+ All Categories
Home > Documents > Accounting Earnings and Stock Prices of Insurance Companies

Accounting Earnings and Stock Prices of Insurance Companies

Date post: 15-Jan-2017
Category:
Upload: george-foster
View: 215 times
Download: 2 times
Share this document with a friend
14
Accounting Earnings and Stock Prices of Insurance Companies Author(s): George Foster Source: The Accounting Review, Vol. 50, No. 4 (Oct., 1975), pp. 686-698 Published by: American Accounting Association Stable URL: http://www.jstor.org/stable/245234 . Accessed: 14/06/2014 16:13 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . American Accounting Association is collaborating with JSTOR to digitize, preserve and extend access to The Accounting Review. http://www.jstor.org This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PM All use subject to JSTOR Terms and Conditions
Transcript
Page 1: Accounting Earnings and Stock Prices of Insurance Companies

Accounting Earnings and Stock Prices of Insurance CompaniesAuthor(s): George FosterSource: The Accounting Review, Vol. 50, No. 4 (Oct., 1975), pp. 686-698Published by: American Accounting AssociationStable URL: http://www.jstor.org/stable/245234 .

Accessed: 14/06/2014 16:13

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

American Accounting Association is collaborating with JSTOR to digitize, preserve and extend access to TheAccounting Review.

http://www.jstor.org

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 2: Accounting Earnings and Stock Prices of Insurance Companies

Accounting Earnings and Stock Prices of Insurance Companies

George Foster

SEVERAL aspects of the relationship be- tween accounting earnings and stock prices (or returns) have been ex-

amined. Early work reported a positive association between accounting earnings and stock prices in the sense that firms which had positive (negative) earnings changes also had positive (negative) stock price changes (e.g., Ball and Brown, 1968). Much subsequent work has examined if this observed positive association is a mechanistic one; i.e., does the aggregate market naively react to a positive (nega- tive) reported earnings change with an up- ward (downward) revaluation of the stock price or does it also look at the economic aspects underlying the reported earnings number?

This subsequent work has examined (among other things): (a) the market reac- tion to accounting changes (e.g., Ball, 1972; Kaplan and Roll, 1972; and Sunder, 1973); and (b) whether any nonreported earnings measures have a higher associa- tion with stock prices than the reported measure (e.g., Beaver and Dukes, 1973). The results of this subsequent work are consistent with the relationship between accounting earnings and stock prices not being mechanistic; i.e., the aggregate market considers the economic aspects underlying accounting earnings. The above and similar papers have also found that the securities market reacts in an unbiased

and rapid manner to the public announce- ment of information pertinent to security price revaluation; i.e., the securities mar- ket appears to be semi-strong efficient.

Prior papers in this area have generally been restricted to New York Stock Ex- change (NYSE) firms.' The generality of the results of these papers to non-NYSE markets is an open issue. In this paper, the association between accounting earnings and stock prices for an important segment of the Over-the-Counter (OTC) market will be examined, i.e., insurance stocks. This topic is of added interest to the ac- counting profession as considerable re- sources have been devoted to analyzing two issues in the insurance industry, i.e., the measurement of underwriting earnings and the measurement of capital gains and losses on marketable equity securities. Several of the arguments relating to these two issues are based on assumptions about the stock market reaction to accounting

This paper was one of the winning manuscripts in the 1975 American Accounting Association Manuscript Contest.

Indebtedness to my Ph.D. committee of Professors W. Beaver, C. Horngren, and R. Litzenberger must be recorded. R. Ball, P. Brown, J. Demski, N. Dopuch, N. Gonedes, and S. Penman also gave useful comments. Financial assistance from an Arthur Andersen Fellow- ship supported this research.

1 May (1971) and Hagerman (1973) are the only known studies on non-NYSE markets.

George Foster is A assistant Professor of Accounting, University of Chicago.

686

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 3: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 687

data, e.g., that there is a mechanistic rela- tion between reported income and stock prices, and that the market for insurance stocks is inefficient in the semi-strong form. The validity of these arguments will be examined in this paper.

ACCOUNTING ISSUES IN THE

INSURANCE INDUSTRY

Statutory vs. Adjusted Underwriting Earnings

For many years, the accounting princi- ples used in the financial reports of in- surance companies (termed "statutory ac- counting principles") differed from gener- ally accepted accounting principles. For instance, the costs of writing new policies were charged to operations as incurred, whereas premiums were taken into income over periods covered by the related poli- cies. Similarly, the interest rate assump- tion used in computing the reserves of life insurance companies was generally less than the actual return on their invest- ments.

