values in the notes of financial statements. 8

9 to the accounts. The fair value model requires investment properties to be measured at fair value, with changes in fair value recognized in the income statement as unrealized gains and losses. Figure 1 depicts possible changes in firms accounting treatments for investment properties at the adoption of IFRS. In this study, based on a firm s accounting model for its investment properties prior to and after the adoption of IFRS, I categorize real estate firms into three groups: (1) firms applying the cost model prior to IFRS and continuing to use the cost model after IFRS (Group A); () firms applying the cost model prior to IFRS and switching to the fair value model after IFRS (Group B); (3) firms applying the revaluation model prior to IFRS and switching to the fair value model after IFRS (Group C). Taking advantage of the variation in the accounting model for investment properties surround IFRS adoption, in this study, I use Group A and Group B as the primary sample to investigate the value relevance of historical cost versus fair value information. More details for research design can be found in Section 3..3 Hypotheses Development Fair values for investment properties are mostly estimated based on unobservable inputs, such as rental streams, levels of occupancy, and discount rates (Muller et al., 011). These future-based estimates can be subject to uncertainties that are out of the company s control. Therefore, it is arguable that estimated investment property fair values are open to measurement error or even management manipulation. For example, using subsequent selling price as a benchmark for the true market value, Dietrich et al. (001) document a conservative bias for investment property fair values. They also find that managers increase investment property fair values prior to issuing debt. Landsman (007) argues that in dealing with the measurement error issue pertaining to fair value accounting, the key question is whether financial statements prepared based on fair value accounting improve the quality of information relative to those prepared under historical 9

10 cost accounting. Notably, historical cost accounting amounts for real estates are not free from bias. This is decided by the economic attributes of investment properties, as well as the accounting requirements of the cost model. Because of the asymmetric recognition of value appreciation and value consumption under the cost model, book value of properties decreases over time even in periods when property value increasees, which can introduce conservative biases to property accounting amounts. Due to increasing real estate values, the bias can become economically large over time. Dietrich et al. (000) find that the conservative bias contained in historical cost measurements is more severe than that contained in fair value measurements. Taken together, I propose the following hypothesis: Hypothesis 1: For investment properties, reported balance sheet numbers prepared under fair value accounting are relatively more value relevant than those prepared under historical cost accounting. Unlike balance sheet numbers, earnings prepared under fair value accounting are not necessarily more or less value relevant than those prepared under historical cost accounting. The main reason is that incorporating fair value gains and losses creates volatility to earnings, whereas restricted by the realization principle, historical cost accounting limits volatility and produces measures of earnings which are more stable. If volatility is perceived by investors to reflect the underlying economic risks of the firm, then fair value-based earnings are more informative than historical cost-based income. However, several reasons can lead to the consequence that the informativeness of volatility is unclear a priori. First, unrealized gains and losses can be transitory. If firms intend to hold properties for a long time, changes in the value of properties may reverse before disposal of asset and therefore provide inefficient information. Second, investment property fair values are mostly obtained from estimation. In situations where estimation error or management manipulation exist, unrealized gains and losses introduce artificial volatility to fair value-based earnings which distorts a firm s performance. Therefore, I propose the following hypothesis in the null form: 10

11 Hypothesis a: For real estate firms, income prepared based on fair value accounting is not relatively more value relevant than income prepared based on historical cost accounting. According to Barth et al. (1990), a certain earnings component might not be valued by the market if it is perceived to be transitory, lacking timeliness or subject to management manipulation. In the case of real estate firms, it is of interest to examine whether the unrealized earnings component, represented mainly by fair value adjustments of investment properties, are incrementally value relevant. Supporters of recognizing fair value gains and losses in the income statement are of the view that capital appreciation represents an essential part of a real estate firm s performance and therefore changes in property fair values should be viewed as profit or loss. Opponents hold the opinion that unrealized gains and losses are merely paper profit which can be transitory and may reverse in the future. Moreover, Barth (1994) suggests that if fair value estimation process is noisy, situation could occur where even if unrealized gains and losses are relevant to investors, errors arising from differential effects in estimating them will hinder researchers from documenting significant results. Therefore, I propose the following hypothesis in the null form: value relevant. Hypothesis b: For real estate firms, unrealized gains and losses are not incrementally 3 Research Design As indicated in Section., in this study, I use firms applying the cost model prior to IFRS and continuing to use the cost model after IFRS (Group A) and firms applying the cost model prior to IFRS and switching to the fair value model after IFRS (Group B) as the sample for empirical analyses. These two groups form a quasi-natural experimental setting to conduct difference-in-differences analyses on the implication of fair value accounting (e.g., Muller et al., 011; Goncharov et al., 015). Given that investment properties occupy a significant proportion of real estate firms assets, it is reasonable to assume that differences in the effects 11

