# Fair values of equity and debt securities and share prices of property-liability insurers.

Introduction

This study assesses the relation between fair values of equity and fixed maturity debt securities and share prices of property-liability insurers. The results provide the first evidence on the valuation implications of financial statement disclosures regarding fair value estimates for the majority of assets held by insurers. This evidence is important because, beginning with financial statements for year-end 1994, the Financial Accounting Standards Board's (FASB) Statement of Financial Accounting Standards No. 115 (SFAS 115), Accounting for Certain Investments in Debt and Equity Securities, requires formal balance sheet recognition of fair values for considerably more investment securities than was previously the case. The new accounting standard could have a substantial impact on earnings, assets, and capital of property-liability insurers because investment securities comprise over 60 percent of the total assets of a typical property-liability insurer.

The economic consequences of balance sheet recognition of fair values for investment securities are debated widely. The controversy centers primarily on whether fair value estimates for investment securities are sufficiently reliable to be valuation-relevant. Advocates of fair value accounting, including former Securities Exchange Commission (SEC) Chairman Breeden, assert that recognition of fair values will enhance the relevance of information in financial statements (Wall Street Journal, January 8, 1992). Opponents, including Federal Reserve Board Chairman Greenspan and various representatives of the financial services industry, assert that estimates of fair values of thinly-traded securities will be unreliable and, thus, valuation-irrelevant (Wall Street Journal, November 8, 1990). In addition, opponents assert that fair value accounting will induce additional volatility in earnings and capital measures that does not reliably reflect underlying economic volatility, leading to inefficient capital allocation decisions by investors and excessive intervention by regulators.(1)

The FASB (1980) cites relevance and reliability as the two principal criteria for choosing among accounting alternatives. The FASB defines relevant information as that which is capable of influencing decisions, and reliable information as that which is verifiable. These two characteristics are clearly not independent. In this setting, fair values of key assets such as investments are likely to be considered relevant by investors, if disclosed fair values are sufficiently reliable.(2) Since fair values must be estimated for certain securities, estimation error could impair the value-relevance of related disclosures.(3) This study tests the relevance and reliability of fair value disclosures by analyzing their association with share prices of property-liability insurers.

Prior research on accounting for investments, which has focused primarily on banking, has shown that fair value estimates help explain bank share prices.(4) However, prior work does not detect a link between fair value gains and losses and bank stock returns, perhaps because of measurement error in changes in fair value estimates (Barth, 1994). The insurance industry is a more powerful setting to test the valuation implications of fair value disclosures, because bank investment portfolios only range from 14 to 21 percent of total assets and almost exclusively include fixed maturity securities.

We find that fair value disclosures for certain types of investment securities are important determinants of share prices and returns of property-liability insurers. In particular, our evidence suggests that property-liability share prices can be explained by fair values of equity securities, which have traditionally been recognized on the balance sheet, and fair values of U.S. Treasury securities, which have been disclosed in footnotes. We also find that fair value disclosures for other types of investment securities (e.g., municipal and corporate bonds and other debt instruments) do not explain share prices. Further, we observe that our sample of insurers tend to hold U.S. Treasury securities with less than five years to maturity and corporate and municipal bonds with more than five years to maturity.

Our results suggest that the reliability of fair value estimates for different types of securities affects the value-relevance of related disclosures. We detect a relation between share prices of insurers and fair values of investments, but only if the fair value estimates are closely related to market prices - for example, investments traded in more liquid markets. These results lend insight into factors that affect share prices of property-liability insurers and illustrate potential effects of fair value accounting.

The remainder of the article provides background by describing accounting for investments by property-liability insurers, describes the sample, presents our empirical tests and results, and provides concluding remarks.

Background: Accounting for Investment Securities

Prior to 1994, under generally accepted accounting principles, changes in fair values of the majority of securities held by a typical property-liability insurer were accounted for as follows. Changes in fair values of equity securities were recognized in the assets and equity sections of the balance sheet but not in the income statement. Changes in fair values of fixed maturity debt securities were not recognized but were disclosed parenthetically on the balance sheet or in a footnote.(5) Of course, realized gains/losses from security sales were recognized in the income statement.

Beginning with financial statements for 1994, SFAS 115 requires that firms classify their securities into trading, available for sale, and held to maturity portfolios. Accounting for changes in fair values of trading portfolios are formally recognized on the balance sheet and income statement. Under SFAS 115, securities in the held to maturity portfolio are accounted for at amortized cost, with fair values disclosed but not recognized. However, under SFAS 115, few securities qualify for this treatment because SFAS 115 makes restrictive the condition that a firm must have the ability and intent to hold these securities to maturity.(6) Securities that do not qualify for the held to maturity portfolio (e.g., securities that are being held indefinitely) are to be classified as available for sale. For securities considered available for sale, SFAS 115 requires formal balance sheet recognition at fair value, with unrealized gains/losses recognized in the owners' equity section.

The most substantive change for property-liability insurers is that SFAS 115 will for the first time require explicit balance sheet recognition of unrealized gains/losses on most fixed maturity securities. In contrast, the insurance industry is a rare setting in which SFAS 115 will not materially change the way in which equity securities are accounted for in the asset section of the balance sheet. Property-liability industry accounting practice has historically included explicit balance sheet recognition of unrealized gains/losses on equity investments. Thus, this setting provides an opportunity to investigate the value-relevance of recognized fair values of equity investments prior to the adoption of SFAS 115.

Sample Selection and Data Description

The sample comprises 56 publicly-held property-liability insurers operating during 1985 through 1991, including 41 property-liability insurers on the 1990 or 1992 Compustat industrial tapes (primary, secondary, tertiary files).(7) An additional 15 firms meeting our sample criteria were identified from a report on publicly-held property-liability insurers by a management consultant and actuarial firm (Tillinghast, 1988). The sample represents greater than 50 percent of the total premiums earned by the industry.

Accounting data related to the investment portfolios of our sample firms were hand-collected directly from annual reports. Data on stock prices, stock splits and stock dividends were collected from both Standard and Poors' Daily Stock Price Records and Compustat. Data on stock returns were obtained from the Center for Research in Security Prices' monthly (NYSE and AMEX firms) or daily (NASDAQ firms) return files.

Financial statements of property-liability insurers indicate that investment portfolios comprise an important component of their economic value. Table 1 reports descriptive statistics of the 344 insurer-year observations in our sample. The book values of equity investments (carried at fair values) and fixed maturity investments (carried at amortized cost) comprise 7.6 percent and 55.4 percent, respectively, of total assets for the mean sample insurer.

[TABULAR DATA FOR TABLE 1 OMITTED]

The statistics shown in Table 1 also indicate that, on average from 1985 through 1991, fair values of equity and fixed maturity investments exceeded their respective book values by 3.2 percent and 1.5 percent of stockholders' equity. Corresponding standard deviations suggest that considerable variation exists across insurers. For example, the difference between fair value and historical cost of equity securities held by the insurer in the 90th (10th) percentile of this distribution was equal to 15.4 percent (-3.3 percent) of stockholders' equity. Likewise, the difference between fair value and book value of fixed maturity securities held by the insurer in the 90th (10th) percentile of this distribution was equal to 15.9 percent (-7.7 percent) of stockholders' equity.

As noted earlier, opponents of fair value accounting assert that fair value estimates may be less reliable and therefore less value-relevant for securities that are not widely traded or are intended to be held to maturity. Tables 1 and 2 present data on the types and maturity distributions of securities held by our sample insurers. For the 282 insurer-years with disclosures about the issuers of their fixed maturity investments, the mean insurer invested approximately 40 percent and 25 percent of the total fixed maturity portfolio in municipal and U.S. Treasury securities, respectively, with the remainder in corporate and other debt instruments (see Table 1).

Table 2 includes correlations between proportions of the fixed maturity portfolio with different maturities and proportions of the fixed maturity portfolio from different types of issuers, for the 112 sample insurer-years with disclosures of types and maturities of investments. Based on this subsample, these correlation coefficients suggest that insurers that hold shorter-term fixed maturities (less than one year and one to five years) invest more heavily in U.S. Treasury securities. In contrast, insurers holding longer-term maturities (from five to ten years and over ten years) invest more heavily in municipal and corporate securities. No discernible pattern of correlation exists between maturity and percentage of investment in other fixed maturity securities.

[TABULAR DATA FOR TABLE 2 OMITTED]

Empirical Tests

If disclosed fair value estimates measure underlying fair values of investments with a sufficiently high degree of error, then these disclosures are not likely to be considered reliable information and, thus, are not likely to be related to share prices or returns. Alternatively, we predict that fair values of both equity and fixed maturity investments are positively related to share prices, and changes in fair values are positively related to returns. If so, the results can be interpreted as evidence that fair values are sufficiently reliable as to be considered valuation-relevant.(8)

Certain investments held by property-liability insurers are likely to be actively traded in securities markets (e.g., Treasury securities). Fair value disclosures for such investments are based on readily observable market prices. Other types of investments (e.g., municipal and corporate bonds), involve a mixture of thickly and thinly traded securities. Some municipal and corporate issues are actively traded (such as general obligation bonds of large cities and states and corporate bonds of large firms), while others are not (such as obligations of small municipalities or corporations). Thus, fair value disclosures for portfolios of municipal and corporate securities likely require more estimation than disclosures for portfolios of U.S. Treasury securities. We provide evidence on the reliability of these disclosures by examining explicitly the share price impact of fair values that are estimated with more (or less) error across different types of securities.

Value-Relevance of Fair Values of Investment Securities

Like all other firms, the market value of a property-liability firm (MV) is a function of the market value of assets (MVA) less the market value of liabilities (MVL). Under the assumption of perfect and complete markets, this valuation relation for firm i at time t can be expressed as [MV.sub.it] = [MVA.sub.it] - [MVL.sub.it]. Under the accounting identity, the book value of equity in a firm (BV) is equal to the book value of assets (BVA) less the book value of liabilities (BVL): [BV.sub.it] = [BVA.sub.it] - [BVL.sub.it]. Thus, the difference between market and book value of equity can be expressed as a function of the differences between market and book values of assets and liabilities, or

[MV.sub.it] = [BV.sub.it] + ([MVA.sub.it] - [BVA.sub.it]) - ([MVL.sub.it] - [BVL.sub.it]) (1)

(see Barth, Beaver, and Landsman, 1994, and others for very similar approaches to modeling the relations between market and book values).