In the 1960's there were attempts by both the accounting profession and finan- cial analysts to develop uniform methods to make insurance company earnings more comparable to the earnings of noninsur- ance companies. The AICPA issued an Industry A audit Guide on Fire and Casualty Companies in 1966 which advocated the allocation of "prepaid acquisition costs applicable to unearned premiums" (p. 60). In 1969 the New York Association of In- surance and Financial Analysts outlined a method for adjusting life insurance statu- tory earnings. The AICPA Industry Guide on A audits of Stock Life Insurance Com- panies was published in 1972 and advo- cated separate adjustments for (a) the acquisition costs of new policies and (b) policy reserve interest rate assumptions. Since 1973 all life insurance companies are required to report adjusted (i.e., GAAP) earnings in their financial statements.

A frequent argument advanced in sup-

port of the accounting profession's role in providing adjusted underwriting measures was that the stock market was taking the reported statutory numbers at their face value. For instance, a member of the AICPA Committee which prepared the Life Insurance Audit Guide, made the fol- lowing comment on the Guide:

Was it worth the cost and effort . . . ? To answer this .. . we must look to the interests and needs of shareholders and the investing public. To the extent that conservative regulatory accounting practices frequently resulted in understated earnings of insurance companies, thus dir ectly affecting the prices of stocks, they clearly did not serve investors' interests and needs (Waterfield, 1974, p. 4, emphasis added).

The descriptive validity of the argument that the aggregate market naively reacted to the reported statutory life underwriting measure will be examined. The null hy- pothesis is that the statutory life under- writing measure will have the highest asso- ciation with stock prices. A finding that an adjusted underwriting measure has a higher association is evidence inconsistent with null hypothesis. Note that there is some evidence that much of the invest- ment community prefers an adjusted mea- sure. For instance, many financial analysts, Standard and Poor's and A. M. Best, re- port adjusted earnings in their investment advisory services. Evidence inconsistent with the null hypothesis would suggest that the above argument used to justify the accounting profession reporting ad- justed life underwriting earnings was invalid.

Marketable Equity Securities

Fire-casualty insurance companies and, to a lesser extent, life insurance companies hold sizable proportions of their assets in common stocks. Given these relatively high holdings of common stocks, it is not surprising that the reaction of the insur- ance industry to proposed changes in com-

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 4: Accounting Earnings and Stock Prices of Insurance Companies

688 The Accounting Review, October 1975

mon stock reporting rules has been a major concern to accounting policy makers. Horngren (1973) notes that the Account- ing Principles Board's initial preference for annual realized and unrealized capital gains and losses being included in the in- come statement (i.e., flow-thru) was blocked by extreme opposition from the insurance industry.

One set of arguments the insurance in- dustry used to justify their opposition to flow-thru-and their preference for capital gains and losses being disclosed in a sepa- rate statement-was the alleged mislead- ing effect on investors and the aggregate market of flow-thru income numbers. For instance, the Allstate Insurance Company argued that the "distortions" induced by flow-thru "may unduly affect investors' opinions and may be a damaging factor to the orderly functioning of the stock mar- ket" (A.P.B. Public Hearing . . ., p. 417). The Monumental Corporation argued that ''reporting capital gains and losses on an annual basis would just be intolerable- for our stockholders" and that it might cause "the stock of the company to decline when it shouldn't" (APB Public Hearing

., p. 71). The argument that disclosing capital

gains and losses in the income statement rather than in some separate statement may adversely affect stock prices implicitly assumes semi-strong inefficiencies in the capital market. In a semi-strong efficient market, all publicly available information is fully impounded into stock prices. The available evidence on semi-strong effi- ciency is generally inconsistent with the inefficiencies implied in the above insur- ance industry argument. This evidence, however, is mainly based on firms listed on the NYSE. As most insurance stocks are listed on the OTC market, policy makers have little evidence with which to assess the likelihood of the above consequences. The research in this paper will provide

evidence on the semi-strong efficiency of the market for insurance stocks. Evidence consistent with semi-strong efficiency would suggest that the above argument by insurance industry officials is invalid.

COMPANIES INCLUDED IN THE STUDY

Earnings information on insurance com- panies was taken from the A. M. Best In- surance Securities Research Service. The Service provides financial data on insurance companies for the prior five years, with the current version of the Service dating back to 1969. By using the Service from 1969 to 1973, earnings data from 1964 to 1972 can be obtained. The A. M. Best Service re- ports both statutory underwriting earnings and an adjusted underwriting measure, in addition to other financial data about in- surance companies.

The 1973 edition of the Service includes reports on 107 insurance companies. To qualify for inclusion in this study, two criteria had to be met: (1) earnings data available from 1964 to 1972 and (2) monthly stock prices and dividend data for the period from January 1965 to April 1973. Seventy-three of the above 107 firms met these criteria.

The majority of these 73 firms specialize in either life-health or fire-casualty insur- ance. Several firms, however, underwrite both types of insurance. The criterion chosen for classifying a firm in either in- dustry was premiums earned in 1972. A company was classified life-health (fire- casualty) if over 50% of its premiums were from life-health (fire-casualty) policies. This classification resulted in 51 life-health firms and 22 fire-casualty firms.

