A test of the market’s mispricing of domestic and foreign earnings

The results indicate the market understates foreign earnings'persistence, causing a positive relation between current changes in foreign earnings and future abnormal stock returns.. For

A test of the market’s mispricing of domestic and foreign earnings

Wayne B Thomas

School of Accounting, University of Utah, Salt Lake City, UT 84112, USA

Received 13 May 1999; received in revised form 17 March 2000

Abstract

This study investigates whether abnormal returns can be earned using public information about firms' domestic and foreign earnings The results indicate that the market understates foreign earnings’ persistence As a result, it is possible to construct a zero-investment hedge portfolio that consistently earns positive returns across years A disproportionate fraction of the positive abnormal returns to the long position is concentrated in the few days surrounding the subsequent year's quarterly earnings announcement dates Furthermore, the abnormal returns do not appear to persist beyond the subsequent year The results are consistent with market

mispricing, and not mis-estimated risk

JEL classification: F23; G14; M41

Key words: Capital markets; Market efficiency; Valuation; Multinational firms; Foreign

earnings

*Corresponding author Tel: (801) 581-8790; fax: (801) 581-7214; e-mail: actwbt@business utah.edu

I wish to thank Neil Bhattacharya, Dan Collins, Peter Easton, Don Herrmann, Marlene Plumlee, J Riley Shaw, James Wahlen and seminar participants at Arizona State University, Oklahoma State University, and the University

of Utah for helpful comments relating to the paper I am especially thankful for the comments provided by S.P Kothari (the editor) and Art Kraft (the referee) that greatly improved this research Eugene Fama and Mark Carhart generously provided factor model data.

1 Introduction

This study investigates whether abnormal returns can be earned using public information about firms’ domestic and foreign earnings Specifically, I test whether the market accurately incorporates the pricing effects of the persistence of the domestic and foreign earnings

components of total earnings reported by a firm The tests in this study add to the existing literature on post-earnings-announcement drift, the pricing of accruals, and the pricing of

domestic and foreign earnings components The results indicate that the market understates foreign earnings’ persistence and that positive abnormal returns can be earned using a trading strategy based on changes in foreign earnings Further analysis indicates that the abnormal returns do not appear to be the result of risk mis-estimation

SEC Regulation §210.4-08(h), General Notes to Financial Statements – Income Tax Expense, requires firms to disclose components of income (loss) before income tax expense (benefit) as either domestic or foreign While domestic income refers to a single country (i.e., the United States), foreign income encompasses countries from around the world differing drastically in terms of economic conditions, political stability, competitive forces, growth

opportunities, governmental regulations, etc Therefore, Rule 4-08(h) may provide only limited information regarding the risks and opportunities of the firm’s foreign earnings

Guidelines set forth in Statement of Financial Accounting Standards No 14, Financial Reporting for Segments of a Business Enterprise (FASB 1978) (SFAS 14) require firms to go

beyond a simple breakdown of earnings into foreign and domestic categories SFAS 14 requires that firms disclose earnings by geographic area (e.g., Canada, Europe, Asia/Pacific), potentially providing information beyond that required by Rule 4-08(h) Many complain, however, that geographic segment earnings disclosures are not useful because of (1) the lack of comparability

and consistency in segment definition both across firms and over time for the same firm, (2) insufficient disaggregation, (3) failure to group foreign operations according to similar risk and return characteristics, and (4) management manipulation through transfer pricing policies and common cost allocations These criticisms come from both the financial community (e.g., Association of the Institute for Management Research 1992 and American Institute of Certified Public Accountants 1994) and the academic community (Bavishi and Wyman 1980, Arnold et al.

1980, Roberts 1989, Balakrishnan et al 1990, Boatsman et al 1993, and Herrmann 1996).1, 2 If the market fails to understand the time-series properties of domestic or foreign earnings, then stock prices will systematically understate/overstate the value of the firm in a predictable

manner That is, if the market perceives the persistence of domestic or foreign earnings to be different than their historical time-series patterns, then stock prices will move in a predictable manner in the subsequent year