To test for an association between share prices and fair values of equity and fixed maturity investments, we estimate the following specification of model (1), pooled across 56 sample insurers and the years 1985 through 1991:(9)

[PRICE.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][NBV.sub.it] + [[Beta].sub.2][UCLAIMS.sub.it] + [[Beta].sub.3][ES_HC.sub.it] + [[Beta].sub.4][ES_FVOHC.sub.it] + [[Beta].sub.5][FM_BV.sub.it] + [[Beta].sub.6][FM_FVOBV.sub.it] + [[Beta].sub.7][lnASSETS.sub.it] + [[Epsilon].sub.it] (2)

PRICE denotes price per share of common stock as of March 31 in year t+1.(10) This specification assumes that the relations are constant across sample firms and over this study period, which is reasonable because property-liability insurers have relatively homogeneous asset and liability structures and production functions. Year-specific intercepts are included to control for systematic changes in levels of stock prices in the insurance industry over the sample period.

The focus of this study is to test whether fair value estimates of investments provide value-relevant information incremental to historical costs. Thus, the key variables of interest in this regression are the difference between fair value and historical cost of equity investments (ES_FVOHC), and the difference between fair value and book value of fixed maturity investments (FM_FVOBV). We predict the coefficients on both ES_FVOHC and FM_FVOBV to be positive.(11) Similarly, historical costs of equity securities (ES_HC) and book values of fixed maturities (FM_BV) also should be positively related to share prices.

The variable NBV denotes the net book value of common equity before investments in equity and fixed maturity securities and before the liability for unpaid claims. This variable is included to control for the market's valuation of all net assets not separately identified in our model.

The variable UCLAIMS represents the liability for unpaid claims. This variable controls for potential differences in valuation across insurers with different levels of outstanding claims, which is typically an insurer's largest liability (representing 67 percent of total liabilities and 49 percent of total assets on average). Ideally, we would include a regressor that reflects fair values of liabilities of our sample firms. In particular, changes in fair values of liabilities that hedge changes in fair values of investments are potentially important for firm valuation decisions. Unfortunately, such measures are not disclosed by property-liability insurers (and SFAS 115 does not require recognition); thus, fair values of liabilities are not included in our model.(12)

Accounting practice of property-liability insurers recognizes explicitly the impact of unexpected changes in inflation on the liability for outstanding claims (UCLAIMS).(13) UCLAIMS is reported on the balance sheet at the nominal amount expected to be paid (i.e., a future value), with the change in the nominal value included as a component of income.(14) Fair value of UCLAIMS would equal the present value of the reported nominal value of this liability. Whether UCLAIMS controls for fair values of this liability depends on the extent to which discount rates applied by investors to future claim loss cash flows change with unexpected inflation in those cash flows. If discount rates applied by investors increase to reflect unexpected inflation in future claim loss cash flows, then the present value of this liability remains the same (or may actually fall), and including UCLAIMS in our tests of share prices does not control for changes in fair values of liabilities. Alternatively, if discount rates increase to only partially reflect unexpected inflation in future claim loss cash flows, then UCLAIMS may partially control for changes in fair values of the liability.(15) Because discount rates used by investors are unobservable, it is difficult to assess the impact of changes in fair values of liabilities on our tests.

The natural logarithm of total assets (lnASSETS) is included as a control for size and potential heteroskedasticity. After including lnASSETS, the White (1980) chi-square test statistic fails to reject the null hypothesis of homoskedastic error terms and correct model specification (p = 0.37). Without this control, the model fails White's test (p = 0.07). The results on our variables of interest, however, are not changed by the inclusion of this control.

Model (2) is estimated using ordinary least squares over 342 insurer-years (see Table 3, Panel A).(16) The regression model has an adjusted [R.sup.2] of 67 percent. As expected, fair values of investments are important explanatory variables for insurer share prices. The coefficients on ES_FVOHC and FM_FVOBV are positive and significant at the 0.001 level.(17) In addition, the coefficients on the historical cost of equity securities and the book value of fixed maturity investments are both reliably positive. The book value of common equity before investments (NBV) is positively associated with share prices, and the liability for unpaid claims (UCLAIMS) is negatively associated with share prices.(18)

Because serial correlation in residuals may be a problem with a pooled regression of price levels, we also estimate the model on a yearly basis. The results, not reported here, are similar to those reported above. The coefficient on ES_FVOHC is positive in all seven years and significant at less than the 0.05 level in six years. A Z-statistic on the pooled significance of the yearly t-statistics is 3.48 (p = 0.0009).(19) The coefficient on FM_FVOBV is positive in all seven years and significant in four years at less than the 0.05 level. The Z-statistic is 2.78 (p = 0.008).

These results indicate that fair value estimates of investment securities are reflected in the share prices of property-liability insurers. These findings hold for fair value estimates of equity securities, which are explicitly recognized on insurer balance sheets, and for fair value estimates of debt securities, which are disclosed.

Fair value disclosures for certain types of fixed maturity investments (e.g., securities traded in liquid markets) are likely to have more explanatory power for share prices than fair values of other securities. Thus, we expand model (2) to estimate separately the incremental value-relevance of fair values for each of the four commonly disclosed categories of fixed maturity securities. The expanded regression model includes both the book value and the difference between fair value and book value for U.S. Treasury securities (US_BV and US_FVOBV), municipal bonds (MUNI_BV and MUNI_FVOBV), corporate bonds (CORP_BV and CORP_FVOBV), and other fixed maturity investments (OTH_BV and OTH_FVOBV).

Regression results based on 265 insurer-years (51 insurers over 1985 through 1991) are presented in Table 3, Panel B.(20) The standard errors and t-statistics reported are the result of ordinary least squares because the White (1980) chi-square test-statistic is not significant at conventional levels (p = 0.29). The coefficients on the equity security variables and the control variables are similar to those found in the simpler model. The coefficients on the four variables capturing book values of fixed maturity investments are all positive and significant as predicted, but none of the four are statistically equivalent to one. In contrast, of the four fair value-based variables, only the coefficient on US_FVOBV is significant and positive at the 0.05 level. The coefficients on MUNI_FVOBV, CORP_FVOBV and OTH_FVOBV are not significant. These results indicate that book values are value-relevant for all types of fixed maturity investment securities, but fair values are value-relevant only for U.S. Treasury securities.

We also conducted several tests of the robustness of the regression results reported in Table 3, Panel B. First, to address potential effects of serial correlation of residuals we obtained separate yearly estimates of a model similar to the expanded regression. The results are qualitatively similar to those reported above. Coefficients on variables capturing market values of equity securities and U.S. Treasury securities are consistently positive and significant in a majority of sample years. Z-statistics on the pooled significance of the yearly t-statistics are significant at conventional levels.

Second, to test whether the results are sensitive to differences across finns in risk, financial health, or other firm-specific factors, we reestimated the model twice, including firm-specific intercepts and a numerical version of A.M. Best's financial condition rating (on a reduced sample of 246 insurer-years).(21) The results under both procedures are similar to those reported above. However, the coefficient on US_FVOBV becomes less significant (t = 1.72, p = 0.0435) when the model includes firm-specific intercepts (requiring 50 additional parameter estimates).(22)

These results suggest that investors consider fair value disclosures related to equity and U.S. Treasury securities to be more reliable and valuation-relevant than similar disclosures for municipal and corporate bonds and other debt instruments. A potential explanation for this is that investors consider fair value estimates for equities and Treasury securities to be more reliable because they are more likely to be based on readily observable market prices.(23)

Value-Relevance of Changes in Fair Values of Investment Securities

Given that we find that fair values help explain share price levels, changes in fair values of investments should be important determinants of returns. Barth (1994), however, does not observe a consistent relation between returns and fair value gains and losses in the banking industry. Barth's results suggest that measurement error in fair value gains and losses may be sufficiently large as to overwhelm the relation across a sample of banks. To test this possibility in the insurance setting, we estimate the following model pooled across 55 sample insurers and the years 1986 through 1991:

[Return.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][Delta][EPS_BRGL.sub.it] + [[Beta].sub.2][RGL.sub.it] + [[Beta].sub.3][Delta][UGL.sub.it] + [[Epsilon].sub.it], (3)

where Return denotes return on common stock in sample firm i from April 1 in year t to March 31 in year t+1.(24) Yearly intercepts are included to control for systematic changes in stock prices over time.

To control for returns associated with income before gains and losses on investment securities, the model includes changes in primary earnings per share before realized gains and losses (after taxes) and before discontinued operations and extraordinary items ([Delta]EPS_BRGL).(25) We predict that the coefficient on [Delta]EPS_BRGL is positive. If changes in earnings are expected to be permanent, then, in an ideal setting (e.g., perfect and complete markets), this coefficient would equal the reciprocal of the cost of capital (Miller and Modigliani, 1966).

Realized gains and losses on investments after taxes (RGL) are included to capture returns associated with sales of investments. Changes in unrealized gains and losses on equity and fixed maturity investments ([Delta]UGL) are included to test whether returns are associated with changes in fair values of investment securities. [Delta]UGL is calculated as the change in the excess of fair value over historical cost for equity securities plus the change in the excess of fair value over book value for fixed maturity securities.(26) In this regression, the coefficient on [Delta]UGL should capture the value-relevance (if any) of changes in the fair value of the securities portfolio that are incremental to realized gains and losses. We predict that this coefficient should be positive. This specification assumes that expectations for realized gains and losses and changes in unrealized gains and losses are zero over the pooled cross-section, which is plausible and consistent with prior work (e.g., Barth, 1994).(27)

Ordinary least squares regression results based on 274 insurer-years are reported in Table 4, Panel A. White's (1980) test does not reject homoskedasticity and correct model specification (p = 0.29). The model has an adjusted [R.sup.2] of 37 percent. As expected, the coefficient on [Delta]EPS_BRGL is positive and significant. This coefficient is statistically equivalent to one, consistent with the possibility that changes in earnings in this sample are on average expected to be transitory rather than permanent.

The coefficient on RGL is positive and significant at the 0.01 level, and is statistically equivalent to one. This is contrary to previous research. Warfield and Linsmeier (1992) and Barth (1994) observe that the association between bank stock returns and realized gains and losses on investment securities is negative on average. Both studies suggest that the findings are dominated by stock price increases associated with firms that selectively realize losses to reduce taxes.(28) To provide evidence on this possibility, we examine separately returns associated with four subsamples: high versus low marginal tax rate insurers (see footnote 11) with net realized gains versus net realized losses. Our findings from these tests (untabulated) indicate no significant negative coefficients on any of the four subsamples. In fact, we obtain a significant positive coefficient on the high tax/net realized losses subsample that appears to drive the results in prior work.(29)

The coefficient on [Delta]UGL is positive and significant (p = 0.01), indicating changes in fair values of investments are value-relevant in the property-liability insurance industry. This result is not consistent with results based on the banking industry (Barth, 1994). It appears that the insurance industry is a more powerful setting for this test.