A. M. Best describe the firms included in their service as "prominent stock insur- ance companies." These firms are much larger and have more widely dispersed stockholdings than many small insurance companies. Consequently, the conclusions are not as easily generalizable to all in-

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 5: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 689

surance stocks than would have been the case had the sample selection criterion been a random selection from the total universe of such stocks.

DESCRIPTION OF TIME-SERIES METHODOLOGY

The association between accounting earnings and security prices will be ex- amined with the time-series "residual methodology." This methodology requires estimates of (1) abnormal security returns and (2) positive and negative earnings changes.

Measurement of A normal Security Returns

Estimates of abnormal returns of indi- vidual securities are obtained with a multi- index version of the market model (Sharpe, 1970). This model is a description of the stochastic process generating security re- turns. The following version of the model is used:

Rit = a + /31Rmt + 132t + Cit

E(Uit) = 0

E(Rmt Uit) = 0

E(It Uit) = (

E(Uit, Uik) {ny if t 5

k

where

R-i= the return on security i in pe- riod t

Rmt the return on a representative market index in period t

i= the return on a representative industry index in period t

a,1 31302=parameters to be estimated

Estimates of the parameters of (1) are de- rived from a time-series OLS regression. The total data base includes price relatives for 100 months-January 1965 to April 1973. In estimating this relationship, the mechanics of least squares regression will

force the sum of the computed residuals to be zero. However, if in the earnings period being examined stock prices adjust to the earnings change, the residuals in this pe- riod will be nonzero. Including these months in the regression will cause a bias in the computed residuals. Hence, in esti- mating (1) a separate regression is run for each earnings period examined. The twelve months prior to the relevant earnings an- nouncement are deleted in each regression. Thus, 88 months are used to estimate (1) for each earnings announcement period.

The general market index used is Stan- dard and Poor's Composite Index of 500 stocks. The industry index is an index that is made orthogonal to the general market index. This orthogonalization is achieved by regressing an industry index on the general market index. The residuals from this first-stage regression are used as the industry index in (1). The industry indexes in the first stage are constructed (assuming equal weighting) from the firms in the study. Separate industry indexes are con- structed for the fire-casualty and life- health samples of firms.

Cross-sectional statistics are presented in Table 1. These statistics are for the 1972 earnings period regression. They are very similar to those for the other earnings an- nouncement periods. In contrast to pre- vious studies,2 the industry variable ex- plains a large proportion of the variation of a firm's monthly security rates of return. This increase in the average adjusted R2 by adding the industry variable occurs for both the fire-casualty and life-health sam- ples of firms. The statistical tests used in this paper assume independence in each firm's abnormal price returns. If these in- dustry cross-sectional commonalities in se- curity returns are ignored, this independence assumption would likely be violated.

The residuals (i.e., the error term) from

2 King (1966) and Meyers (1972).

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 6: Accounting Earnings and Stock Prices of Insurance Companies

690 The Accounting Review, October 1975

TABLE 1

CROSS-SECTIONAL DISTRIBUTION OF PRICE REGRESSION STATISTICS

(1st Statistic Is Mean; 2nd Statistic in Brackets Is Standard Deviation)

Durbin- Durbin- Watson Watson

Adj.R2 Adj.R2 Statistic Statistic (industry (including (industry (including variable industry variable industry

a! 1 32 suppressed) variable) suppressed) variable)

FIRE-CASUALTY 0.010 0.868 0.959 0.141 0.477 2.137 1.759 (.007) (.242) (.190) (.064) (.301) (.240) (.679)

LIFE-HEALTH -.0002 1.056 .991 .166 .448 2.229 2.272 (.006) (.357) (.303) (.083) (.148) (.197) (.238)

(1) are aggregated to measure the abnor- mal security return. First, for each firm the residuals are accumulated for the 12 months up to the earnings announcement date. These cumulative residuals are then averaged cross-sectionally across all firms (CAR) = Cumulative Average Residual.

1 N 12

(CAR) = - Uit N j=1 t=1

Earnings announcements are assumed to be issued in the third month following the end of the calendar year. This is the month insurance companies are required to file their Convention Statement with state in- surance commissioners.'

Measurement of Earnings Changes

Three sub-earnings series of insurance companies will be separately examined: (1) underwriting; (2) investment-com- prising mainly interest coupons from bonds or mortgages and dividends from common stocks; and (3) capital gains and losses (both realized and unrealized) on marketable equity securities. Positive and negative earnings changes are estimated via a linear regression of each firm's earn- ings on the average of the earnings of the other firms in the sample, i.e., via an "ac- counting market model." This model is estimated in both the levels and the first differences versions.