A growing body of literature questions the market’s efficiency with respect to earnings and other accounting information (for reviews see Ball 1992, Bernard et al 1993, and Bernard et

al 1997) Most of this research centers on the premise that the market does not fully understand the time-series behavior of earnings or its components For example, Bernard and Thomas (1989, 1990) show that the well-documented post-earnings-announcement drift is characteristic

of a market that expects (naively) seasonal changes in quarterly earnings to follow a random walk process, even though the time-series pattern over the past several decades shows that this is not the case When subsequent earnings are released, the market acts “surprised” and stock prices move in a predictable direction and magnitude

1 See Pacter (1993) for a thorough discussion of the alleged shortcomings of segment reporting practices.

2 The FASB recently issued SFAS 131, Disclosures about Segments of an Enterprise and Related Information, which supersedes SFAS 14 However, the new statement appears to have reduced the quantity and quality of

disclosure of foreign operations compared to that provided under SFAS 14 (Herrmann and Thomas 2000).

In related work, Sloan (1996) investigates the accrual and cash flow components of earnings Sloan (1996) finds that even though the cash flow component of earnings persists into future earnings more heavily than the accrual component, stock prices act as if the opposite were true In the subsequent period, the market appears to revise its prior (incorrect) belief in a predictable manner Furthermore, these adjustments are concentrated around future earnings announcements.

Abarbanell and Bushee (1997, 1998) examine nine of the twelve fundamentals signals identified in Lev and Thiagarajan (1993) Abarbanell and Bushee (1997) find that (some of) these fundamentals are significantly associated with future earnings but analysts tend to

underreact to the fundamental signals (i.e., analysts do not fully understand the impact that these fundamental signals have on future earnings) Abarbanell and Bushee (1998) devise an

investment strategy based on these fundamental signals that earns significant abnormal returns inthe following period The market tends to underreact to fundamental signals about future

earnings and in the subsequent period when earnings are different than expectations, the market corrects its apparent mispricing in a predictable manner

This study adds to the existing literature by testing whether the market correctly

incorporates the pricing effects of the persistence of domestic and foreign earnings components

of total earnings reported by a firm The results indicate the market understates foreign earnings'persistence, causing a positive relation between current changes in foreign earnings and future abnormal stock returns As with any study in this area, conclusions should be made with the caveat that the apparent abnormal returns could be the result of the researcher’s inability to adequately measure and control for underlying risk factors The change in foreign earnings may

be a proxy for (or source of) risk, and the positive relation with abnormal returns is the result of unidentified risk premia and not market mispricing.

To help disentangle these two competing hypotheses, I employ three additional tests

First, Bernard et al (1997) suggest that any risky hedge portfolio that requires zero-investment

should produce positive returns in some years and negative returns in other years A

zero-investment hedge portfolio that consistently produces positive returns is more likely to be the result of market mispricing and not unidentified risk Using the relation between changes in totalearnings and foreign earnings, a zero-investment hedge portfolio consisting of firms expecting to

do well (poorly) is constructed This hedge portfolio produces positive abnormal returns in nine out of ten years, which supports the conclusion of market mispricing and not omitted risk factors

The second test suggested by Bernard et al (1997) involves observing market reactions tofuture earnings announcements If abnormal returns are the result of a market that does not fully understand the persistence of current earnings (or its components), then market corrections are most likely to occur when future earnings are announced As such, abnormal returns should be concentrated in the few days surrounding subsequent earnings announcements I do find that the abnormal returns to the long position of the hedge portfolio are concentrated in the few days surrounding the subsequent year’s earnings announcements For firms that experience a large, positive increase in foreign earnings in year t (holding the change in total earnings constant), a disproportionately large, positive reaction occurs in the few days surrounding the quarterly earnings announcements in year t+1 This is characteristic of a market that underestimates the persistence of foreign earnings and corrects for this mispricing in the subsequent year when earnings are announced higher than expected

The final test involves estimating the relation between long-term stock returns and current changes in foreign earnings Long-term stock returns are defined as stock returns two or three years in the future If the change in foreign earnings is a proxy for (or source of) risk, then one might expect abnormal returns to persist beyond the subsequent year A permanent shift in the firm’s systematic risk will be related to higher returns in subsequent years If, however, the market does not fully understand the persistence of foreign earnings, then abnormal returns should exist in the immediate subsequent year only and should not continue It is not likely that mispricing could occur for several subsequent years The market will correct for its (incorrect) prior belief when earnings are realized above or below expectations in the subsequent year The results in this study show no relation between long-term stock returns and current changes in foreign earnings The market appears to correct fully for its mispricing in the subsequent year sothat abnormal returns do not persist for more than one year All three tests support the notion that the abnormal returns are the result of market pricing, and not mis-estimated risk.