Results from our stock price regressions suggest that investors price fair values for certain types of investments (e.g., equities and Treasury securities) more than other types. Ideally, we would test the information content in changes in fair values for equity securities and each of the four categories of fixed maturity securities. This test requires data on realized security gains and losses after taxes for each category. Otherwise, results would be difficult to interpret because the predicted response to changes in unrealized gains and losses of each category should be conditional on the realized gains and losses of that category. Unfortunately, insurers do not separately disclose realized gains and losses after taxes for the equity and four types of fixed maturity securities.

Insurers generally do, however, disclose separately realized gains and losses before taxes for portfolios of equity and fixed maturity securities (but not by fixed maturity category). This allows us to estimate realized gains and losses after taxes for the two portfolios and thus test the association between returns and the change in unrealized gains and losses on the equity and fixed maturity portfolios separately using an expanded form of model (3). In place of RGL, the expanded model includes after-tax estimates of the three disclosed components of RGL: equity securities (ES_RGL), fixed maturity securities (FM_RGL), and all other investments - e.g., mortgages and real estate - (MISC_RGL). The tax effects of the realized gains and losses are estimated based upon the tax position of the insurers as disclosed in the footnotes to the financial statements (see footnote 11). Our tax estimates are reasonable but add noise to the regression model since the Pearson correlation coefficient between the sum of ES_RGL, FM_RGL, and MISC_RGL, and the total after-tax realized gains and losses as reported in the financial statements (RGL) is 0.83. In addition, instead of [Delta]UGL, the expanded model includes two disclosed components: the change in the excess of the fair value over historical cost of equity securities (ES_[Delta]UGL) and the change in the excess of the fair value over book value of fixed maturity securities (FM_[Delta]UGL).

Regression results based on 270 insurer-years (55 insurers from 1986 through 1991) are reported in Table 4, Panel B.(30) White's test does not reject homoskedasticity and correct model specification (p = 0.50). The model has an adjusted [R.sup.2] of 36 percent. Again, the coefficient on [Delta]EPS_BRGL is positive and significant. As expected, the coefficients on realized gains and losses on equity securities and changes in unrealized gains and losses on equity securities are both significantly positive at the 0.05 level. The coefficients on realized gains and losses on fixed maturities and changes in unrealized gains and losses on fixed maturities are positive but not significant.

The results reported in Table 4 are robust to several specification tests. First, the results are similar when the models include firm-specific intercepts and lnASSETS to control for firm-specific differences such as risk or size. Separate yearly estimates of each model also produce similar results. Z-statistics indicate that pooled annual t-statistics are significant at conventional levels for coefficients on [Delta]UGL and ES_[Delta]UGL but not for coefficients on FM_[Delta]UGL.

The results indicate fair value gains and losses on equity securities in particular help explain returns for property-liability insurers, but fair value gains and losses on fixed maturity securities do not. These results are not necessarily inconsistent with our other findings. Our ability to detect a relation between returns and changes in unrealized gains and losses on fixed maturity securities is limited because changes in unrealized gains and losses on fixed maturities reflect municipal, corporate, and other fixed maturity investments, which were not found to be value-relevant in our share price regression results.

Conclusion

This research enhances our understanding of how property-liability insurers are valued and contributes to the ongoing debate concerning the relevance and reliability of fair value accounting information for investment securities and other financial assets and liabilities. We find that, although book values for all categories of investments are valuation-relevant, fair values for only certain categories of investment securities are reflected in share prices and returns of property-liability insurers. It appears that fair values for certain categories of investments, such as equities and U.S. Treasury securities, which are more likely to be traded in active markets, are valuation-relevant. Share prices are not explained by fair values for other types of investments - e.g., municipal and corporate bonds and other debt instruments - which may be less actively traded and which tend to have longer terms to maturity than U.S. Treasury securities in this sample.

One interpretation of these findings is that concern over application of fair value accounting techniques is warranted, especially for assets and liabilities without readily observable market prices. Alternatively, one can also view these results as evidence that sophisticated investors discriminate between fair value estimates of different assets and rely only on fair value disclosures that are more closely aligned with readily observable market prices.

The authors appreciate comments from M. Barth, J. Hand, W. Landsman, T. Linsmeier, R. Pfeiffer, T. Warfield, two anonymous reviewers, the editor, and workshop participants at the University of Georgia, Michigan State University, the University of Oregon, Penn State University, and the 1994 meeting of the American Accounting Association. The authors are grateful for the research assistance of Norman Godwin, Luann Lynch, and Daqing Qi and financial support from Michigan State University, the Center for Financial and Accounting Research at the University of North Carolina, and the KPMG Peat Marwick Research Fellowship program.

1 However, there is no indication that statutory accounting principles (SAP) will change as a result of SFAS 115. Under current SAP requirements, equity securities are reported at fair value, and fixed maturity securities are generally reported at amortized cost.

2 However, certain opponents of fair value accounting argue that fair value disclosures are irrelevant for debt securities that are to be held to maturity, on the basis that all contractual cash flows will be received over the life of the instrument; thus, changes in fair values are temporary. Evidence in Barth, Landsman, and Wahlen (1995) does not support this argument.

3 The FASB adopted the term "fair value" instead of market value to encompass estimated values for securities that are not traded in active markets.

4 For example, samples of banks have been used to examine the value-relevance of fair value disclosures in Barth, Beaver, and Wolfson (1990), Warfield and Linsmeier (1992), Ahmed and Takeda (1993), Barth (1994), and Barth, Landsman, and Wahlen (1995). Foster (1975, 1977) used a sample of 22 insurers to examine the value-relevance of total gains and losses on equity securities. A sample of closed-end mutual funds is examined in concurrent research on the relevance of fair values by Carroll and Linsmeier (1994).

5 Technically, SFAS 60, effective in 1982, requires that fixed maturity securities held in trading portfolios be recognized separately and accounted for at fair value. Curiously, a separate trading portfolio disclosure does not appear on any of our sample firms' balance sheets until such disclosure is provided by one firm in 1990 and three firms in 1991; only six sample insurers provide footnote disclosures about trading portfolios during this period.

6 The intent of SFAS 115 and current SEC enforcement practice indicate that this is a stringent condition. For example, although no formal position has been issued, the SEC has, through the use of comment letters, already begun to require many financial institutions to classify some or all of their investment portfolio as available for sale (Price Waterhouse, 1994). Aetna Life and Liability Co., for instance, which adopted SFAS 115 in 1993 (one year early), classified less than 10 percent of all equity and debt securities in their held to maturity portfolio.

7 This sample was constructed originally by Anthony and Petroni (1994). The set of firms used here includes all insurers in Anthony and Petroni (1994) that are classified primarily as property-liability insurers (e.g., a primary Standard Industrial Classification code of 6331 for fire, marine, and liability insurance) and are considered to be actively traded (e.g., share price changes on greater than 50 percent of the trading days in the sample period). Three insurers on Compustat are omitted from the sample because they are not actively traded, and one is omitted because the stock was traded for less than one year.

8 In addition, recognition of unrealized gains/losses could lead to contractual implications (e.g., regulatory intervention or binding debt covenants, or greater scrutiny by auditors) that could affect share prices (cf., Barth, Landsman, and Wahlen, 1995). Opponents of fair value accounting also have asserted that these implications could alter investment behavior of financial institutions, making them more reluctant to hold securities with volatile fair values.

9 All variables (except lnASSETS) are stated on a per share basis to reduce the presence of heteroskedastic error terms. All per share data are adjusted for stock splits and stock dividends.

10 Stock prices as of March 31 in year t+1 are used because all sample insurers have December 31 fiscal year ends, and disclosures regarding investments are not likely to be public until the release of the annual report. Because all other variables are measured as of December 31 in year t, using March 31 in year t+1 stock prices adds noise to our tests.

11 More precisely, under ideal estimation conditions, the coefficients should equal one minus the present value of the expected future marginal tax rate, because appreciation/depreciation in market values of securities does not explicitly affect taxes of an insurer until realized. The reported results are based on differences between market and book values measured on a before-tax basis. To examine the effects of taxes on these results, however, tax implications of differences between market and book values are estimated using the statutory tax rate and the tax status of the insurer. We employ the standard technique of assuming that the insurer faces a zero marginal tax rate if it has a net operating loss or investment tax credit carryforward; otherwise, the marginal rate is the statutory rate (see, e.g., Scholes, Wilson, and Wolfson, 1990). The results are similar using either a before- or after-tax basis.

12 Prior research on the association between fair values in investments and share prices generally has omitted changes in the fair value of liabilities, potentially biasing their coefficients on fair value-based variables toward zero. Armed and Takeda's (1993) analysis includes a variable that reflects net assets maturing or repricing after one year as a proxy for interest-rate sensitivity. Their results indicate that this variable is an important factor affecting bank stock prices; however, it does not change their conclusion that realized and unrealized security gains and losses are positively related to bank stock returns. Also, in concurrent research, Carroll and Linsmeier (1994) show that fair value disclosures for both equity and debt securities are value-relevant for share prices and returns of closed-end mutual funds, which are all-equity firms that invest exclusively in securities and thus do not have any omitted correlated variable problems.

13 Unearned premiums is the second largest liability of a property-liability insurer (representing 22 percent of total liabilities and 16 percent of total assets). It is a short-term liability reflecting prepaid premiums and is not significantly influenced by unexpected changes in interest rates. Thus, book value approximates fair value. The reported statistics are based on 1987 stock property-liability industry totals from A.M. Best Company (1988).

14 Weiss (1985) and Petroni (1992) find evidence that the liability for outstanding claims is subject to management manipulation.

15 Evidence in Petroni and Beasley (1996) suggests that the level of the liability for unpaid claims relative to total liabilities is a proxy for the length of the claim cycle (i.e., the lag between the incurrence of a loss and the final claim settlement), which is likely to be a good proxy for the interest rate sensitivity of the liabilities of an insurer. To further control for interest rate sensitivity of the liabilities of an insurer (as a potential correlated omitted variable), all models were also estimated including the ratio of unpaid claims to total liabilities as an independent variable. This additional control variable is not included in our tabulated results, however, because this ratio is never significant and the results on all other variables are unchanged.

16 Conventional influence diagnostics reveal the presence of two observations with an undue influence on this test. The reported results are based on a sample that excludes these two observations. Similar results are obtained when these observations are included.

17 All reported probabilities on t-statistics are for one-tailed tests if the sign of the association is predicted; otherwise, two-tailed probabilities are reported.

18 Some insurers in our sample also write some life insurance. Life insurers tend to have longer-term liabilities than property-liability insurers. To test for possible differences across firm types, we reran all of our regressions on the subset of 40 insurers that write only property-liability insurance. The results from this subsample are similar to those from the full sample.