E,=a + ?Imt + ?it (2a)

A =it = A + ?,A .n.t + fit (2b)

E(it)= 0 E(Pmt iit) = 0 and E(APmt*iit) = 0

{02 ifk #1 E(i Elk) = 4f2 if I= k

where

Pit= the earnings variable of the ith firm in period t

Pmt = the earnings variable for the repre- sentative index in period t

A= a first differences operator

The representative index for each earnings variable is constructed (assuming equal weighting) from the firms in the study. Separate indexes are constructed for the fire-casualty and life-health samples of firms. A firm is classified as having a posi- tive earnings change if in period t the re- sidual in (2) is positive; a negative re- sidual would indicate a negative earnings change in period t.

The index in (2a) and (2b) will capture

3 Obtaining estimates of the actual announcement dates of earnings for the sample of firms is difficult. For instance, most NYSE firms' earnings announcements are reported in the Wall Street Journal. Very little in- formation on the firms in this study is published in this source, e.g., for only 7 of the 51 life-health companies and 4 of the 22 fire-casualty companies were announce- ments of 1970 annual earnings reported in the Wall Street Journal Index.

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 7: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 691

industry-wide commonalities and those economy-wide commonalities captured by the industry index. For the underwriting earnings series of insurance companies, economy-wide commonalities not cap- tured by the index could be a potential source of cross-sectional correlation. In the sample, however, this additional source of cross-sectional correlation is quite minor. In every case examined, the adjusted R2 for (2a) and (2b) with an additional vari- able representing economy-wide earnings is less than the adjusted R2 for the models without the additional variable. All earn- ings variables are for rates of return series. The use of a deflated series is necessitated due to mergers in the insurance industry (Brown and Ball, 1967).

Separate analysis of sub-earnings series is important from a methodological view- point for several reasons.4 First, it enables the dependent and independent variables in (2a) and (2b) to be relatively homoge- neous; i.e., underwriting earnings of firm i are regressed against average underwriting earnings of the other firms, etc. If (2a)and (2b) were estimated for only the aggregate of the three sub-earnings series, the index (A3 or i.EMt) would represent the average mix of underwriting, investment, and capital gains for the industry. If there are substantial differences between firms in the proportion of aggregate earnings de- rived from these three sources, the esti- mate of A in (2a) and (2b) could be down- ward biased. The second reason for sepa- rate analysis of sub-series is the problem caused by the sub-series having different changes in sign. Suppose the security price revaluation implications of an under- writing earnings change and an investment earnings change exactly offset each other. In this case a test based on the aggregate of the two series would report no associa- tion, whereas a test based on the separate series would be able to detect the revalua- tions associated with these different earn-

ings changes. The final reason for separate analysis is that one hypothesis being ex- amined relates specifically to the relative associations of the statutory and adjusted underwriting measures. By separately esti- mating (2a) and (2b) for each underwriting measure, the analysis can focus on this hy- pothesis more directly than if (2a) and (2b) had been estimated with the aggre- gate earnings series.

The mean and standard deviation of parameters estimated with equations (2a) and (2b) and summary statistics are pre- sented in Table 2. Columns (1) to (4) re- port these statistics for fire-casualty com- panies; columns (5) to (8) contain the life- health statistics. The statistics in Table 2 are for the 1972 earnings period (i.e., esti- mated with 1964-71 data). The statistics for other earnings periods are similar to those presented. Evidence from prior studies (e.g., Gonedes, 1973) suggests that (2b) is a better-specified linear model than (2a). In general, the statistics (e.g., Dur- bin-Watson Statistic) in Tables 3 and 4 are consistent with this prior evidence. Note, however, the cross-sectional average O3's of the differences model for statutory and adjusted life-health underwriting earn- ings do not approximate unity.5 Note, also, the large standard deviations of d in the earnings regressions (Table 2) relative to the standard deviations of id and 12 in the security return regressions (Table 1). This greater estimation efficiency in Table 1 is

I Strictly speaking, the following analysis assumes that earnings changes on each sub-series are independent of earnings changes on the other sub-series. In the Appendix to this paper it is argued that violations of the independence assumption in the sample do not change the substance of the results. See also Foster (1974).

5 If the assumptions of the linear model are met for each firm, and the index is constructed from firms in the sample, the average intercept should approximate 0 and the average f3 should approximate 1. See Gonedes (1973).

Note also the large standard deviation of the differ- ences model for the life-health investment series. This is due to one outlier. After deleting this company, the average ,3 is 0.973 with a cross-sectional standard devi- ation of 1.462.