The paper proceeds as follows The research design is outlined in section 2 The sample selection is discussed in section 3 The results of the primary and supplemental tests are reported

in section 4 and the paper concludes in section 5

2 Research Design

The primary focus of this paper is to test whether the market correctly prices

multinational firms’ securities relative to the persistence of the domestic and foreign earnings components To test this, I use the Mishkin (1983) framework Mishkin (1983) devises a test to determine whether the market rationally prices information In the context of this study, the test for market rationality would involve simultaneously estimating the following equations.3

3 For a more extensive, generalized discussion of this test, see Mishkin (1983) or Sloan (1996).

1 0

abnormal return in year t+1

Equation (1) represents the actual time-series relation of changes in domestic and foreign

earnings to future changes in total earnings (i.e., the extent to which changes in domestic and foreign earnings persist into changes in total earnings in the subsequent year) D represents a measure of the persistence of changes in domestic earnings and F represents a measure of the persistence of changes in foreign earnings, controlling for one another A slope coefficient equal

to -1 would suggest that earnings are purely transitory whereas a slope coefficient equal to 0 would suggest that earnings follow a random walk A slope coefficient greater than 0 would indicate growth in earnings If F is greater than D, then foreign earnings are considered to be more persistent than domestic earnings

Equation (2) estimates the relation between unexpected movements in stock prices and the unexpected portion of the change in total earnings The expected change in total earnings is based on last year’s change in domestic and foreign earnings Mishkin (1983) suggests that the

second equation provides an estimate of the market’s perceived time-series behavior of domestic

and foreign earnings The notion is that unexpected movements in stock prices in the current period are related only to unexpected information received that same period Therefore, the second equation can be used to estimate the market’s perception of the unexpected change in total earnings based on the change in domestic and foreign earnings in the previous year *is

an estimate of the extent to which the market perceives changes in domestic earnings to persist

into future years Likewise, * is an estimate of the extent to which the market perceives

changes in foreign earnings to persist into future years Since domestic and foreign earnings are public information, market efficiency requires that D = * and F = * If either equality does not hold, then the market’s perception of earnings persistence differs from the historical time-series pattern

The two equations are estimated simultaneously using non-linear least squares linear least squares is required because of equation (2) The equality of the coefficients across equations is tested using the likelihood ratio statistic suggested by Mishkin (1983)

Non-) (

~ ) /

log(

2n SSR C SSR U 2 q

 (3)where n = the number of observations, SSRC = the sum of squared residuals from the constrainedweighted system, SSRU = the sum of squared residuals from the unconstrained weighted system, and q = the number of constraints imposed by market efficiency A significant Chi-square statistic would suggest that the coefficients are not equal across equations and market efficiency would be rejected

3 Sample Selection, Variable Measurement, and Descriptive Statistics

The sample consists of all firms that have the necessary data available in the intersection

of the 1998 versions of the Compustat annual industrial and research files and the CRSP monthlystock returns file Compustat contains data on firms’ foreign and domestic earnings back to

1984 Since the tests require the change in foreign earnings, 1985 is the first year of usable observations The sample is restricted to U.S multinational firms so that domestic and foreign earnings are reported for each firm and have the same meaning across sample firms To control for influential observations, any observation that has a change in total earnings, change in

domestic earnings, or change in foreign earnings (scaled by average total assets) in year t greater than 25 is deleted This criterion resulted in approximately 4% of the observations being

deleted The final sample consists of 8,051 firm-year observations over the 1985-1995 period

No control for extreme observations of returns or earnings in t+1 is made since this would introduce hindsight bias in the results

Compustat reports both domestic and foreign earnings on a pretax basis (Compustat data item #272 and data item #273, respectively) To be consistent, total earnings is defined as pretaxincome (data item #122) To control for differences in size across firms and over time, all earnings variables are scaled by average total assets (data item #7) Average total assets are defined as total assets in year t plus total assets in year t-1, divided by 2.4