19 As in Barth (1994) and Rosenthal (1991, pp. 89-109), the Z-statistic is computed as follows:

[Mathematical Expression Omitted].

20 Sixty-two observations are omitted from this test because they do not meet additional data requirements, and 14 observations are omitted because the footnote disclosure and the balance sheet do not agree. Three observations are withheld because conventional influence diagnostics (e.g., Belsley, Kuh, and Welsch, 1980) show that they have an undue influence on this test.

21 A.M. Best is the leading supplier of ratings of the financial condition of insurers. To quantify the letter rating assigned yearly by Best to each firm, we convert it to a number such that a rating of A+ is assigned a 1, A is assigned a 2, ... C is assigned an 8, and Below C is assigned a 9. The rating is not found to have explanatory power in our model.

22 We also examined the sensitivity of the results reported in Table 3, Panel A using the same checks. The results for both sensitivity checks are similar to those reported above; however, when 55 firm-specific intercepts are included, the coefficient on FM_FVOBV becomes insignificant (t = 0.52). These additional coefficients reduce the efficiency of the model, and firm-specific intercepts appear to capture some of the firm-specific effects of FM_FVOBV in this specification.

23 Ideally, we would test this explanation by examining whether valuation implications differ across types of securities with fair values that are based on readily available market prices vs. estimates. Unfortunately, we do not have data on the availability of market prices of the specific investments held by these sample insurers.

24 All variables are stated on a per-share basis, scaled by price as of April 1 in year t (Christie, 1987), with per-share data adjusted for stock splits and stock dividends.

25 This specification assumes that [Delta]EPS_BRGL follows a random walk, which is consistent with prior research. Results are qualitatively similar when the level of earnings per share is used.

26 Changes in unrealized gains and losses of insurers are not taxed until realized. Thus, the reported results are based on changes in unrealized gains and losses on a before-tax basis. The results are similar, however, whether changes in unrealized gains and losses are used on a before- or estimated after-tax basis.

27 Of course, expectations for realized gains and losses and for changes in unrealized gains and losses may not be zero under certain scenarios (e.g., investors may form nonzero expectations for realized gains and losses for certain firms in years they face low marginal tax rates or low levels of regulatory capital).

28 The results in Warfield and Linsmeier (1992), however, do not hold for gains/losses realized in the fourth quarter, when year-end tax-planning activities are most likely to occur. As they note, this result could be due to noise associated with realized gains and losses due to fourth-quarter earnings or regulatory capital management.

29 Warfield and Linsmeier (1992) examine two-day returns, which may have more power to detect market reactions to gains/losses realized for tax purposes. Additional explanations may be found in differences in tax or regulatory treatments across banks versus insurers and across study periods.

30 Four insurer-years are excluded in the expanded model regression because the footnote disclosures do not separate realized gains and losses between the equity and fixed maturity investment portfolios.

References

Ahmed, A. and C. Takeda, 1993, Stock Market Valuation of Gains and Losses on Commercial Banks' Investment Securities: An Empirical Analysis, Unpublished manuscript, University of Florida, Gainesville.

A. M. Best Company, 1988, Best's Aggregates and Averages (Oldwick, N.J.: A. M. Best).

Anthony, J. and K. Petroni, 1994, Accounting Estimation Errors and Firm Valuation, Unpublished manuscript, Michigan State University, East Lansing.

Barth, M., 1994, Fair Value Accounting: Evidence from Investment Securities and the Market Valuation of Banks, Accounting Review, 69: 1-25.

Barth, M., W. Beaver, and W. Landsman, 1994, Value-Relevance of Banks' Fair Value Disclosures under SFAS 107, Unpublished manuscript, Harvard University, Cambridge, Massachusetts.

Barth, M., W. Beaver, and M. Wolfson, 1990, Components of Bank Earnings and the Structure of Bank Share Prices, Financial Analysts Journal, 46: 53-60.

Barth, M., W. Landsman, and J. Wahlen, 1995, Fair Value Accounting: Effects on Banks' Earnings Volatility, Regulatory Capital, and Value of Contractual Cash Flows, Journal of Banking and Finance, 19:577-605.

Belsley, D., E. Kuh, and R. Welsch, 1980, Regression Diagnostics (New York: John Wiley).

Carroll, T. and T. Linsmeier, 1994, Fair Value Accounting: Evidence from Closed-End Mutual Funds, Unpublished manuscript, University of Iowa, Iowa City.

Christie, A., 1987, On Cross-Sectional Analysis in Accounting Research, Journal of Accounting and Economics, 9(3): 231-258.

Financial Accounting Standards Board, 1980, Statement of Financial Accounting Concepts No. 2 Qualitative Characteristics of Accounting Information, Stamford, Connecticut, May.

Foster, G., 1975, Accounting Earnings and Stock Prices of Insurance Companies, Accounting Review, 50(4): 686-698.

Foster, G., 1977, Valuation Parameters of Property-Liability Companies, Journal of Finance, 32(2): 823-835.

Miller, M. and F. Modigliani, 1966, Some Estimates of the Cost of Capital to the Electric Utility Industry, 1954-1957, American Economic Review, 56(3): 333-391.

Petroni, K., 1992, Optimistic Reporting in the Property-Casualty Insurance Industry, Journal of Accounting and Economics, 15: 485-508.

Petroni, K. and M. Beasley, 1996, Accounting Estimation Error Disclosures and Firm Valuation, Journal of Accounting Research, 34 (1), forthcoming.

Price Waterhouse, 1994, PW Researcher (New York: Price Waterhouse).

Rosenthal, R., 1991, Meta-Analytic Procedures for Social Research (Newbury Park, Cal.: Sage).

Scholes, M., G. P. Wilson, and M. Wolfson, 1990, Tax Planning, Regulatory Capital Planning, and Financial Reporting Strategy for Commercial Banks, Review of Financial Studies, 3(4): 625-650.

Tillinghast, 1988, 1987 SEC Loss Reserve Disclosures (Stamford, Conn.: Tillinghast).

Warfield, T. and T. Linsmeier, 1992, Tax Planning, Earnings Management, and the Differential Information Content of Bank Earnings Components, Accounting Review, 67: 546-562.

Weiss, M., 1985, A Multivariate Analysis of Loss Reserving Estimates in Property-Liability Insurers, Journal of Risk and Insurance, 52:199-221.

White, H., 1980, A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity, Econometrica, 48: 817-838.

Kathy Ruby Petroni is Assistant Professor of Accounting in the Eli Broad Graduate School of Management, Michigan State University. James Michael Wahlen is Assistant Professor of Accounting in the Kenan-Flagler School of Business, University of North Carolina at Chapel Hill.

This study assesses the relation between fair values of equity and fixed maturity debt securities and share prices of property-liability insurers. The results provide the first evidence on the valuation implications of financial statement disclosures regarding fair value estimates for the majority of assets held by insurers. This evidence is important because, beginning with financial statements for year-end 1994, the Financial Accounting Standards Board's (FASB) Statement of Financial Accounting Standards No. 115 (SFAS 115), Accounting for Certain Investments in Debt and Equity Securities, requires formal balance sheet recognition of fair values for considerably more investment securities than was previously the case. The new accounting standard could have a substantial impact on earnings, assets, and capital of property-liability insurers because investment securities comprise over 60 percent of the total assets of a typical property-liability insurer.

The economic consequences of balance sheet recognition of fair values for investment securities are debated widely. The controversy centers primarily on whether fair value estimates for investment securities are sufficiently reliable to be valuation-relevant. Advocates of fair value accounting, including former Securities Exchange Commission (SEC) Chairman Breeden, assert that recognition of fair values will enhance the relevance of information in financial statements (Wall Street Journal, January 8, 1992). Opponents, including Federal Reserve Board Chairman Greenspan and various representatives of the financial services industry, assert that estimates of fair values of thinly-traded securities will be unreliable and, thus, valuation-irrelevant (Wall Street Journal, November 8, 1990). In addition, opponents assert that fair value accounting will induce additional volatility in earnings and capital measures that does not reliably reflect underlying economic volatility, leading to inefficient capital allocation decisions by investors and excessive intervention by regulators.(1)

The FASB (1980) cites relevance and reliability as the two principal criteria for choosing among accounting alternatives. The FASB defines relevant information as that which is capable of influencing decisions, and reliable information as that which is verifiable. These two characteristics are clearly not independent. In this setting, fair values of key assets such as investments are likely to be considered relevant by investors, if disclosed fair values are sufficiently reliable.(2) Since fair values must be estimated for certain securities, estimation error could impair the value-relevance of related disclosures.(3) This study tests the relevance and reliability of fair value disclosures by analyzing their association with share prices of property-liability insurers.

Prior research on accounting for investments, which has focused primarily on banking, has shown that fair value estimates help explain bank share prices.(4) However, prior work does not detect a link between fair value gains and losses and bank stock returns, perhaps because of measurement error in changes in fair value estimates (Barth, 1994). The insurance industry is a more powerful setting to test the valuation implications of fair value disclosures, because bank investment portfolios only range from 14 to 21 percent of total assets and almost exclusively include fixed maturity securities.

We find that fair value disclosures for certain types of investment securities are important determinants of share prices and returns of property-liability insurers. In particular, our evidence suggests that property-liability share prices can be explained by fair values of equity securities, which have traditionally been recognized on the balance sheet, and fair values of U.S. Treasury securities, which have been disclosed in footnotes. We also find that fair value disclosures for other types of investment securities (e.g., municipal and corporate bonds and other debt instruments) do not explain share prices. Further, we observe that our sample of insurers tend to hold U.S. Treasury securities with less than five years to maturity and corporate and municipal bonds with more than five years to maturity.

Our results suggest that the reliability of fair value estimates for different types of securities affects the value-relevance of related disclosures. We detect a relation between share prices of insurers and fair values of investments, but only if the fair value estimates are closely related to market prices - for example, investments traded in more liquid markets. These results lend insight into factors that affect share prices of property-liability insurers and illustrate potential effects of fair value accounting.

The remainder of the article provides background by describing accounting for investments by property-liability insurers, describes the sample, presents our empirical tests and results, and provides concluding remarks.

Background: Accounting for Investment Securities

Prior to 1994, under generally accepted accounting principles, changes in fair values of the majority of securities held by a typical property-liability insurer were accounted for as follows. Changes in fair values of equity securities were recognized in the assets and equity sections of the balance sheet but not in the income statement. Changes in fair values of fixed maturity debt securities were not recognized but were disclosed parenthetically on the balance sheet or in a footnote.(5) Of course, realized gains/losses from security sales were recognized in the income statement.