12 of IFRS adoption on the two groups irrelevant to the choice between the cost model and the fair value model provided by IAS 40 are likely to be marginal. Accordingly, the option to switch accounting model for investment properties at the adoption of IFRS can be viewed as a treatment applied to the sample firms 4, where Group A can be viewed as a control group and Group B can be viewed as a treatment group. I employ this difference-in-differences setting to investigate the value relevance of historical cost and fair value accounting information. 3.1 Value Relevance of Balance Sheet Information Following prior research, I employ the balance sheet model to investigate the relative value relevance of fair value-based and historical cost-based balance sheet information (e.g., Landsman, 1986; Barth, 1991). I partition assets into investment property assets and other assets and restate the balance sheet model: MVE it Pre = α 0 + α 1 InvProp it Pre + α OthAssets it Pre + α 3 Liabilities it Pre + α 4 NumShares it Pre + ε it (1) MVE it Post = α 0 + α 1 InvProp it Post + α OthAssets it Post + α 3 Liabilities it Post + α 4 NumShares it Post + ε it () The dependent variable is MVE it, firm i s market value of equity at the end of the third month after fiscal year end t. I use market value of equity three months after fiscal year end to ensure the accounting information is available to the market. Independent variables include: InvProp it, firm i s recognized value of investment properties for fiscal year t; OthAssets it, firm i s other assets (i.e., total assets excluding investment properties) for fiscal year t ; Liabilities it, firm i s total liabilities for fiscal year t; NumShares it, firm i s number of shares outstanding at the end of fiscal year t. NumShares it is included as a scale proxy to mitigate 4 It is worth noting that the difference-in-differences setting in this study is not a strict quasi-experimental setting, because it is not a random allocation of the cost and fair value model among the sample firms. Rather, all firms are provided with a free choice between the cost model and the fair value model to account for investment properties at the adoption of IFRS. 1

13 scale differences in this level-based regression (Barth and Kallapur, 1996). For all the variables, Pre means that a certain variable is prepared in the pre-ifrs period in accordance with domestic GAAP and Post means that the variable is prepared following IFRS. [Insert Table ] I estimate each of Equations (1) and () using Group A and Group B separately. This procedure yields four R s (see Table ). Comparison of R s between Equation (1) and Equation () (i.e., R A_Pre versus R A_Post and R B_Pre versus R B_Post ) provide insights into the relative value relevance of balance sheet information prepared under domestic GAAP and that prepared under IFRS. Under the assumption that the effect of systematic changes in the institutional structures surrounding IFRS adoption is similar across firm in the control group (i.e., Group A) and firms in the treatment group (i.e., Group B), changes in the value relevance of accounting information upon IFRS adoption for the control group can be attributed to the effect of systematic changes of IFRS adoption, as measured by R A_Post R A_Pre. The effect of switching accounting model on the value relevance of balance sheet information can be measured by (R B_Post R B_Pre ) (R A_Post R A_Pre ). Following prior research (e.g., Joos and Lang 1994; Harris et al., 1994; Arce and Mora, 00; Tsalavoutas et al., 01), I statistically compare the R s of Equation (1) and that of Equation () using a test based on Cramer (1987). 3. Measurement Bias of Fair Value versus Historical Cost Measurements To investigate whether fair value measurements mitigate conservative bias contained in historical cost measurements, I augment the basic balance sheet model by adding two indicator variables, Post it and Switch i, and interacting each of them with InvProp it. In Equation (3) below, Post it equals to 1 if an observation belongs to the post-ifrs period, and 0 otherwise. Switch i equals to 1 if firm i applies the cost model pre-ifrs and switches to the fair value model post-ifrs, and 0 if firm i applies the cost model both pre-ifrs and post-ifrs. 13