The Mishkin (1983) test requires the use of abnormal returns As in Daniel, Grinblatt, Titman, and Wermers (1997), I calculate abnormal returns using a comparison portfolio

approach This approach entails matching a firm’s security return with the value-weighted return

of a portfolio consisting of firms that are similar in size, book-to-market-ratio, and prior year return The comparison portfolios are created by first selecting all stocks that have year end capitalization values available on CRSP and book value data available on COMPUSTAT Only firms with positive book values are included These stocks are then sorted into quintiles based

on their capitalization values at the beginning of the year of portfolio formation NYSE

breakpoints are used so that there are an equal number of NYSE stocks in each quintile Next, within each of the size quintiles, stocks are sorted into quintiles based on their industry-adjusted

4 For some firms, domestic earnings plus foreign earnings does not equal total earnings To test the sensitivity of the results to this inequality, observations were eliminated if domestic earnings and foreign earnings did not sum to within 1% of total earnings (scaled by average total assets) This resulted in approximately 2.2% of the sample observations being eliminated The results for this reduced sample are similar to those reported The effect of this inequality was also tested by including the difference between total earnings and domestic plus foreign earnings (i.e., “nonallocated” earnings) as an additional explanatory variable Inclusion of this additional explanatory variable has no qualitative effect on the reported results.

book-to-market ratios The book-to-market ratio is defined as the book value at the end of the fiscal year prior to the year of portfolio formation divided by capitalization value at the

beginning of the calendar year of portfolio formation The book-to-market ratio is adjusted by subtracting the mean industry book-to-market ratio over the sample period from the individual stock's book-to-market ratio, where industries are defined along two-digit SIC codes Finally, within the 25 size/book-to-market portfolios, stocks are sorted based on their prior twelve-month return ending one month prior to portfolio formation Excluding the month just prior to the portfolio formation date avoids problems associated with the bid-ask spread bounce and monthly return reversals (Jegadeesh 1990) Thus, there are 125 size/book-to-market/prior year return portfolios for each fiscal year

industry-Abnormal returns are calculated as the stock’s twelve-month buy-and-hold return

beginning three months after the fiscal year-end minus the buy-and-hold value-weighted return

of the comparable size/book-to-market/prior year return portfolio over the same twelve-month period Extending the return interval three months beyond the fiscal year end helps to ensure that information related to domestic versus foreign earnings is publicly available by the

beginning of the return interval

Table 1 reports descriptive statistics for the variables in this study The average annual abnormal return for the firms in the sample is 92% The average change in domestic earnings

is 46% while the average change in total foreign earnings is 31% (both scaled by average total assets) Total earnings has an average change of 77%.5 The rank correlation between the change in domestic earnings and the change in foreign earnings is 15

[insert table 1 about here]

5 Average domestic earnings is 5.13% and average foreign earnings 2.36%.

4 Empirical Analysis

4.1 Primary Results

Table 2 reports the results of the Mishkin (1983) test for total earnings alone (Panel A) and for domestic and foreign earnings (Panel B) Results are reported using (1) a pooled model and (2) estimated coefficients from the Fama-MacBeth (1973) procedure where the coefficient represents the mean of the annual regression coefficients and the standard error is based on the time-series variation of the annual coefficients Assuming independence through time, the Fama-MacBeth procedure has the advantage of controlling for cross-sectional correlation in the residuals In the presence of positive cross-correlation in the residuals, the standard errors can bebiased downward and the t-statistics biased upward in the pooled regression For this study, the results of these two procedures are similar

[insert table 2 about here]

In Panel A, the estimate of actual total earnings persistence (T) is -.209 for the pooled model The magnitude of the total earnings coefficient suggests that total earnings contain both transitory and permanent components Positive (negative) changes in earnings tend to be

followed by negative (positive) changes in earnings, but the subsequent reversal is less than the

current change The estimate of the market’s perceived total earnings persistence (*) is -.194

The difference in actual total earnings persistence and the market’s perception of total earnings persistence is not significant (2 (1) = 162 with p-value = 687) The difference in the cross-

sectional means is also not significant (t = 073 with p-value = 943) The market appears to understand the time-series behavior of total earnings and incorporates this information into security prices appropriately