Beginning with financial statements for 1994, SFAS 115 requires that firms classify their securities into trading, available for sale, and held to maturity portfolios. Accounting for changes in fair values of trading portfolios are formally recognized on the balance sheet and income statement. Under SFAS 115, securities in the held to maturity portfolio are accounted for at amortized cost, with fair values disclosed but not recognized. However, under SFAS 115, few securities qualify for this treatment because SFAS 115 makes restrictive the condition that a firm must have the ability and intent to hold these securities to maturity.(6) Securities that do not qualify for the held to maturity portfolio (e.g., securities that are being held indefinitely) are to be classified as available for sale. For securities considered available for sale, SFAS 115 requires formal balance sheet recognition at fair value, with unrealized gains/losses recognized in the owners' equity section.

The most substantive change for property-liability insurers is that SFAS 115 will for the first time require explicit balance sheet recognition of unrealized gains/losses on most fixed maturity securities. In contrast, the insurance industry is a rare setting in which SFAS 115 will not materially change the way in which equity securities are accounted for in the asset section of the balance sheet. Property-liability industry accounting practice has historically included explicit balance sheet recognition of unrealized gains/losses on equity investments. Thus, this setting provides an opportunity to investigate the value-relevance of recognized fair values of equity investments prior to the adoption of SFAS 115.

Sample Selection and Data Description

The sample comprises 56 publicly-held property-liability insurers operating during 1985 through 1991, including 41 property-liability insurers on the 1990 or 1992 Compustat industrial tapes (primary, secondary, tertiary files).(7) An additional 15 firms meeting our sample criteria were identified from a report on publicly-held property-liability insurers by a management consultant and actuarial firm (Tillinghast, 1988). The sample represents greater than 50 percent of the total premiums earned by the industry.

Accounting data related to the investment portfolios of our sample firms were hand-collected directly from annual reports. Data on stock prices, stock splits and stock dividends were collected from both Standard and Poors' Daily Stock Price Records and Compustat. Data on stock returns were obtained from the Center for Research in Security Prices' monthly (NYSE and AMEX firms) or daily (NASDAQ firms) return files.

Financial statements of property-liability insurers indicate that investment portfolios comprise an important component of their economic value. Table 1 reports descriptive statistics of the 344 insurer-year observations in our sample. The book values of equity investments (carried at fair values) and fixed maturity investments (carried at amortized cost) comprise 7.6 percent and 55.4 percent, respectively, of total assets for the mean sample insurer.

[TABULAR DATA FOR TABLE 1 OMITTED]

The statistics shown in Table 1 also indicate that, on average from 1985 through 1991, fair values of equity and fixed maturity investments exceeded their respective book values by 3.2 percent and 1.5 percent of stockholders' equity. Corresponding standard deviations suggest that considerable variation exists across insurers. For example, the difference between fair value and historical cost of equity securities held by the insurer in the 90th (10th) percentile of this distribution was equal to 15.4 percent (-3.3 percent) of stockholders' equity. Likewise, the difference between fair value and book value of fixed maturity securities held by the insurer in the 90th (10th) percentile of this distribution was equal to 15.9 percent (-7.7 percent) of stockholders' equity.

As noted earlier, opponents of fair value accounting assert that fair value estimates may be less reliable and therefore less value-relevant for securities that are not widely traded or are intended to be held to maturity. Tables 1 and 2 present data on the types and maturity distributions of securities held by our sample insurers. For the 282 insurer-years with disclosures about the issuers of their fixed maturity investments, the mean insurer invested approximately 40 percent and 25 percent of the total fixed maturity portfolio in municipal and U.S. Treasury securities, respectively, with the remainder in corporate and other debt instruments (see Table 1).

Table 2 includes correlations between proportions of the fixed maturity portfolio with different maturities and proportions of the fixed maturity portfolio from different types of issuers, for the 112 sample insurer-years with disclosures of types and maturities of investments. Based on this subsample, these correlation coefficients suggest that insurers that hold shorter-term fixed maturities (less than one year and one to five years) invest more heavily in U.S. Treasury securities. In contrast, insurers holding longer-term maturities (from five to ten years and over ten years) invest more heavily in municipal and corporate securities. No discernible pattern of correlation exists between maturity and percentage of investment in other fixed maturity securities.

[TABULAR DATA FOR TABLE 2 OMITTED]

Empirical Tests

If disclosed fair value estimates measure underlying fair values of investments with a sufficiently high degree of error, then these disclosures are not likely to be considered reliable information and, thus, are not likely to be related to share prices or returns. Alternatively, we predict that fair values of both equity and fixed maturity investments are positively related to share prices, and changes in fair values are positively related to returns. If so, the results can be interpreted as evidence that fair values are sufficiently reliable as to be considered valuation-relevant.(8)

Certain investments held by property-liability insurers are likely to be actively traded in securities markets (e.g., Treasury securities). Fair value disclosures for such investments are based on readily observable market prices. Other types of investments (e.g., municipal and corporate bonds), involve a mixture of thickly and thinly traded securities. Some municipal and corporate issues are actively traded (such as general obligation bonds of large cities and states and corporate bonds of large firms), while others are not (such as obligations of small municipalities or corporations). Thus, fair value disclosures for portfolios of municipal and corporate securities likely require more estimation than disclosures for portfolios of U.S. Treasury securities. We provide evidence on the reliability of these disclosures by examining explicitly the share price impact of fair values that are estimated with more (or less) error across different types of securities.

Value-Relevance of Fair Values of Investment Securities

Like all other firms, the market value of a property-liability firm (MV) is a function of the market value of assets (MVA) less the market value of liabilities (MVL). Under the assumption of perfect and complete markets, this valuation relation for firm i at time t can be expressed as [MV.sub.it] = [MVA.sub.it] - [MVL.sub.it]. Under the accounting identity, the book value of equity in a firm (BV) is equal to the book value of assets (BVA) less the book value of liabilities (BVL): [BV.sub.it] = [BVA.sub.it] - [BVL.sub.it]. Thus, the difference between market and book value of equity can be expressed as a function of the differences between market and book values of assets and liabilities, or

[MV.sub.it] = [BV.sub.it] + ([MVA.sub.it] - [BVA.sub.it]) - ([MVL.sub.it] - [BVL.sub.it]) (1)

(see Barth, Beaver, and Landsman, 1994, and others for very similar approaches to modeling the relations between market and book values).

To test for an association between share prices and fair values of equity and fixed maturity investments, we estimate the following specification of model (1), pooled across 56 sample insurers and the years 1985 through 1991:(9)

[PRICE.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][NBV.sub.it] + [[Beta].sub.2][UCLAIMS.sub.it] + [[Beta].sub.3][ES_HC.sub.it] + [[Beta].sub.4][ES_FVOHC.sub.it] + [[Beta].sub.5][FM_BV.sub.it] + [[Beta].sub.6][FM_FVOBV.sub.it] + [[Beta].sub.7][lnASSETS.sub.it] + [[Epsilon].sub.it] (2)

PRICE denotes price per share of common stock as of March 31 in year t+1.(10) This specification assumes that the relations are constant across sample firms and over this study period, which is reasonable because property-liability insurers have relatively homogeneous asset and liability structures and production functions. Year-specific intercepts are included to control for systematic changes in levels of stock prices in the insurance industry over the sample period.

The focus of this study is to test whether fair value estimates of investments provide value-relevant information incremental to historical costs. Thus, the key variables of interest in this regression are the difference between fair value and historical cost of equity investments (ES_FVOHC), and the difference between fair value and book value of fixed maturity investments (FM_FVOBV). We predict the coefficients on both ES_FVOHC and FM_FVOBV to be positive.(11) Similarly, historical costs of equity securities (ES_HC) and book values of fixed maturities (FM_BV) also should be positively related to share prices.

The variable NBV denotes the net book value of common equity before investments in equity and fixed maturity securities and before the liability for unpaid claims. This variable is included to control for the market's valuation of all net assets not separately identified in our model.

The variable UCLAIMS represents the liability for unpaid claims. This variable controls for potential differences in valuation across insurers with different levels of outstanding claims, which is typically an insurer's largest liability (representing 67 percent of total liabilities and 49 percent of total assets on average). Ideally, we would include a regressor that reflects fair values of liabilities of our sample firms. In particular, changes in fair values of liabilities that hedge changes in fair values of investments are potentially important for firm valuation decisions. Unfortunately, such measures are not disclosed by property-liability insurers (and SFAS 115 does not require recognition); thus, fair values of liabilities are not included in our model.(12)

Accounting practice of property-liability insurers recognizes explicitly the impact of unexpected changes in inflation on the liability for outstanding claims (UCLAIMS).(13) UCLAIMS is reported on the balance sheet at the nominal amount expected to be paid (i.e., a future value), with the change in the nominal value included as a component of income.(14) Fair value of UCLAIMS would equal the present value of the reported nominal value of this liability. Whether UCLAIMS controls for fair values of this liability depends on the extent to which discount rates applied by investors to future claim loss cash flows change with unexpected inflation in those cash flows. If discount rates applied by investors increase to reflect unexpected inflation in future claim loss cash flows, then the present value of this liability remains the same (or may actually fall), and including UCLAIMS in our tests of share prices does not control for changes in fair values of liabilities. Alternatively, if discount rates increase to only partially reflect unexpected inflation in future claim loss cash flows, then UCLAIMS may partially control for changes in fair values of the liability.(15) Because discount rates used by investors are unobservable, it is difficult to assess the impact of changes in fair values of liabilities on our tests.

The natural logarithm of total assets (lnASSETS) is included as a control for size and potential heteroskedasticity. After including lnASSETS, the White (1980) chi-square test statistic fails to reject the null hypothesis of homoskedastic error terms and correct model specification (p = 0.37). Without this control, the model fails White's test (p = 0.07). The results on our variables of interest, however, are not changed by the inclusion of this control.