14 MVE it = δ 0 + δ 1 InvProp it + δ Post it +δ 3 Switch i + δ 4 InvProp it Post it + δ 5 InvProp it Switch i + δ 6 Post it Switch i + δ 7 InvProp it Switch i Post it + δ 8 OthAssets it + δ 9 Liabilities it + δ 10 NumShares it + ζ it (3) In Equation (3), the coefficient on InvProp it captures the association between the accounting amounts of investment properties and market value of equity for firms in the control group pre-ifrs. The coefficient on InvProp it Switch i captures difference (if any) in this association between the control group and the treatment group prior to IFRS adoption. The coefficient on InvProp it Post it captures effects of IFRS adoption on this association for both the control group and the treatment group. If historical cost measurements contain conservative biases, it is expected that the valuation coefficients on accounting amounts prepared under the cost model are greater than 1. A statistically significant negative coefficient on InvProp it Switch i Post it suggests that switching to the fair value model mitigates conservative bias contained historical cost measurements, after controlling for the effect of systematic changes of IFRS adoption (captured by coefficient on InvProp it Post it ) and the difference between control group and treatment group pre-ifrs (captured by coefficient on InvProp it Switch i ). Meanwhile, an F-test of (δ 1 + δ 4 + δ 5 + δ 7 ) insignificantly different from one provides evidence that fair values recognized under IAS 40 are relevant and reliable. 3.3 Value Relevance of Income Statement Information I then examine the relative value relevance of historical cost-based net income and fair value-based net income using the returns model. Return Pre it = γ 0 + γ 1 NI Pre it + γ ΔNI Pre it + η it (4) Return Post it = γ 0 + γ 3 NI Post it + γ 4 ΔNI Post it + η it (5) 14

15 In Equations (4) and (5), the dependent variable is Return it, firm i s equity returns after adjustments of dividends for fiscal year t, measured from three months after year-end for fiscal year t 1 to three months after year-end for fiscal year t, plus dividends for fiscal year t. Independent variable is NI it, firm i s net income for fiscal year t. denotes annual change. All variables are deflated by the beginning-of-year market value of equity. The rationale underlying tests of the impact of accounting model on the value relevance of income is similar to that introduced in balance-sheet-focused tests in Section 3.1. Given that real estate firms earnings sources are composed mainly of rental income and capital appreciation, it is reasonable to assume that whether to recognize unrealized gains and losses in the income statement is the main difference in the effect of IFRS adoption on net income between the control group and the treatment group. I estimate Equations (4) and (5) using Group A and Group B separately, which also yields four R s. Changes in R for the control group, R A_Post R A_Pre, captures the confounding effects of IFRS adoption on the value relevance of net income. The effect of switching accounting model on the value relevance of net income amounts can be measured by (R B_Post R B_Pre ) (R A_Post R A_Pre ). Finally, for firms switching to the fair value model after IFRS adoption, I examine the incremental value relevance of earnings components (i.e., realized earnings and unrealized earnings 5 ) prepared under fair value accounting by restating the returns model: Return it = γ 0 + γ 1 Earnings_Realized it + γ Earnings_Unrealized it + γ 3 Earnings_Oth it + γ 4 ΔEarnings_Realized it + γ 5 ΔEarnings_Unrealized it + γ 6 ΔEarnings_Oth it + η it (6) In Equation (6), the dependent variable Return it is defined the same as before. Independent variables include: Earnings_Unrealized it, defined as firm i s unrealized gains 5 For the firms examined in this study, realized earnings consist mainly of net rental income and gains and losses from disposal of properties and unrealized earnings consist mainly of fair value adjustments for investment properties. 15

