The contribution of fundamental analysis after a currency devaluation.
ABSTRACT: For a sample of companies traded on the Mexican Bolsa, fundamental analysis is used to investigate the value of financial statement information to investors after the December 1994 currency devaluation. Associations with contemporary returns show that earnings in the year of the devaluation
Keywords: currency devaluation; financial statement analysis; Mexico; economic shock.
Data Availability: All data are available from public databases identified in the paper.
I. INTRODUCTION
Investors' need for forward-looking accounting information is greatly increased when an economic shock occurs during a reporting period. This is particularly true when the shock occurs late in the reporting period, so that current earnings information cannot be extrapolated to the future. Under these conditions, earnings lose value relevance and investors are forced to rely on other information. The basic idea underlying this paper is that, when the value-relevance of earnings is reduced by severe economic change, the detailed performance information provided in financial statements may still be useful in predicting future earnings and cash flows. Currency devaluations provide a natural experimental setting to investigate the effects of severe economic change. The findings from this setting likely extend to other instances in which firms encounter an economic shock during a reporting period (e.g., a terrorist attack, an oil embargo, or a labor strike).
The specific setting we investigate is the December 1994 peso devaluation in Mexico. The economic shock from the peso devaluation was typical of other currency devaluations. Inflation increased dramatically from 7 percent in calendar 1994 to 52 percent in 1995, and stock prices dropped by about 50 percent (see Figure 1, Panel A). A unique aspect of the 1994 Mexican devaluation is that most of the exchange rate decline occurred during the final week of December. The timing of the peso devaluation produces an ideal research setting. First, 1994 earnings clearly cannot be extrapolated to the future, so investors need alternative information. Second, we can use annual report data to test whether pre-shock accounting information can provide forward-looking information about post-shock operating performance and stock returns.
The primary finding to-date from research on currency devaluations is that earnings lose their value relevance (Graham et al. 2000; Ho et al. 2001; Davis-Friday and Gordon 2002). This study differs from prior research by using the fundamental analysis approach developed by Lev and Thiagarajan (1993) (hereafter LT) as an alternative way to investigate the value relevance of accounting information after a devaluation. Fundamental analysis uses the detailed information provided in financial statement line items to provide a finer analysis of a firm's performance than can be obtained using a summary measure, such as earnings. By considering changes in key components of operating performance (e.g., changes in gross margin or changes in expenses relative to sales), fundamental analysis has the potential to capture more completely than earnings the value relevance of accounting information after a macroeconomic shock.
The sample consists of companies that traded on the Mexican Bolsa during the period from 1992 to 1998. The results begin with 1993, however, since we use 1992 data to calculate changes. We find that annual earnings levels have considerable value relevance during the non-devaluation years and including the fundamental signals adds additional explanatory power. In contrast, earnings are not value relevant in the 1994 devaluation year. Importantly, a model that includes both the fundamental signals and earnings has considerable value relevance in 1994 ([R.sup.2] of 25 percent), primarily due to the selling and administrative (S&A) expense signal. Additional tests using interim accounting disclosures show that earnings remain value relevant during the first three quarters of 1994. In the fourth (devaluation) quarter, earnings lose value relevance, but fundamental analysis, primarily the selling and administrative expense signal, results in significant value relevance ([R.sup.2] of 24.7 percent). The tests using interim information indicate the increased contribution of fundamental analysis is due to the devaluation event, rather than a gradual decline in economic conditions throughout 1994.
To provide further information about whether the detailed accounting data used in fundamental analysis provide forward-looking information in 1994, we conduct several additional analyses. These analyses do not rely on an assumption that contemporaneous market reactions capture the full extent to which accounting information may be of value to investors. Using the approach in Abarbanell and Bushee (1997), we find that fundamental signals based on 1994 financial statements predict one-year-ahead earnings changes, analysts' forecast revisions, and analysts' forecast errors. This finding indicates that the signals are useful in predicting future earnings and that analysts use the signals in revising forecasts, but their revision does not fully reflect all the information about future earnings. In particular, two fundamental signals are consistently significant in the analyses: selling and administrative expense, and gross margin. Companies that are able to control the level of selling and administrative expenses relative to sales provide higher earnings after the devaluation, so Mexican companies must pay particular attention to controlling overhead costs. The rationale is that such costs are sticky, tending to increase more with an economic upturn than the decrease with a downturn (Anderson et al. 2002). In addition, firms that are able to maintain or increase their gross margin provide significantly higher future earnings. These firms are better able to pass on to their customers the higher costs from the increase in inflation. Tests using quarterly accounting information and forecast revisions immediately before and after the devaluation indicate that the increased contribution of the fundamental signals is due to the devaluation event, rather than to a gradual decline in economic conditions throughout 1994.