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 8: Accounting Earnings and Stock Prices of Insurance Companies

692 The Accounting Review, October 1975

TABLE 2

CROSS-SECTIONAL DISTRIBUTION OF EARNINGS REGRESSION STATISTICS

(1st Statistic Is Mean; 2nd Statistic Is Standard Deviation)

FIRE-CASUALTY LIFE-HEALTH

Col.4 Col. 8 Durbin- Durbin-

Col. 1 Col. 2 Col. 3 Watson Col. 5 Col. 6 Col. 7 Watson a f3 Adj. R2 Statistic aC A dj. R2 Statistic

Levels Statutory .0005 .900 .289 1.640 -.004 1.005 .456 1.773

Underwriting (.059) (.634) (.309) (.568) (.009) (2.020) (.335) (.509)

Best's Adjusted .0032 .876 .303 1.638 .101 .860 .162 1.405 Underwriting (.059) (.811) (.308) (.549) (.239) (3.657) (.255) (.490)

Investment .005 .968 .775 1.694 .003 .985 .635 1.898 (.055) (.641) (.239) (.478) (.191) (1.487) (.371) (.477)

Capital Gains and Losses -.0002 .991 .789 2.166 -.002 1.022 .497 1.823 (.018) (.317) (.343) (.781) (.084) (.608) (.392) (.845)

Differences Statutory Underwriting .0007 .906 .232 2.168 -.004 .244 .014 2.056

(.016) (.657) (.308) (.500) (.017) (3.108) (.258) (.548)

Best's Adjusted .0009 .854 .179 2.234 -.013 .255 .033 2.156 Underwriting (.103) (.826) (.328) (.504) (.011) (1.989) (.198) (.570)

Investment .0002 .935 .386 1.875 -.002 1.206 .062 2.102 (.001) (.750) (.321) (.495) (.053) (5.499) (.282) (.778)

Capital Gains and Losses .00003 .994 .801 2.304 -.005 1.180 .533 2.352 (.007) (.335) (.354) (.639) (.023) (1.098) (.386) (.798)

believed to be due to using monthly se- curity return data to estimate (1)-88 ob- servations-and using annual earnings data to estimate (2a) and (2b)-8 observa- tions. This relatively large amount of sam- pling error in (2a) and (2b) is, however, not expected to produce biased estimates of positive or negative earnings changes.

RESULTS FOR SEPARATE EARNINGS SERIES

Results for the complete sample over the 1965-72 period are presented in Table 3 (columns 1-4). Columns (1) and (2) are the CAR's for the positive and negative earnings change groups respectively. Col- umn (3) is the composite CAR-this as- sumes an investment policy of buying long the positive earnings change firms and selling short the negative earnings change

firms. Column (4) presents the Mann- Whitney U Test (see Hollander and Wolfe, 1973) for statistical significance. This test requires that the CAR's for the 12-month period up to each earning's announcement be ranked from the highest to the lowest. Then the average ranks for firms with posi- tive and negative changes are computed. The null hypothesis of no association be- tween earnings changes and stock price changes implies that the average rank of the two earnings groups is the same.

Underwriting earnings has the strongest association with stock price changes, e.g., the composite CAR for adjusted under- writing is 0.054 for the levels model (2a) and 0.051 for the differences earnings model (2b). The investment series has a lower CAR than underwriting, but the de- gree of association is still statistically sig-

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 9: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 693

TABLE 3 CUMULATIVE AVERAGE RESIDUAL RESULTS

COMPLETE SAMPLE FIRE-CASUALTY LIFE-HEALTH

Col. I Col. 2 Col. 3 Col. 4 Col. 5 Col. 6 Col. 7 Col. 8 Col. 9 Col. 10 Col. 11 Col. 12 Positive Negative Com- Mann- Positive Negative Com- Mann- Positive Negative Com- Mann-

Earnings Earnings posite Whitney Earnings Earnings posite Whitney Earnings Earnings posite Whitney Change Change CAR U Change Change CAR U Change Change CAR U

Levels Statutory

Underwriting .036 -.036 .036 3.116*** .058 -.059 .058 2.845*** .025 -.026 .025 1. 804* Best's Adjusted

Underwriting .056 -.053 .054 5.283*** .057 -.058 .058 3.050*** .055 -.051 .053 4.308*** Investment .030 -.027 .028 2.374*** .058 -.041 .048 2.593*** .019 -.017 .017 1.140 Capital Gains

and Losses .012 -.010 .011 l.654** .015 -.015 .015 1.18 .010 -.008 .009 -.001 Differences

Statutory Underwriting .049 -.050 .050 4.123*** .067 -.071 .069 3.266*** .041 -.041 .041 2.769***

Best's Adjusted Underwriting .050 -.052 .051 4.628*** .047 -.050 .048 1. 974** .052 -.053 .052 4.160***

Investment .024 -.023 .023 2.311** .038 -.037 .037 2.157** .017 -.017 .017 1.367* Capital Gains

and Losses .008 -.011 .009 .943 -.006 .013 -.009 -.560 .015 -.014 .014 1.057

Significance Levels .01*** 2.33 .05** 1.64 . 10* 1.28

nificant at the 0.05 level. Of the three earnings series, capital gains and losses has the lowest CAR and the association is sta- tistically less significant.6 Columns 5-12 contain an industry breakdown of the re- sults-columns 5-8 for fire-casualty firms and columns 9-12 for life-health firms. For both sets of firms the ranking of the three sub-earnings series, in terms of strength of association, is the same, i.e., (1) under- writing, (2) investment, and (3) capital gains and losses.