Model (2) is estimated using ordinary least squares over 342 insurer-years (see Table 3, Panel A).(16) The regression model has an adjusted [R.sup.2] of 67 percent. As expected, fair values of investments are important explanatory variables for insurer share prices. The coefficients on ES_FVOHC and FM_FVOBV are positive and significant at the 0.001 level.(17) In addition, the coefficients on the historical cost of equity securities and the book value of fixed maturity investments are both reliably positive. The book value of common equity before investments (NBV) is positively associated with share prices, and the liability for unpaid claims (UCLAIMS) is negatively associated with share prices.(18)

Table 3

Panel A: Valuation Implications of Fair Value Disclosures for Equity and Fixed Maturity Debt Securities

[PRICE.sub.it] = [[Alpha].sub.it] + [[Beta].sub.1][NBV.sub.it] + [[Beta].sub.2][UCLAIMS.sub.it] + [[Beta].sub.3][ES_HC.sub.it] + [[Beta].sub.4][ES_FVOHC.sub.it] + [[Beta].sub.5][FM_BV.sub.it] + [[Beta].sub.6][FM_FVOBV.sub.it] + [[Beta].sub.7][lnASSETS.sub.it] + [[Epsilon].sub.it]

Independent Variable Coefficient Standard Error T-statistic

NBV 1.35 0.15 8.88(**) UCLAIMS -1.55 0.18 -8.55(**) ES_HC 1.92 0.18 10.40(**) ES_FVOHC 2.38 0.32 7.36(**) FM_BV 1.16 0.13 8.63(**) FM_FVOBV 1.12 0.29 3.87(**) lnASSETS 4.63 0.69 6.69(**)

Panel B: Valuation Implications of Fair Value Disclosures for Equity and Fixed Maturity Debt Securities by Investment Type

[PRICE.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][NBV.sub.it] + [[Beta].sub.2][UCLAIM.sub.it] + [[Beta].sub.3][ES_HC.sub.it] + [[Beta].sub.4][ES_FVOHC.sub.it] + [[Beta].sub.5][US_BV.sub.it] + [[Beta].sub.6][MUNI_BV.sub.it] + [[Beta].sub.7][CORP_BV.sub.it] + [[Beta].sub.8][OTH_BV.sub.it] + [[Beta].sub.9][US_FVOBV.sub.it] + [[Beta].sub.10][MUNI_FVOBV.sub.it] + [[Beta].sub.11][CORP_FVOBV.sub.it] + [[Beta].sub.12][OTH_FVOBV.sub.it] + [[Beta].sub.13][lnASSETS.sub.it] + [[Epsilon].sub.it]

Independent Variable Coefficient Standard Error T-statistic

NBV 0.63 0.14 4.61(**) UCLAIMS -0.97 0.16 -6.10(**) ES_HC 1.71 0.15 11.10(**) ES_FVOHC 1.01 0.22 4.63(**) US_BV 0.54 0.13 4.24(**) MUNI_BV 1.28 0.12 11.00(**) CORP_BV 0.54 0.12 4.39(**) OTH_BV 0.57 0.13 4.42(**) US_FVOBV 3.15 1.05 2.99(*) MUNI_FVOBV -0.70 0.50 -1.40 CORP_FVOBV 0.40 0.50 0.81 OTH_FVOBV -0.73 0.93 -0.78 lnASSETS 2.96 0.62 4.79(**)

Note: Panel A presents results of an ordinary least squares regression of share prices on book values and fair values of equity and fixed maturity securities for a sample of 56 insurers over 1985 through 1991 (342 insurer-years). Adjusted [R.sup.2] = 0.67. Panel B presents results of an ordinary least squares regression of share prices on book values and fair values of equity securities and fixed maturity securities by security type. The sample includes 51 insurers over 1985 through 1991 (265 insurer-years). Adjusted [R.sup.2] = 81.

PRICE = price per share of common equity.

NBV = book value of common equity before investments in equity and fixed maturity securities and before the liability for unpaid claims.

UCLAIMS = liability for unpaid claims.

ES_HC = historical cost of equity securities.

ES_FVOHC= fair value estimate of equity securities minus historical cost.

FM_BV = book value of fixed maturity investments.

FM_FVOBV= fair value estimate of fixed maturity investments minus book value.

US_BV = book value of fixed maturity U.S. Treasury securities.

MUNI_BV = book value of fixed maturity municipal securities.

CORP_BV = book value of fixed maturity corporate securities.

OTH_BV = book value of other fixed maturity securities.

US_FVOBV= fair value estimate of fixed maturity U.S. Treasury securities minus book value.

MUNI_FVOBV = fair value estimate of fixed maturity municipal securities minus book value.

CORP_FVOBV = fair value estimate of fixed maturity corporate securities minus book value.

OTH_FVOBV= fair value estimate of other fixed maturity securities minus book value.

lnASSETS= natural log of assets.

[Epsilon] = error term.

i = insurer index for up to 56 insurers.

t = year index for 1985 through 1991.

All variables (except lnASSETS) are in per share terms and are adjusted for stock splits and stock dividends.

* Statistically significant at the 0.005 level, one-sided.

** Statistically significant at the 0.0001 level, one-sided.

Because serial correlation in residuals may be a problem with a pooled regression of price levels, we also estimate the model on a yearly basis. The results, not reported here, are similar to those reported above. The coefficient on ES_FVOHC is positive in all seven years and significant at less than the 0.05 level in six years. A Z-statistic on the pooled significance of the yearly t-statistics is 3.48 (p = 0.0009).(19) The coefficient on FM_FVOBV is positive in all seven years and significant in four years at less than the 0.05 level. The Z-statistic is 2.78 (p = 0.008).

These results indicate that fair value estimates of investment securities are reflected in the share prices of property-liability insurers. These findings hold for fair value estimates of equity securities, which are explicitly recognized on insurer balance sheets, and for fair value estimates of debt securities, which are disclosed.

Fair value disclosures for certain types of fixed maturity investments (e.g., securities traded in liquid markets) are likely to have more explanatory power for share prices than fair values of other securities. Thus, we expand model (2) to estimate separately the incremental value-relevance of fair values for each of the four commonly disclosed categories of fixed maturity securities. The expanded regression model includes both the book value and the difference between fair value and book value for U.S. Treasury securities (US_BV and US_FVOBV), municipal bonds (MUNI_BV and MUNI_FVOBV), corporate bonds (CORP_BV and CORP_FVOBV), and other fixed maturity investments (OTH_BV and OTH_FVOBV).

Regression results based on 265 insurer-years (51 insurers over 1985 through 1991) are presented in Table 3, Panel B.(20) The standard errors and t-statistics reported are the result of ordinary least squares because the White (1980) chi-square test-statistic is not significant at conventional levels (p = 0.29). The coefficients on the equity security variables and the control variables are similar to those found in the simpler model. The coefficients on the four variables capturing book values of fixed maturity investments are all positive and significant as predicted, but none of the four are statistically equivalent to one. In contrast, of the four fair value-based variables, only the coefficient on US_FVOBV is significant and positive at the 0.05 level. The coefficients on MUNI_FVOBV, CORP_FVOBV and OTH_FVOBV are not significant. These results indicate that book values are value-relevant for all types of fixed maturity investment securities, but fair values are value-relevant only for U.S. Treasury securities.

We also conducted several tests of the robustness of the regression results reported in Table 3, Panel B. First, to address potential effects of serial correlation of residuals we obtained separate yearly estimates of a model similar to the expanded regression. The results are qualitatively similar to those reported above. Coefficients on variables capturing market values of equity securities and U.S. Treasury securities are consistently positive and significant in a majority of sample years. Z-statistics on the pooled significance of the yearly t-statistics are significant at conventional levels.

Second, to test whether the results are sensitive to differences across finns in risk, financial health, or other firm-specific factors, we reestimated the model twice, including firm-specific intercepts and a numerical version of A.M. Best's financial condition rating (on a reduced sample of 246 insurer-years).(21) The results under both procedures are similar to those reported above. However, the coefficient on US_FVOBV becomes less significant (t = 1.72, p = 0.0435) when the model includes firm-specific intercepts (requiring 50 additional parameter estimates).(22)

These results suggest that investors consider fair value disclosures related to equity and U.S. Treasury securities to be more reliable and valuation-relevant than similar disclosures for municipal and corporate bonds and other debt instruments. A potential explanation for this is that investors consider fair value estimates for equities and Treasury securities to be more reliable because they are more likely to be based on readily observable market prices.(23)

Value-Relevance of Changes in Fair Values of Investment Securities

Given that we find that fair values help explain share price levels, changes in fair values of investments should be important determinants of returns. Barth (1994), however, does not observe a consistent relation between returns and fair value gains and losses in the banking industry. Barth's results suggest that measurement error in fair value gains and losses may be sufficiently large as to overwhelm the relation across a sample of banks. To test this possibility in the insurance setting, we estimate the following model pooled across 55 sample insurers and the years 1986 through 1991:

[Return.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][Delta][EPS_BRGL.sub.it] + [[Beta].sub.2][RGL.sub.it] + [[Beta].sub.3][Delta][UGL.sub.it] + [[Epsilon].sub.it], (3)

where Return denotes return on common stock in sample firm i from April 1 in year t to March 31 in year t+1.(24) Yearly intercepts are included to control for systematic changes in stock prices over time.

To control for returns associated with income before gains and losses on investment securities, the model includes changes in primary earnings per share before realized gains and losses (after taxes) and before discontinued operations and extraordinary items ([Delta]EPS_BRGL).(25) We predict that the coefficient on [Delta]EPS_BRGL is positive. If changes in earnings are expected to be permanent, then, in an ideal setting (e.g., perfect and complete markets), this coefficient would equal the reciprocal of the cost of capital (Miller and Modigliani, 1966).

Realized gains and losses on investments after taxes (RGL) are included to capture returns associated with sales of investments. Changes in unrealized gains and losses on equity and fixed maturity investments ([Delta]UGL) are included to test whether returns are associated with changes in fair values of investment securities. [Delta]UGL is calculated as the change in the excess of fair value over historical cost for equity securities plus the change in the excess of fair value over book value for fixed maturity securities.(26) In this regression, the coefficient on [Delta]UGL should capture the value-relevance (if any) of changes in the fair value of the securities portfolio that are incremental to realized gains and losses. We predict that this coefficient should be positive. This specification assumes that expectations for realized gains and losses and changes in unrealized gains and losses are zero over the pooled cross-section, which is plausible and consistent with prior work (e.g., Barth, 1994).(27)

Ordinary least squares regression results based on 274 insurer-years are reported in Table 4, Panel A. White's (1980) test does not reject homoskedasticity and correct model specification (p = 0.29). The model has an adjusted [R.sup.2] of 37 percent. As expected, the coefficient on [Delta]EPS_BRGL is positive and significant. This coefficient is statistically equivalent to one, consistent with the possibility that changes in earnings in this sample are on average expected to be transitory rather than permanent.