16 and losses on investment properties for fiscal year t; Earnings_Realized it, defined as firm i s earnings before interest and tax excluding unrealized gains and losses for fiscal year t; and Earnings_Oth it, firm i s net income minus earnings before interest and tax for fiscal year t. denotes annual changes of corresponding variables. A significantly positive coefficient on Earnings_Realized it (Earnings_Unrealized it ) suggests that realized earnings (unrealized gains and losses) are incrementally value relevant in explaining share returns. 4 Sample Selection and Sample Description 4.1 Sample Selection [Insert Table 3] Table 3 presents the sample selection process. I start with the Thomson Reuters Worldscope population of firms that are classified as either Real Estate Investment Services or Real Estate Investment Trusts in European countries (hereafter, real estate firms) and are in active status as at December 15, 006. To examine changes in the value relevance of accounting information upon IFRS adoption, the sample period spans 16 fiscal years (000 to 015), five years prior to mandatory IFRS adoption and ten years after mandatory IFRS adoption 6. I check Worldscope item Accounting Standards Followed to identify mandatory IFRS adopters, which brings an initial sample of 38 unique firms and 3,808 firmyear observations. I require annual reports for all firm-years to be available to identify a firm s accounting model for investment properties. After merging hand-collected variables obtained from annual reports with other variables obtained from Thomson Reuters Worldscope database, I exclude 633 firm-year observations with insufficient data. I also exclude 16 firm-year observations transiting from IFRS to domestic GAAP after mandatory IFRS adoption. Finally, because difference-in-differences analyses require each individual firm to exist in both the pre- 6 The sample year starts from 000 because the annual reports of the EU companies are generally not publicly available in electronic form prior to

17 IFRS and post-ifrs periods, I eliminate firms without sufficient data pre-ifrs. These procedures lead to a final sample of,036 firm-year observations (134 unique firms) from 15 European countries over the period 000 to Sample Distribution and Descriptive Statistics [Insert Table 4] Table 4 reports firms accounting model for investment properties surrounding IFRS adoption by each country. I define fiscal years 000 through to 004 as the pre-ifrs period and fiscal years 006 through to 015 as the post-ifrs period. As a transition year, fiscal year 005 has been excluded from the post-ifrs period. In summary, prior to IFRS adoption, 84 out of the 134 firms account for investment properties under the cost model and 50 of them employ the revaluation model 7. At the adoption of IFRS, almost all the 50 firms applying the revaluation model pre-ifrs preferred the fair value model. However, for the 84 firms applying the cost model pre-ifrs, only 36 of them remained on the cost model and the other 48 firms switched to the fair value model. Accordingly, the number of firms for difference-in-differences analyses is summarized in Table 5. Table 5 Sample Summary for Difference-in-Differences Tests Pre-IFRS (Post = 0) Post-IFRS (Post = 1) Total Control Group (Group A & Switch = 0) Measurement Base HC HC HC to HC Unique Firms (Firm-Year Observations) (160) (330) (490) Treatment Group (Group B & Switch = 1) Measurement Base HC FV HC to FV N of Unique Firms (Firm-Year Observations) (7) (457) (684) Note: This table summarizes the sample that will be used for difference-in-differences analyses. HC means that a firm employs the cost model to account for investment properties and FV means that a firm employs the fair value 7 Among the 18 pre-ifrs firm-year observations from countries where revaluation was permitted under domestic GAAP (i.e., Greece, Finland, Sweden, Switzerland, Poland, and Belgium), only 50 of them conducted revaluation and 45 of these 50 observations are from Belgium. To simplify research design, I classify pre-ifrs firm-years domiciling in countries permitting ad hoc revaluation (i.e., Finland, Greece, Sweden, Switzerland, and Spain) into the cost model category, and firms conducting periodic annual revaluations into the revaluation model category (i.e., Belgium). 17