One method of quantifying the relative association of these sub-earnings series is the clairvoyant technique used by Ball and Brown (1968). For the years 1965-72, fore- knowledge of the sign of the year's security price change enables an abnormal return of 17.7%0. The proportion of the 17.7%0 re- turn that can be explained by foreknowl- edge of the sign of the sub-earnings change (model 2a) is:

Underwriting-30%0 Investment-16%o Capital Gains and Losses-6%o

Thus, changes in the three sub-earnings series jointly explain approximately 50%0

of the abnormal price changes in insurance stock prices over the 1965-72 period.7

A djusted vs. Statutory Underwriting

In many cases, the two underwriting measures imply the same earnings change. This reflects the many elements that the two measures have in common. To gain a better picture of possible differences be- tween the two measures, CAR's were com- puted on those instances where the changes in statutory and adjusted earnings differed in sign.

In the 1965-68 period, reference to ad- justed life-health measures in annual re- ports was restricted to footnotes, chair- man's address, etc. Over the 1969-72 pe- riod, there was both an increasing use of adjusted measures in the financial state-

6 Several reasons for the low CAR for the capital gains and losses series are discussed in Foster (1974). These include: (a) the time-series behavior of the capital gains and losses series and (b) the change on the capital gains and losses series being highly correlated with changes in the market portfolio.

7More detailed analysis of CAR specification issues is in Foster (1974). This analysis separately allowed for (a) nonstationarities in systematic risk and (b) an additional commonality in security price returns (the return on the "zero-beta" asset). In both cases the results are similar to those reported in this paper.

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 10: Accounting Earnings and Stock Prices of Insurance Companies

694 The Accounting Review, October 1975

ments and an increasing reference to the adjusted measures in footnotes, etc. In both these periods, the CAR from follow- ing adjusted underwriting is positive.

CAR Z Statistic Follow- (Mann- No. of

ing Whitney Obser- Adjusted Test) nations

Levels 1. 1965-68 .049 1.947** 57 2. 1969-72 .037 1.446* 69

Differences 1. 1966-68 .010 .555 40 2. 1969-72 .034 1.678** 48

There does not appear to be a mechanistic relation between the reported statutory underwriting measure and stock price changes in the 1965-68 period. Indeed, in both the 1965-68 and the 1969-72 periods, the adjustments A. M. Best made to statu- tory are at least consistent with those the aggregate market was making to the statu- tory life-health underwriting measure.

For the fire-casualty group, there is little difference between the adjusted and statu- tory measures for the levels model. For the differences model, statutory has a stronger association. In this model, however, there is a marked sub-period behavior for the 1966-68 and 1969-72 periods. The relevant data for both models over these sub- periods are:

U Statistic CAR (Mann- No. of

Following Whitney Obser- Adjusted Test) nations

Levels 1. 1965-68 .010 32 15 2. 1969-72 -.010 55 21

Differences 1. 1966-68 -.107 8** 13 2. 1969-72 -.012 49 22

The only case where statutory and ad- justed earnings are significantly different is for 1966-68 using the differences model.8 In interpreting the above results, the ex- tremely small samples should be noted.

Indeed, given the small samples, it is diffi- cult to attribute economic significance to the comparisons between statutory and adjusted underwriting measures for fire- casualty firms.

SEMI-STRONG EFFICIENCY OF THE MARKET

FOR INSURANCE STOCKS9

Prior studies into the relationship be- tween accounting earnings and security price changes have found that once annual earnings are publicly announced, there is little difference between the security price behavior of firms with positive earnings changes and firms with negative earnings changes. This result is consistent with a semi-strong efficient market in which new information is rapidly and unbiasedly im- pounded into security prices.

Similar conclusions appear to hold for the market for insurance stocks over the 1965-71 period. (As the analysis in this section concentrates on the CAR for the 12-month period subsequent to each earn- ings announcement, the 1972 earnings an- nouncement could not be examined.) Fig- ure 1 presents results for the 73 firms for the adjusted underwriting and investment series (using model 2a). The behavior of the CAR for the other sub-earnings series, and for all sub-earnings series with model

8 This case is the only statistically significant one (at the 0.05 level) for fire-casualty sub-periods. The Mann-Whitney U statistic, rather than the Z statistic (normal approximation), is appropriate due to the small samples.

I Strictly speaking, the above analysis assumed semi- strong efficiency of the market for insurance stocks. That is, in interpreting the CAR as a measure of associ- ation, it is assumed that abnormal security price changes reflect new information becoming available rather than new insights into previously publicly avail- able information. The results in this section are con- sistent with the assumption of semi-strong efficiency.