The coefficient on RGL is positive and significant at the 0.01 level, and is statistically equivalent to one. This is contrary to previous research. Warfield and Linsmeier (1992) and Barth (1994) observe that the association between bank stock returns and realized gains and losses on investment securities is negative on average. Both studies suggest that the findings are dominated by stock price increases associated with firms that selectively realize losses to reduce taxes.(28) To provide evidence on this possibility, we examine separately returns associated with four subsamples: high versus low marginal tax rate insurers (see footnote 11) with net realized gains versus net realized losses. Our findings from these tests (untabulated) indicate no significant negative coefficients on any of the four subsamples. In fact, we obtain a significant positive coefficient on the high tax/net realized losses subsample that appears to drive the results in prior work.(29)

Table 4

Tests of the Relation Between Returns and Realized and Unrealized Gains and Losses on Equity and Fixed Maturity Debt Securities

Panel A: [Return.sub.it] = [[Alpha].sub.t] + [[Beta].sub.1][EPS_BRGL.sub.it] + [[Beta].sub.2][RGL.sub.it] + [[Beta].sub.3][Delta][UGL.sub.it] + [[Epsilon].sub.it]

Independent Variable Coefficient Standard Error T-statistic

[Delta]EPS_BRGL 0.89 0.10 8.96(**) RGL 0.72 0.26 2.78(*) [Delta]UGL 0.19 0.08 2.40(*)

Panel B: [Return.sub.it] = [[Alpha].sub.it] + [[Beta].sub.1][Delta][EPS_BRGL.sub.it] + [[Beta].sub.2][ES_RGL.sub.it] + [[Beta].sub.3][ES_[Delta]UGL.sub.it] + [[Beta].sub.4][FM_RGL.sub.it] + [[Beta].sub.5][FM_[Delta]UGL.sub.it] + [[Beta].sub.6][MISC_RGL.sub.it] + [[Epsilon].sub.it]

Independent Variable Coefficient Standard Error T-statistic

[Delta]EPS_BRGL 0.82 0.10 7.97(**) ES_RGL 0.73 0.37 2.00(*) ES_[Delta]UGL 0.21 0.12 1.70(*) FM_RGL 0.36 0.42 0.85 FM_[Delta]UGL 0.17 0.13 1.30 MISC_RGL 1.43 0.97 1.48

Note: Panel A presents ordinary least squares results of a regression of returns on change in earnings before realized gains and losses on investments, realized gains and losses on investments, and the change in unrealized gains and losses on investments. The sample includes 55 insurers over 1986 through 1991 (274 insurer-years). Adjusted [R.sup.2] = 0.37. Panel B presents ordinary least squares results of a regression of returns on change in earnings before realized gains and losses on investments, realized gains and losses by investment type, and the change in unrealized gains and losses by investment type. The sample includes 55 insurers over 1986 through 1991 (270 insurer-years). Adjusted [R.sup.2] = 0.36.

Return = common stock returns from April 1 in year t through March 31 in year t+1.

[Delta]EPS_BRGL = change in primary earnings per share before realized gains and losses on investments, discontinued operations, and extraordinary items.

RGL = realized gains and losses (after taxes) on investments.

[Delta]UGL = change in the excess of market value over historical cost for equity securities plus the change in the excess of market value over book value for fixed maturity securities.

ES_RGL = realized gains and losses after estimated taxes on equity securities.

ES_[Delta]UGL = change in the excess of market value over historical cost for equity securities.

FM_RGL = realized gains and losses after estimated taxes on debt securities.

FM_[Delta]UGL = change in the excess of market value over book value for debt securities.

MISC_RGL = realized gains and losses after estimated taxes on investments other than equity and debt securities.

[Epsilon] = error term.

i = insurer index for 55 insurers.

t = year index for 1986 through 1991. All variables are stated in per share terms, scaled by beginning period price, and adjusted for stock splits and stock dividends.

* Statistically significant at the 0.05 level, one-sided.

** Statistically significant at the 0.0001 level, one-sided.

The coefficient on [Delta]UGL is positive and significant (p = 0.01), indicating changes in fair values of investments are value-relevant in the property-liability insurance industry. This result is not consistent with results based on the banking industry (Barth, 1994). It appears that the insurance industry is a more powerful setting for this test.

Results from our stock price regressions suggest that investors price fair values for certain types of investments (e.g., equities and Treasury securities) more than other types. Ideally, we would test the information content in changes in fair values for equity securities and each of the four categories of fixed maturity securities. This test requires data on realized security gains and losses after taxes for each category. Otherwise, results would be difficult to interpret because the predicted response to changes in unrealized gains and losses of each category should be conditional on the realized gains and losses of that category. Unfortunately, insurers do not separately disclose realized gains and losses after taxes for the equity and four types of fixed maturity securities.

Insurers generally do, however, disclose separately realized gains and losses before taxes for portfolios of equity and fixed maturity securities (but not by fixed maturity category). This allows us to estimate realized gains and losses after taxes for the two portfolios and thus test the association between returns and the change in unrealized gains and losses on the equity and fixed maturity portfolios separately using an expanded form of model (3). In place of RGL, the expanded model includes after-tax estimates of the three disclosed components of RGL: equity securities (ES_RGL), fixed maturity securities (FM_RGL), and all other investments - e.g., mortgages and real estate - (MISC_RGL). The tax effects of the realized gains and losses are estimated based upon the tax position of the insurers as disclosed in the footnotes to the financial statements (see footnote 11). Our tax estimates are reasonable but add noise to the regression model since the Pearson correlation coefficient between the sum of ES_RGL, FM_RGL, and MISC_RGL, and the total after-tax realized gains and losses as reported in the financial statements (RGL) is 0.83. In addition, instead of [Delta]UGL, the expanded model includes two disclosed components: the change in the excess of the fair value over historical cost of equity securities (ES_[Delta]UGL) and the change in the excess of the fair value over book value of fixed maturity securities (FM_[Delta]UGL).

Regression results based on 270 insurer-years (55 insurers from 1986 through 1991) are reported in Table 4, Panel B.(30) White's test does not reject homoskedasticity and correct model specification (p = 0.50). The model has an adjusted [R.sup.2] of 36 percent. Again, the coefficient on [Delta]EPS_BRGL is positive and significant. As expected, the coefficients on realized gains and losses on equity securities and changes in unrealized gains and losses on equity securities are both significantly positive at the 0.05 level. The coefficients on realized gains and losses on fixed maturities and changes in unrealized gains and losses on fixed maturities are positive but not significant.

The results reported in Table 4 are robust to several specification tests. First, the results are similar when the models include firm-specific intercepts and lnASSETS to control for firm-specific differences such as risk or size. Separate yearly estimates of each model also produce similar results. Z-statistics indicate that pooled annual t-statistics are significant at conventional levels for coefficients on [Delta]UGL and ES_[Delta]UGL but not for coefficients on FM_[Delta]UGL.

The results indicate fair value gains and losses on equity securities in particular help explain returns for property-liability insurers, but fair value gains and losses on fixed maturity securities do not. These results are not necessarily inconsistent with our other findings. Our ability to detect a relation between returns and changes in unrealized gains and losses on fixed maturity securities is limited because changes in unrealized gains and losses on fixed maturities reflect municipal, corporate, and other fixed maturity investments, which were not found to be value-relevant in our share price regression results.

Conclusion

This research enhances our understanding of how property-liability insurers are valued and contributes to the ongoing debate concerning the relevance and reliability of fair value accounting information for investment securities and other financial assets and liabilities. We find that, although book values for all categories of investments are valuation-relevant, fair values for only certain categories of investment securities are reflected in share prices and returns of property-liability insurers. It appears that fair values for certain categories of investments, such as equities and U.S. Treasury securities, which are more likely to be traded in active markets, are valuation-relevant. Share prices are not explained by fair values for other types of investments - e.g., municipal and corporate bonds and other debt instruments - which may be less actively traded and which tend to have longer terms to maturity than U.S. Treasury securities in this sample.

One interpretation of these findings is that concern over application of fair value accounting techniques is warranted, especially for assets and liabilities without readily observable market prices. Alternatively, one can also view these results as evidence that sophisticated investors discriminate between fair value estimates of different assets and rely only on fair value disclosures that are more closely aligned with readily observable market prices.

The authors appreciate comments from M. Barth, J. Hand, W. Landsman, T. Linsmeier, R. Pfeiffer, T. Warfield, two anonymous reviewers, the editor, and workshop participants at the University of Georgia, Michigan State University, the University of Oregon, Penn State University, and the 1994 meeting of the American Accounting Association. The authors are grateful for the research assistance of Norman Godwin, Luann Lynch, and Daqing Qi and financial support from Michigan State University, the Center for Financial and Accounting Research at the University of North Carolina, and the KPMG Peat Marwick Research Fellowship program.

1 However, there is no indication that statutory accounting principles (SAP) will change as a result of SFAS 115. Under current SAP requirements, equity securities are reported at fair value, and fixed maturity securities are generally reported at amortized cost.

2 However, certain opponents of fair value accounting argue that fair value disclosures are irrelevant for debt securities that are to be held to maturity, on the basis that all contractual cash flows will be received over the life of the instrument; thus, changes in fair values are temporary. Evidence in Barth, Landsman, and Wahlen (1995) does not support this argument.

3 The FASB adopted the term "fair value" instead of market value to encompass estimated values for securities that are not traded in active markets.

4 For example, samples of banks have been used to examine the value-relevance of fair value disclosures in Barth, Beaver, and Wolfson (1990), Warfield and Linsmeier (1992), Ahmed and Takeda (1993), Barth (1994), and Barth, Landsman, and Wahlen (1995). Foster (1975, 1977) used a sample of 22 insurers to examine the value-relevance of total gains and losses on equity securities. A sample of closed-end mutual funds is examined in concurrent research on the relevance of fair values by Carroll and Linsmeier (1994).

5 Technically, SFAS 60, effective in 1982, requires that fixed maturity securities held in trading portfolios be recognized separately and accounted for at fair value. Curiously, a separate trading portfolio disclosure does not appear on any of our sample firms' balance sheets until such disclosure is provided by one firm in 1990 and three firms in 1991; only six sample insurers provide footnote disclosures about trading portfolios during this period.

6 The intent of SFAS 115 and current SEC enforcement practice indicate that this is a stringent condition. For example, although no formal position has been issued, the SEC has, through the use of comment letters, already begun to require many financial institutions to classify some or all of their investment portfolio as available for sale (Price Waterhouse, 1994). Aetna Life and Liability Co., for instance, which adopted SFAS 115 in 1993 (one year early), classified less than 10 percent of all equity and debt securities in their held to maturity portfolio.

7 This sample was constructed originally by Anthony and Petroni (1994). The set of firms used here includes all insurers in Anthony and Petroni (1994) that are classified primarily as property-liability insurers (e.g., a primary Standard Industrial Classification code of 6331 for fire, marine, and liability insurance) and are considered to be actively traded (e.g., share price changes on greater than 50 percent of the trading days in the sample period). Three insurers on Compustat are omitted from the sample because they are not actively traded, and one is omitted because the stock was traded for less than one year.