In Panel B of Table 2, the total change in earnings in year t+1 is regressed on changes in both domestic and foreign earnings in year t The coefficient on the change in domestic earnings(D) is -.233 for the pooled model, similar to the findings in Panel A for total earnings The coefficient on the change in foreign earnings (F) is -.096 for the pooled model Foreign

earnings show less reversal (i.e., are more persistent) than domestic earnings D and F are statistically different (2(1) = 9.17 with p-value = 003).6 The estimates of the market’s

perception of the persistence of domestic earnings (*) and foreign earnings (*) changes in the pooled model are -.168 and -.338, respectively The difference in the domestic earnings

coefficients (D - *= -.065) is not significant for the pooled model (2(1) = 1.71 with p-value

= 191) or for the means of the cross-sectional models (t = 859 with p-value = 401)

The difference in the foreign earnings coefficients (F - *= 243) is significant for the pooled model ((2(1) = 5.622 with p-value = 0177) and for the means of the cross-sectional models (t = 2.09 with p-value = 049) These results suggest that stock prices do not reflect accurately the time-series properties of foreign earnings Specifically, stock prices underestimatethe extent to which changes in foreign earnings persist A market that fixates on changes in total earnings without considering whether the changes are attributable to domestic versus foreign earnings causes stock prices to lag earnings

Finding that the market correctly prices the domestic component of earnings may seem inconsistent with the findings of extant research that the market fails to understand and price correctly the time-series trend of total earnings [e.g., Bernard and Thomas (1989, 1990) for the post-earnings-announcement drift] The tests in this study are conducted based on annual

earnings data The post-earnings-announcement drift is more pronounced for quarterly earnings

6 This result is consistent with that found by Bodnar and Weintrop (1997) who show that foreign operations provide greater growth opportunities than domestic operations Greater growth opportunities should lead to more persistent earnings.

data (Ball et al 1993) Furthermore, the firms in this study are larger than average and the earnings-announcement drift is less pronounced for larger firms (Bernard and Thomas 1989) These two factors contribute to the insignificant relation between domestic earnings and future abnormal returns documented in this study.

post-4.2 Distinguishing Between Market Mispricing and Failure to Control for Risk

Given the results of the previous section, a positive relation should exist between the change in foreign earnings in the current year and stock returns in the subsequent year

(controlling for changes in total earnings).7 This would be consistent with the evidence that the market does not fully understand (i.e., impound into security prices) the implications of foreign earnings on the value of the firm

However, such a result is also consistent with the change in foreign earnings being a proxy for (or source of) risk In rational markets, investors require higher returns on assets with greater risks Since foreign operations likely have greater risk than domestic operations, the higher future returns may simply be compensation for risk Even though commonly identified risk factors were controlled for in the measure of abnormal returns, there may be other

unidentified risk factors This is the classical problem for all studies in this area Are abnormal returns the result of market mispricing or the researcher’s inability to accurately specify and control for differences in risk?

Bernard et al (1997) discuss how researchers may best disentangle the issues of market mispricing from premia for unidentified risks They suggest two simultaneous tests With respect to earnings information, they suggest that market mispricing is more likely when (1)

7 Evidence for the positive relation between the change in foreign earnings in year t and abnormal returns in year t+1

is shown in Table 4 These results will be discussed in more detail below.

abnormal returns on zero-investment hedge portfolios are consistently positive each period and (2) these abnormal returns are concentrated around subsequent earnings announcement dates.

The first test involves taking a long (short) position in stocks expecting to do well

(poorly) If equal amounts are invested in each portfolio and the market prices securities

correctly, then any nonzero average return must be attributable to differences in risk However, a

risky zero-investment portfolio should result in positive returns in some periods and negative

returns in others, with the average return being perhaps positive If the annual returns are

consistently positive, it is difficult to attribute the results to unidentified risk

The second test suggested by Bernard et al (1997) determines whether the abnormal returns cluster around future earnings announcement dates If abnormal returns are the result of the market failing to understand fully the time-series properties of earnings (or its components), then absent other information events, earnings announcements are the points at which the marketrealizes that earnings are different than expectations and the price corrects Bernard et al (1997)demonstrate that for the post-earnings-announcement drift anomaly about 25% of the abnormal returns occur around earnings announcements They suggest that such a result is difficult to explain as risk-based Instead, it is more reasonable to conclude that the market failed to

understand the time-series properties of seasonal changes in quarterly earnings When

subsequent earnings are announced, the market revises its prior (incorrect) belief in a predictable manner Bernard et al (1997) show that risk-based anomalies (e.g., book-to-market, earnings-to-price) produce positive abnormal returns evenly over the return interval with no disproportionate cluster of returns around earnings announcements