It is important to note that in measuring "associ- ation" the behavior of the residuals from (1) in the twleve months up to and including the earnings an- nouncement month are analyzed. In examining semi- strong efficiency the behavior of the residuals from (1) in the twelve months subsequent to the earnings an- nouncement month are analyzed. That is, the same data period is not used to examine both "association" and market efficiency.

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 11: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 695

FIGURE 1

TIME-SERIES BEHAVIOR OF CUMULATIVE AVERAGE RESIDUALS

A. ADJUSTED UNDERWRITING

Positive Earnings Changes

+05 ++ ? + ?t + t

W ~~~~~+ a d + ] j

-.05tt +++++ + ++

_ _ ~~~~~~~~~~~~Negative Earnings Changes , I I I I I j t I ~~~ ~ ~~~ ~~~~~~~I I . I I I

-12 -9 -6 -3 0 +3 +6 +9 +12 Month from Earnings Announcement Date

B. INVrESTMENT

+.04 positive Earnings Changes

-12 -9 -6 -3 0 +3 +6 +9 +12 Month from Earnings Announcement Date

+.04~~

+

++ + U -.02 +~~~~~~~~~~~~~ +

Negative Earnings Changes

-12 -9 -6 -3 0 +3 +6 +9 +12

Month from Earnings Announcement Date

(2b), correspond closely to those in Figure 1.

For all earnings series examined, the stock market is progressively better able to discriminate between positive earnings change firms and negative earnings change

firms in the 12 months prior to the earnings announcement. In the 2-3 months after the month companies must file their statu- tory reports, the residuals have a slight positive drift for positive change firms and a slight negative drift for negative change

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 12: Accounting Earnings and Stock Prices of Insurance Companies

696 The Accounting Review, October 1975

firms. A trading strategy based on this drift, however, would yield a maximum abnormal return of 1%0.

A 12-month trading strategy, based on the earnings sign change (model 2a) and starting after the annual earnings are an- nounced, would yield:

0.007 abnormal return for statutory un- derwriting,

zero abnormal returns for adjusted un- derwriting,

zero abnormal returns for investment, and

0.010 abnormal returns for capital gains and losses.

In general, these results are consistent with a semi-strong efficient market for in- surance stocks.'0

CONCLUSIONS

At the methodological level, this study differs in two main respects from prior "residual-type" studies in accounting. First, there is more explicit analysis of in- dustry-wide commonalities in both secur- ity returns and accounting earnings than has been undertaken in prior studies. This explicit analysis was necessitated by the independence assumptions of the statisti- cal significance tests used in the paper. Second, this study has decomposed the total earnings number into several sub- earnings series. This decomposition en- abled more effective analysis of two ac- counting issues in the insurance industry.

The results provide some evidence on the underwriting earnings issue. They fail to support the arguments advanced by many accountants that the stock prices of life-health companies were adversely af- fected by the use of statutory underwriting measures in financial reports. The aggre- gate market appears to behave as if it made adjustments to the life-health statu- tory measure in the years before the ac- counting profession required the reporting

of GAAP underwriting earnings. This finding, however, does not necessarily im- ply that an adjusted underwriting measure is the preferred reporting policy. Making preference rankings on financial reporting alternatives involves difficult issues of social choice that are beyond the scope of this paper (see Demski, 1974).

This study also provides some evidence on the marketable securities issue. One argument against the inclusion of annual capital gains and losses in the income statement, rather than a separate surplus statement, was that the income statement alternative "may be a damaging factor to the orderly functioning of the stock mar- ket" and may cause "the stock of the com- pany to decline when it shouldn't." The evidence in this paper suggests there is a small likelihood of such consequences. The above insurance industry argument im- plies a semi-strong inefficient market. My results, however, are consistent with the market for insurance stocks being reason- ably semi-strong efficient. In such a mar- ket, the substantive issue (as regards security price determination) is not how to report capital gains and losses on market- able equity securities, but whether to dis- close them at all.

APPENDIX

SEPARATE ANALYSIS OF SUB-EARNINGS SERIES WITH CORRELATION BETWEEN

SUB-EARNINGS CHANGES

In interpreting the CAR as an indicator of the degree of association between stock prices and earnings changes for each sub- earnings series, an internal validity prob- lem may exist. This problem is due to possible correlations between the earnings changes of the various sub-earnings series.

10 The 73 firms include 10 NYSE and 63 OTC firms. The behavior of the cumulative residuals from (1) for the OTC firms are very similar to those in Figure 1. These results are reported in Foster (1974).

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 13: Accounting Earnings and Stock Prices of Insurance Companies

Foster: Earnings and Stock Prices 697

Suppose the classification of earnings changes for two sub-earnings series (with 300 observations) are those below:

Investment Earnings

Positive Negative Change Change

Underwriting Earnings Positive

Change + 100 65

Negative Change- 40 95

In this case there is a strong association between positive (negative) changes for in- vestment and positive (negative) changes for underwriting. Even if only one of these two series (say, underwriting) had a strong association between earnings changes and price changes, it is possible that the CAR test on both series would show significant results. The CAR for investment earnings could be significant merely because of the strong correlation between its earnings changes and those for changes in under- writing earnings.