18 model to account for investment properties. Number in each cell without parentheses represents number of unique firms and number with parentheses represents number of firm-year observations. Table 6 presents descriptive statistics, which can be interpreted in combination with Figure and Figure 3, which show sample firms average accounting amounts for investment properties and total assets over the sample period respectively. [Insert Table 6] Figure Sample Firms Book Value of Investment Properties Note: This graph shows the sample firms average book value of investment properties over fiscal years 001 to 015. All values are in millions of Euros. HCHC represents firms applying the cost model both pre- and post- IFRS; HCFV represents firms applying the cost model pre-ifrs and switching to the fair value model post-ifrs; FVFV represents firms applying the revaluation model pre-ifrs and remaining on the fair value model post-ifrs; DisFV represents the disclosed fair value amounts for firms applying the cost model under the IFRS regime. Figure 3 Sample Firms Book Value of Total Assets Note: This graph shows the sample firms average book value of total assets over fiscal years 001 to 015. All values are in millions of Euros. HCHC represents firms applying the cost model both pre- and post-ifrs; HCFV represents firms applying the cost model pre-ifrs and switching to the fair value model post-ifrs; FVFV represents firms applying the revaluation model pre-ifrs and remaining on the fair value model post-ifrs. 18

19 Figure and Figure 3 show that for firms switching from the cost model to the fair value model upon IFRS adoption, there is a substantial increase in the book value of investment properties and total assets around fiscal year 005, which is the year when firms switched from the cost model to the fair value model to account for investment properties. This increase is not observed in firms remaining on the cost model or firms applying the revaluation model pre- IFRS and applying the fair value model post-ifrs. The substantial increase around the year of IFRS adoption appearing only for firms switching the accounting model provides some evidence that the accounting amounts of investment properties is underestimated under the cost model compared to that provided under the fair value model. Moreover, it can be observed that fair values (both recognized and disclosed) decline during the period 007 to 009. However, the decrease is not observed for firms applying the cost model. Through reading annual reports, I find that that few of the firms applying the cost model recognize impairments losses on their investment properties during this period. An explanation is that increases in property value cannot be captured under the cost model. To trigger an impairment charge, the decreases in the market value of property (if any) during the crisis period should be sufficiently large and exceed past value increments that did not previously get recorded under the historical cost accounting system. Finally, over the sample period 000 to 015, investment properties occupy nearly 50 percent of total assets for firms in Panel A and around 70 percent of sample firms total assets for firms in Panel B. Figure and Figure 3 show that for all three groups of firms, trends in the changes of book value for total assets are consistent with that of book value for investment properties, indicating that the accounting amounts for investment properties have significant impacts on real estate firms financial statements. 19

20 5 Empirical Results 5.1 Value Relevance of Balance Sheet Information [Insert Table 7] Table 7 summarizes the R s based on estimating Equation (1) and Equation () and reports results of the tests that I conducted to compare the difference in the R s. I first Untabulated results of estimating Equation (1) and Equation () show that all coefficients have the expected signs, with coefficients on assets variables being positive and those on liabilities variables being negative. Panel A shows that for the control group (Group A), the R for the regression based on pre-ifrs firm-years is 0.95, whereas the R for the regression based on post-ifrs firm-years is Cramer s test indicates that the difference is statistically significant (z-statistic = -.185, p-value = 0.000), suggesting that balance sheet information prepared under the IFRS regime is relatively less value relevant than that prepared under domestic GAAP for firms in the control group. However, for the treatment group (Group B), the R for the regression based on pre-ifrs subsample is 0.951, while the R for the regression based on post-ifrs subsample is Although there is a slight decrease upon IFRS adoption, Cramer s test fails to reject the null hypothesis that they are equal with each other (z-statistic = ). Further, a comparison of the R for the control group and that of the treatment group in the pre-ifrs period shows that there is no significant difference between the two (0.95 for the control group and for the treatment group), suggesting that balance sheet amounts are of similar value relevance between the two groups prior to IFRS adoption. This provides some evidence to support the validity of the difference-in-differences research design, given that the rationale underlying difference-in-differences tests assumes that the control group and the treatment group are homogenous in nature prior to the treatment. 0