In this section, the two tests suggested by Bernard et al (1997) are conducted in the context of domestic and foreign earnings A zero-investment hedge portfolio is created using

firms’ foreign earnings in year t The abnormal returns to the long and short positions are then measured over the twelve-month period beginning three months after year t So that the

variables used to create the hedge portfolios coincide chronologically and are publicly available

at the time of portfolio creation, only firms with December year-ends are included in the test The positions of the hedge portfolio are held for one year and then a new hedge portfolio is created To insure that the length of the earnings announcement period relative to the

nonannouncement period remains constant across firms, firms are required to have

announcement dates for all four quarters in the subsequent year provided on the Compustat quarterly tapes and these dates must fall within the twelve-month holding period These criteria reduce the sample to 3,614 firm/year observations.8

To create the long and short positions of the hedge portfolio, firms are first ranked on the magnitude of the change in total earnings and assigned in equal numbers to twenty portfolios each year Next, within each of the twenty portfolios, firms are sorted evenly into quintiles based on the change in foreign earnings Firms in the lowest quintile are firms that have

experienced the most negative change in foreign earnings for a given change in total earnings Firms in the highest quintile are firms that have experienced the most positive change in foreign earnings for that same change in total earnings Creating the portfolios in this manner allows firms to be ranked on changes in foreign earnings while controlling for changes in total earnings.The long (short) position consists of all firms in the top (bottom) quintile of the change in foreignearnings and has an average of 67.5 (63.3) firms per year The ranking procedure is performed

8 While requiring firms to have all four quarterly earnings announcement dates available within the one year return interval enhances interpretation of the results, it reduces the sample size The requirement, however, has little affect

on the results The average abnormal return to the hedge portfolio after relaxing this requirement is similar to that reported In addition, the average abnormal return of the firms deleted in this analysis is close to zero (.3%).

each year and the number of observations in the long and short position in each year is

approximately the same, except for sample size differences over time.9

If the market fixates on changes in total earnings, then those firms experiencing the largest increase in foreign earnings are more likely to have undervalued stocks In the

subsequent year when total earnings are higher than expected because of the higher persistence

of foreign earnings, stock prices will correct upward Likewise, firms that have experienced the largest decrease in foreign earnings are more likely to have overvalued stocks These firms have experienced a shift in earnings from more persistent foreign sources to less persistent domestic sources In the subsequent year when the market realizes that earnings expectations were too high, stock prices will correct downward If the market fails to understand the persistence of foreign earnings versus domestic earnings in this manner, then it should be possible to exploit this to earn positive abnormal returns in each year

The annual abnormal returns to the hedge portfolio are shown in Figure 1.10 The hedge portfolio produces a positive abnormal return in every year of the test period except one.11 The hedge portfolio abnormal returns for the single negative year is close to zero at -.36% The average abnormal return over the ten-year period is 6.8% If the results were due to omitted risk factors, the portfolio would return both negative and positive returns with perhaps the net return being positive over the entire period A consistent pattern of positive abnormal returns is more

9 The hedge portfolio was also created by first sorting firms evenly into deciles based on the change in total earnings each year Then, within each of the deciles, firms were sorted evenly into deciles based on their change in foreign earnings The long (short) position consisted of all firms in the highest (lowest) change in foreign earnings decile The results are similar to those reported.

10 To control for observations that have an extreme influence on the average portfolio returns, any observation that has an abnormal return greater than 200% is omitted from this analysis This results in seventeen observations

or 5% of the sample being deleted The average abnormal return of the hedge portfolio when these observations are not deleted is 7.1%, which is similar to the reported average abnormal return of 6.8% when these observations are deleted Extreme abnormal returns in the short position approximately offset those in the long position Also, deleting observations with abnormal returns greater than 200% had no qualitative effect on the regression results reported in Tables 2, 4, and 5.