As a preliminary check on this problem, correlations between earnings changes on combinations of the sub-series were ex- amined. The only combination to have significant (at the 0.05 level using a chi- square test) correlation between earnings changes was the investment and capital gains and losses series.

As a further check on this problem in the sample the following technique was adopted to control for possible correlations between earnings changes when estimating the CAR for each series.

a. The number of observations were computed for each cell of the earnings change matrix for the series being compared.

b. The cell with the lowest number of observa- tions was used as a base. In the above example, this cell is the -underwriting/+investment category (40 observations).

c. From each other cell, observations were ran- domly deleted until the number of observa- tions in each cell was equal to that in the cell with the smallest number of observations, i.e., 60 were deleted from +underwriting/+invest- ment, 25 from +underwriting/-investment, and 55 from -underwriting/-investment.

d. The CAR for underwriting earnings was then computed for the reduced sample. Steps (a), (b), and (c) enable examination of the associ- ation between underwriting earnings changes and stock price changes, holding changes in investment earnings constant.

e. Steps (a)-(d) were repeated 50 times to damp- en out fluctuations in the CAR due to any particular set of firms being randomly chosen in any one trial. The estimated CAR is the average CAR over these 50 trials.

The maximum difference between the esti- mated CAR from the above procedure and the CAR reported in Table 3 was 0.005 (see Foster, 1974). None of the conclusions in this paper was altered when possible correlations between earnings changes were controlled for.

REFERENCES

A.I.C.P.A. Committee on Insurance Accounting and Auditing, Audits of Fire and Casualty Companies, A.I.C.P.A., 1966.

, The Audit Guide for Stock-Life Insurance Companies, A.I.C.P.A., 1972. A.P.B. Public Hearing on Accounting for Investments in Equity Securities Not Qualifying for the Equity

Method (Arthur Andersen, 1971). Ball, R., "Changes in Accounting Techniques and Stock Prices," Empirical Research in Accouliting:

Selected Studies, 1972. Supplement to Journal of Accounting Research, 1972. and Brown, P., "An Empirical Evaluation of Accounting Income Numbers," Journal of Account-

ing Research (Autumn 1968). Beaver, W. H., and Dukes, R. E., "Interperiod Tax Allocation and a-Depreciation Methods: Some

Empirical Results," THE ACCOUNTING REVIEW (July 1973). Brown, P., and Ball, R., "Some Preliminary Findings on the Association Between the Earnings of a

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions

Page 14: Accounting Earnings and Stock Prices of Insurance Companies

698 The Accounting Review, October 1975

Firm, Its Industry and the Economy." Empirical Research in Accounting: Selected Studies, 1967. Supplement to Journal of Accounting Research (1967).

Demski, J., "Choice Among Financial Reporting Alternatives," THE ACCOUNTING REVIEW (April 1974). Foster, G., "The Association Between Accounting Earnings and Stock Prices of Insurance Companies,"

unpublished Ph.D. dissertation, (Stanford University, 1974). Gonedes, N. J., "Properties of Accounting Numbers: Models and Tests," Journal of Accounting Research

(forthcoming). Hagerman, R. L., "The Efficiency of the Market for Bank Stocks: An Empirical Test," Journal of

Money, Credit and Banking (August 1973). Hollander, M. and Wolfe, D. A., Nonparametric Statistical Methods (Wiley, 1973). Horngren, C., "The Marketing of Accounting Standards," Journal of Accountancy (October 1973). Kaplan, R. A. and Roll, R., "Investor Evaluation of Accounting Information: Some Empirical Evi-

dence," Journal of Business (April 1972). King, B., "Market and Industry Factors in Stock Price Behavior," Journal of Business (January 1966). May, R. G., "The Influence of Quarterly Earnings Announcements on Investor Decisions as Reflected

in Common Stock Price Changes," Empirical Research in Accounting: Selected Studies, 1971, Supple- ment to Journal of Accounting Research, 1971.

Meyers, S. L., "A Re-Examination of Market and Industry Factors in Stock Price Behavior," Journal of Finance (June 1973).

Sharpe, W., Portfolio Theory and Capital Markets (McGraw-Hill, 1970). Sunder, S., "Relationship Between Accounting Changes and Stock Prices: Problems of Measurement

and Some Empirical Evidence," Empirical Research in Accounting: Selected Studies, 1973, Supplement to Journal of Accounting Research, 1973.

Waterfield, R. H., "Our Approach to Serving the Life Insurance Industry," Arthur Young Journal (Winter 1974).

This content downloaded from 62.122.79.31 on Sat, 14 Jun 2014 16:13:44 PMAll use subject to JSTOR Terms and Conditions


Recommended