21 The difference in R between pre-ifrs regression and post-ifrs regression can be attributed to two aspects: (a) systematic changes in institutional structures surrounding IFRS adoption, which apply to both the control group and the treatment group; and (b) change in the accounting model for investment properties, which applies only to the treatment group. Therefore, the difference in the changes of R upon IFRS adoption between the two groups can be attributed to the latter explanation, measured by (R B_Post R B_Pre ) (R A_Post R A_Pre ). Therefore, Table 7 Panel A shows that switching from cost model to fair value model increases the explanatory power of balance sheet information by 8.4 percent. 5. Measurement Bias Test of Balance Sheet Amounts for Investment Properties [Insert Table 8] Table 8 summarizes the valuation coefficients of investment properties from estimating Equation (3). Un-tabulated regression results show that the coefficient on InvProp it Switch it is not significant at the conventional levels (coefficient = , t-statistic = 0.835), which is consistent with expectation since all sample firms accounted for investment properties at depreciated cost prior to IFRS. The coefficient on InvProp it Post it is significantly positive (coefficient = 0.01, t-statistic = 3.571), suggesting that the adoption of IFRS in general increases the valuation coefficient of accounting amounts for investment properties. Because the confounding effects of IFRS adoption applying to all observations has been captured by InvProp it Post it, InvProp it Switch it Post it efficiently captures changes in valuation coefficient arising from switching the accounting model for investment properties. The coefficient is significantly negative (coefficient = , t-statistic = -3.45), indicating that switching from the cost model to the fair value model decreases the valuation coefficient on the accounting amounts for investment properties. I also conduct F-tests to assess whether the valuation coefficients on investment properties prepared under different accounting models are statistically different from the 1

22 theoretical value of one. Table 8 Panel A shows that in the post-ifrs period, the valuation coefficient for historical cost amounts is significantly greater than the theoretical value of one (coefficient = 1.394, F-statistic = 3.79, p-value = 0.055), while the valuation coefficient for fair value amounts is insignificantly different from one (coefficient = 1.094, F-statistic = 0.88, p- value = 0.351). F-tests results provide evidence that fair value measurements are more aligned with the information implicit in firm valuation. 5.3 Value Relevance of Income Statement Information [Insert Table 9] Table 9 summarizes the R s from estimating Equation (4) and Equation (5) using the control group and the treatment group respectively. For all the comparisons of R s between the pre-ifrs and pre-ifrs regressions, R of the post-ifrs regression is significantly larger than that of the pre-ifrs regression, suggesting that net income prepared under the IFRS regime is relatively more value relevant than that prepared under domestic GAAP. For example, in Table 9 Panel A, the R for the control group in the pre-ifrs period is 0.113, while the R for this group in the post-ifrs period increases to Cramer s test shows that the difference is statistically significant (z-statistic =.657, p-value = 0.000). Results for the treatment group are similar, with R equalling to in the pre-ifrs period and 0.33 in the post-ifrs period. Under the assumption that the two groups are homogenous in nature pre-ifrs, the impacts of switching accounting model can be measured by the difference in changes of R upon IFRS adoption between the control group and the treatment group, denoted by (R B_Post R B_Pre ) (R A_Post R A_Pre ). Therefore, switching to the fair value model upon IFRS adoption increases the explanatory power of net income to share returns by 7.6 percent. [Insert Table 10] Table 10 reports results based on estimating Equation (6). Results in columns (1), (3) and (5) are estimated using the sample of firms whose domestic GAAP does not permit

23 revaluation (i.e., Group B). In particular, for Group B, coefficients on realized earnings are positive and statistically significant in both the non-crisis period and the crisis period (noncrisis period: coefficient = 3.003, t-statistic = 1.77; crisis period: coefficient = 1.600, t-statistic = 3.446). The results suggest that for firms from countries revaluation was not permitted, only realized earnings are value relevant and unrealized earnings are not associated with the information implicit in share returns. 6 Additional Analyses 6.1 An Alternative Setting: within-firm Tests on the Value Relevance of Historical Cost and Fair Value Accounting Information Under IAS 40, firms applying the cost model are required to disclose fair values in the notes to the financial statements, which enables me to conduct within-firm tests to reassess the results of my primary analyses. A major advantage of within-firm tests is that the sample firmyears act as their own control, thereby reducing the amount of interruptions arising from variance between individual firms or changes happening to a single firm over time. Regarding the value relevance of balance sheet information, I regress the basic balance sheet model based on historical cost amounts (recognized) and fair value amounts (disclosed) for investment properties respectively and compare the R-squares of the two regressions. Untabulated results show that the R of historical cost-based regression is 8.8 percent and the R of fair value-based regression is 85.5 percent. Vuong (1989) s test shows that the difference between the two is statistically significant (p-value = 0.000), which confirms the result from primary tests that accounting amounts for investment properties measured at fair value are relatively more value relevant than those measured at depreciated cost. To reassess the value relevance of earnings prepared under different accounting models, I use disclosed fair values to construct analogous amounts for unrealized gains and losses 3