8 In addition, recognition of unrealized gains/losses could lead to contractual implications (e.g., regulatory intervention or binding debt covenants, or greater scrutiny by auditors) that could affect share prices (cf., Barth, Landsman, and Wahlen, 1995). Opponents of fair value accounting also have asserted that these implications could alter investment behavior of financial institutions, making them more reluctant to hold securities with volatile fair values.

9 All variables (except lnASSETS) are stated on a per share basis to reduce the presence of heteroskedastic error terms. All per share data are adjusted for stock splits and stock dividends.

10 Stock prices as of March 31 in year t+1 are used because all sample insurers have December 31 fiscal year ends, and disclosures regarding investments are not likely to be public until the release of the annual report. Because all other variables are measured as of December 31 in year t, using March 31 in year t+1 stock prices adds noise to our tests.

11 More precisely, under ideal estimation conditions, the coefficients should equal one minus the present value of the expected future marginal tax rate, because appreciation/depreciation in market values of securities does not explicitly affect taxes of an insurer until realized. The reported results are based on differences between market and book values measured on a before-tax basis. To examine the effects of taxes on these results, however, tax implications of differences between market and book values are estimated using the statutory tax rate and the tax status of the insurer. We employ the standard technique of assuming that the insurer faces a zero marginal tax rate if it has a net operating loss or investment tax credit carryforward; otherwise, the marginal rate is the statutory rate (see, e.g., Scholes, Wilson, and Wolfson, 1990). The results are similar using either a before- or after-tax basis.

12 Prior research on the association between fair values in investments and share prices generally has omitted changes in the fair value of liabilities, potentially biasing their coefficients on fair value-based variables toward zero. Armed and Takeda's (1993) analysis includes a variable that reflects net assets maturing or repricing after one year as a proxy for interest-rate sensitivity. Their results indicate that this variable is an important factor affecting bank stock prices; however, it does not change their conclusion that realized and unrealized security gains and losses are positively related to bank stock returns. Also, in concurrent research, Carroll and Linsmeier (1994) show that fair value disclosures for both equity and debt securities are value-relevant for share prices and returns of closed-end mutual funds, which are all-equity firms that invest exclusively in securities and thus do not have any omitted correlated variable problems.

13 Unearned premiums is the second largest liability of a property-liability insurer (representing 22 percent of total liabilities and 16 percent of total assets). It is a short-term liability reflecting prepaid premiums and is not significantly influenced by unexpected changes in interest rates. Thus, book value approximates fair value. The reported statistics are based on 1987 stock property-liability industry totals from A.M. Best Company (1988).

14 Weiss (1985) and Petroni (1992) find evidence that the liability for outstanding claims is subject to management manipulation.

15 Evidence in Petroni and Beasley (1996) suggests that the level of the liability for unpaid claims relative to total liabilities is a proxy for the length of the claim cycle (i.e., the lag between the incurrence of a loss and the final claim settlement), which is likely to be a good proxy for the interest rate sensitivity of the liabilities of an insurer. To further control for interest rate sensitivity of the liabilities of an insurer (as a potential correlated omitted variable), all models were also estimated including the ratio of unpaid claims to total liabilities as an independent variable. This additional control variable is not included in our tabulated results, however, because this ratio is never significant and the results on all other variables are unchanged.

16 Conventional influence diagnostics reveal the presence of two observations with an undue influence on this test. The reported results are based on a sample that excludes these two observations. Similar results are obtained when these observations are included.

17 All reported probabilities on t-statistics are for one-tailed tests if the sign of the association is predicted; otherwise, two-tailed probabilities are reported.

18 Some insurers in our sample also write some life insurance. Life insurers tend to have longer-term liabilities than property-liability insurers. To test for possible differences across firm types, we reran all of our regressions on the subset of 40 insurers that write only property-liability insurance. The results from this subsample are similar to those from the full sample.

19 As in Barth (1994) and Rosenthal (1991, pp. 89-109), the Z-statistic is computed as follows:

[Mathematical Expression Omitted].

20 Sixty-two observations are omitted from this test because they do not meet additional data requirements, and 14 observations are omitted because the footnote disclosure and the balance sheet do not agree. Three observations are withheld because conventional influence diagnostics (e.g., Belsley, Kuh, and Welsch, 1980) show that they have an undue influence on this test.

21 A.M. Best is the leading supplier of ratings of the financial condition of insurers. To quantify the letter rating assigned yearly by Best to each firm, we convert it to a number such that a rating of A+ is assigned a 1, A is assigned a 2, ... C is assigned an 8, and Below C is assigned a 9. The rating is not found to have explanatory power in our model.

22 We also examined the sensitivity of the results reported in Table 3, Panel A using the same checks. The results for both sensitivity checks are similar to those reported above; however, when 55 firm-specific intercepts are included, the coefficient on FM_FVOBV becomes insignificant (t = 0.52). These additional coefficients reduce the efficiency of the model, and firm-specific intercepts appear to capture some of the firm-specific effects of FM_FVOBV in this specification.

23 Ideally, we would test this explanation by examining whether valuation implications differ across types of securities with fair values that are based on readily available market prices vs. estimates. Unfortunately, we do not have data on the availability of market prices of the specific investments held by these sample insurers.

24 All variables are stated on a per-share basis, scaled by price as of April 1 in year t (Christie, 1987), with per-share data adjusted for stock splits and stock dividends.

25 This specification assumes that [Delta]EPS_BRGL follows a random walk, which is consistent with prior research. Results are qualitatively similar when the level of earnings per share is used.

26 Changes in unrealized gains and losses of insurers are not taxed until realized. Thus, the reported results are based on changes in unrealized gains and losses on a before-tax basis. The results are similar, however, whether changes in unrealized gains and losses are used on a before- or estimated after-tax basis.

27 Of course, expectations for realized gains and losses and for changes in unrealized gains and losses may not be zero under certain scenarios (e.g., investors may form nonzero expectations for realized gains and losses for certain firms in years they face low marginal tax rates or low levels of regulatory capital).

28 The results in Warfield and Linsmeier (1992), however, do not hold for gains/losses realized in the fourth quarter, when year-end tax-planning activities are most likely to occur. As they note, this result could be due to noise associated with realized gains and losses due to fourth-quarter earnings or regulatory capital management.

29 Warfield and Linsmeier (1992) examine two-day returns, which may have more power to detect market reactions to gains/losses realized for tax purposes. Additional explanations may be found in differences in tax or regulatory treatments across banks versus insurers and across study periods.

30 Four insurer-years are excluded in the expanded model regression because the footnote disclosures do not separate realized gains and losses between the equity and fixed maturity investment portfolios.

References

Ahmed, A. and C. Takeda, 1993, Stock Market Valuation of Gains and Losses on Commercial Banks' Investment Securities: An Empirical Analysis, Unpublished manuscript, University of Florida, Gainesville.

A. M. Best Company, 1988, Best's Aggregates and Averages (Oldwick, N.J.: A. M. Best).

Anthony, J. and K. Petroni, 1994, Accounting Estimation Errors and Firm Valuation, Unpublished manuscript, Michigan State University, East Lansing.

Barth, M., 1994, Fair Value Accounting: Evidence from Investment Securities and the Market Valuation of Banks, Accounting Review, 69: 1-25.

Barth, M., W. Beaver, and W. Landsman, 1994, Value-Relevance of Banks' Fair Value Disclosures under SFAS 107, Unpublished manuscript, Harvard University, Cambridge, Massachusetts.

Barth, M., W. Beaver, and M. Wolfson, 1990, Components of Bank Earnings and the Structure of Bank Share Prices, Financial Analysts Journal, 46: 53-60.

Barth, M., W. Landsman, and J. Wahlen, 1995, Fair Value Accounting: Effects on Banks' Earnings Volatility, Regulatory Capital, and Value of Contractual Cash Flows, Journal of Banking and Finance, 19:577-605.

Belsley, D., E. Kuh, and R. Welsch, 1980, Regression Diagnostics (New York: John Wiley).

Carroll, T. and T. Linsmeier, 1994, Fair Value Accounting: Evidence from Closed-End Mutual Funds, Unpublished manuscript, University of Iowa, Iowa City.

Christie, A., 1987, On Cross-Sectional Analysis in Accounting Research, Journal of Accounting and Economics, 9(3): 231-258.

Financial Accounting Standards Board, 1980, Statement of Financial Accounting Concepts No. 2 Qualitative Characteristics of Accounting Information, Stamford, Connecticut, May.

Foster, G., 1975, Accounting Earnings and Stock Prices of Insurance Companies, Accounting Review, 50(4): 686-698.

Foster, G., 1977, Valuation Parameters of Property-Liability Companies, Journal of Finance, 32(2): 823-835.

Miller, M. and F. Modigliani, 1966, Some Estimates of the Cost of Capital to the Electric Utility Industry, 1954-1957, American Economic Review, 56(3): 333-391.

Petroni, K., 1992, Optimistic Reporting in the Property-Casualty Insurance Industry, Journal of Accounting and Economics, 15: 485-508.

Petroni, K. and M. Beasley, 1996, Accounting Estimation Error Disclosures and Firm Valuation, Journal of Accounting Research, 34 (1), forthcoming.

Price Waterhouse, 1994, PW Researcher (New York: Price Waterhouse).

Rosenthal, R., 1991, Meta-Analytic Procedures for Social Research (Newbury Park, Cal.: Sage).

Scholes, M., G. P. Wilson, and M. Wolfson, 1990, Tax Planning, Regulatory Capital Planning, and Financial Reporting Strategy for Commercial Banks, Review of Financial Studies, 3(4): 625-650.

Tillinghast, 1988, 1987 SEC Loss Reserve Disclosures (Stamford, Conn.: Tillinghast).

Warfield, T. and T. Linsmeier, 1992, Tax Planning, Earnings Management, and the Differential Information Content of Bank Earnings Components, Accounting Review, 67: 546-562.

Weiss, M., 1985, A Multivariate Analysis of Loss Reserving Estimates in Property-Liability Insurers, Journal of Risk and Insurance, 52:199-221.

White, H., 1980, A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity, Econometrica, 48: 817-838.

Kathy Ruby Petroni is Assistant Professor of Accounting in the Eli Broad Graduate School of Management, Michigan State University. James Michael Wahlen is Assistant Professor of Accounting in the Kenan-Flagler School of Business, University of North Carolina at Chapel Hill.

Printer friendly Cite/link Email Feedback | |

Author: | Petroni, Kathy Ruby; Wahlen, James Michael |
---|---|

Publication: | Journal of Risk and Insurance |

Date: | Dec 1, 1995 |

Words: | 8656 |

Previous Article: | The relation between capital structure, interest rate sensitivity, and market value in the property-liability insurance industry. |

Next Article: | External impacts on the property-liability insurance cycle. |

Topics: |