11 The hedge portfolio total return (i.e., the return unadjusted for the comparable portfolio performance) is positive in all ten years.

characteristic of a market that fails to understand the implications of current changes in foreign earnings for future earnings The market corrects for its mispricing in the next year when subsequent earnings are reported different than expected.12

[insert figure 1 about here]

The next test involves determining whether the market corrections to the hedge portfolio are centered on future earnings announcements If the market fails to understand differences in the persistence of domestic and foreign earnings and corrects any mispricing when future

quarterly earnings are announced, then the correcting returns should be centered on future earnings announcements Without other information, future earnings announcements are the points at which the market realizes that it has under or overestimated the extent to which last year’s current earnings would persist into the future To perform this analysis, abnormal returns are separately measured in the announcement and nonannouncement periods

The announcement period is defined as the four three-day intervals around the quarterly earnings announcement dates in the subsequent year Each three-day interval begins two days before and ends on the day of the earnings announcement The announcement period abnormal return for a given year is then defined as the twelve-day compounded return surrounding the fourquarterly earnings announcements minus the compounded value-weighted return of the

comparable size/book-to-market/prior year return portfolio over the same twelve-day interval

The nonannouncement period includes all trading days over the twelve-month period, excluding the twelve announcement period days The nonannouncement period abnormal

12 An additional test to determine whether the hedge portfolio's positive abnormal returns were the result of already commonly identified risk premia was conducted by regressing the hedge portfolio total returns (i.e., the raw return

of the long positive minus the raw return of the short position) on the Fama-French (1993) three-factor model and the Fama-French-Carhart (1997) four-factor model For both models, the book-to-market factor was related to the hedge portfolio total return at the 06 significance level and the other risk factors were never significantly related The intercept for both models is positive and significant at the 01 level, indicating that positive abnormal returns remain after controlling for commonly identified risk factors These results can be obtained from the author upon request.

returns are measured as the compounded daily return over the nonannouncement period minus the compounded value-weighted return of the comparable size/book-to-market/prior year return portfolio over the corresponding days The average number of trading days in the

nonannouncement period is 240.7

Table 3 reports the average abnormal returns of the long and short positions for the announcement period, nonannouncement period, and the total period For the long position of the hedge portfolio, the average abnormal return for the announcement period is 1.21% The average abnormal return for the total period for the long position is 3.21% Approximately 38%

of the abnormal returns to this portfolio are concentrated around subsequent quarterly earnings announcement dates The twelve-day announcement interval represents approximately 5% of theannual return interval The average positive abnormal return combined with an unusually large proportion of the return being centered on future earnings announcements provides evidence consistent with market mispricing

[insert table 3 about here]

To provide a benchmark as to how unusual the pattern of abnormal returns exhibited by the long position of the hedge portfolio is, I use a random portfolio design That is, portfolios arecreated using a random sample of firms with characteristics similar to those of the firms in the long position All firms on CRSP that have the necessary data to calculate abnormal returns and

on Compustat that have all four quarterly earnings announcements between April 1 of the currentyear and March 30 of the following year are included The announcement period abnormal return is measured over the twelve-day interval, consisting of the four three-day periods around quarterly earnings announcements The annual abnormal return is measured from April 1 of the current year to March 31 of the following year The average annual and announcement period

abnormal returns for each random portfolio are then computed Such an analysis provides an expectation of how often the pattern of abnormal returns exhibited by the long position randomlyoccurs.

There are 13,544 firm/year observations over the ten-year test period meeting the

necessary data requirements, excluding firm/years in the long position So that the average size

of the firms used to construct the random portfolios approximately equals the average size of the firms in the long position, firms in the two smallest NYSE capitalization deciles are eliminated The mean (median) capitalization decile of firms in the random portfolios is 6.6 (7), which approximately equals the mean (median) capitalization decile of 6.3 (7) for the firms in the long position The final sample used to create random portfolios consists of 8,824 firm/year

observations Thirteen random portfolios are created each year for the ten-year hedge portfolio period This allows the average number of firms in each random portfolio (67.9) approximately

to equal the average number of firms in the long position (67.5) Thus, the data requirements, average firm size, average number of firms, length of the annual and announcement period returnintervals, and the test period of the random portfolios are similar to those of the long position The randomization procedure is repeated until 1,000 random portfolios are created

The results reveal that the pattern of abnormal returns for the long position is indeed unusual In only 79 of the 1,000 random portfolios does the average twelve-day announcement period abnormal return exceed the twelve-day announcement period abnormal return of the long position Furthermore, in only three of the 1,000 random portfolios does the average annual abnormal return exceed 3.21% (i.e., the annual abnormal return of the long position) with at least38% of the return occurring on earnings announcement dates Thus, the significant, positive abnormal return of the long position along with a disproportionately large fraction of this