24 arising from changes in the fair value of investment properties 8 :Fair Value Adjustments it = [InvProp_Disc it (InvProp_Rec it + AccuDepre it )] [InvProp_Disc i,t 1 (InvProp_Rec i,t 1 + AccuDepre i,t 1 )]. To ensure the earnings amounts are comparable, I use earnings before interest and tax as a proxy for historical cost-based earnings. Fair value-based earnings are calculated using earnings before interest, tax, depreciation and amortization plus the analogous amounts for fair value adjustments. Un-tabulated results show that the R for returns model using historical cost-based earnings as explanatory variable is and the R based on fair value earnings is However, Vuong (1989) s test fails to reject the null hypothesis that the two R s are equal (z-statistic = , p-value = 0.15). Therefore, results regarding the relative value relevance of historical cost-based and fair value-based income from the primary tests are only marginally supported by within-firm test results. I further investigate whether the analogous amounts of unrealized gains and losses are incrementally value relevant to stock returns after controlling for historical cost-based earnings recognized in the income statement. The regression model is structured the same as Equation (6), where realized earnings has been re-defined as earnings before interest and tax for fiscal year t and unrealized earnings has been re-defined as the analogous fair value adjustments of investment properties for fiscal year t. Earnings_Oth it is defined as firm i s net income minus earnings before interest and tax for fiscal year t. Consistent with that revealed from primary tests, un-tabulated results based on within-firm tests also show that coefficient on unrealized earnings is insignificant, suggesting that unrealized earnings amounts (analogous) are not useful in explaining share returns for the sample firms. 8 To simplify the construction of analogous earnings prepared under fair value accounting, I construct FV Earnings it based on the assumption that firm i has zero disposal gains and losses and impairments. Also, I use earnings before interest and tax instead of net income to avoid the impact of tax rate on earnings. 4

25 6. Impact of Custom of Revaluation Prior study finds that institutional environment impacts the value relevance of fair value information (e.g., Barlev et al., 007). Results from primary tests, as well as those from withinfirm tests, are all based solely on firms domiciled in countries where revaluation was not permitted pre-ifrs. It is arguable that the implications of fair value accounting may vary across countries with different customs of revaluation. First, countries have long been permitting revaluation are more likely to have established regulations for asset revaluation, which have the potential to reduce investors information processing costs. Meanwhile, investors in these countries are more likely to be familiar with revaluation therefore are of higher ability to process fair value information. Therefore, I compare the value relevance of fair value information prepared by firms domiciled in countries with different customs of asset revaluation. Regarding balance sheet amounts, I augment Equation () by adding an indicator variable, PermitFV i, and interacting it with InvProp it Post to differentiate the valuation coefficient on fair value amounts for investment properties prepared by the two types of firms. Un-tabulated results show that for firms whose domestic GAAP permits revaluation, the valuation coefficient is slightly lower than that of the base group and more approaches the theoretical value of one, which suggests that fair values prepared by firms whose domestic GAAP permitting revaluation contain less measurement error. Regarding income statement information, I regress Equation (6) using the sample of firms domiciled in countries permitting revaluation pre-ifrs compare the results with those obtained in Section 5.4. Results in Table 10 show that the value relevance of earnings components vary considerably between Group B (i.e., firms whose domestic GAAP not permitting revaluation) and Group C (i.e., firms whose domestic GAAP permitting revaluation). For Group B, only realized earnings are incrementally value relevant, whereas for Group C, only unrealized earnings are incrementally value relevant. Taken overall, both balance sheet- 5