Because the analysis of forecasts provides evidence that analysts underutilize the information contained in fundamental signals, we then examine if a profitable trading strategy can be developed. Using the approach in Abarbanell and Bushee (1998), we construct a hedge portfolio in 1994 and document abnormal returns from a trading strategy based on the selling and administrative expense, and gross margin fundamental signals. This trading strategy does not produce significant abnormal returns in years other than 1994. We conclude from these various tests that fundamental analysis assumes additional importance in the wake of a macroeconomic shock but the market does not fully impound this fact.
The findings from this study are of potential interest to several groups. Accounting academics are interested in the general topic of how accounting information is priced by the market. The finding that the incremental value relevance of detailed financial statement information (about changes in assets, liabilities, revenues, and expenses) over that provided by earnings is increased by an economic shock is a new insight into how accounting information is priced. Investors should value knowing the economic conditions under which fundamental analysis provides information that is incremental to earnings. Investors, especially hedge fund managers, should be interested in the finding that the market does not fully impound the information in specific signals, since they may be able to develop a profitable trading strategy by going long and short on stocks, based on fundamental signals. Finally, accounting standard-setters, such as the FASB, International Accounting Standards Board, and the U.K.'s Accounting Standards Board, may be interested in our results for their project on reporting performance, which is considering the predictive value of various financial statement displays (see http://www.FASB.org for a project summary and related links).
The remainder of the paper consists of six sections. The second section describes the source of the data and the accounting system in Mexico. The third section discusses the research methodology. The fourth section provides a summary of the fundamental signals and their hypothesized effects in Mexico. The fifth section presents our empirical results. The final section provides a summary and conclusion.
II. DESCRIPTION OF DATA SOURCE AND MEXICAN ACCOUNTING SYSTEM
The financial statement and stock price data used in this study were obtained from the Economatica database, which provides accounting data and stock returns for Latin American companies. Economatica reports accounting information based on the standardized format used in filing with the Mexican Bolsa. This format provides the intermediate components of financial statements that are needed for fundamental analysis. The sample consists of all companies that traded on the Mexican Stock Exchange during the period 1992-1998 (the 1992 year is used to calculate changes). The sample period begins with 1992 because this is the first year for which the Economatica database provides an extensive sample. (1)
In accord with Bulletin B-10 (Instituto Mexicano de Contadores Publicos [IMCP] 1984), Recognition of the Effects of Inflation in Financial Information, Mexican companies have reported replacement-cost, price-level-adjusted (RCPLA) information in the primary financial statements since 1984 (under an evolving set of rules). Under RCPLA, monetary assets and liabilities for the latest year do not require restatement because they are already stated in terms of year-end purchasing power. However, companies do report purchasing power gains and losses on monetary assets and liabilities held during the year as a component of earnings. Inventory and property, plant, and equipment are revalued at year-end and the holding gain/loss is reported in stockholders' equity. Through 1996, inventory was reported at estimated cost to manufacture or repurchase at year-end prices, and appraisals were used to value property, plant, and equipment at replacement cost. Beginning January 1, 1997, Mexican companies have been required under the Fifth Amendment to Bulletin B-10 to use the consumer price index to revalue all nonmonetary items, including inventory and property, plant, and equipment. Valuations during our sample period, therefore, reflect a mix of replacement cost and general-price-level-adjusted amounts. (For expository convenience, we use the acronym RCPLA for all periods.)
Under RCPLA accounting, amounts for every time period are reported on the financial statements in purchasing power at the year-end of the most recent reporting period. Assets and liabilities for the current year are already valued in terms of year-end purchasing power (as discussed above). Operating results during the current year and financial statements from prior years need to be restated. The Mexican Consumer Price Index is used as the measure of purchasing power. In accord with the Third Amendment of Bulletin B-10, which became effective in 1990, the restatement procedure works as follows: If inflation during the current year has been 30 percent, then every amount from the prior year's financial statement is multiplied by 1.3 to obtain the balances used for comparative purposes in the current year's statement. Revenues and expenses that occur during the current year are rolled forward into year-end prices using monthly price indexes. If inflation occurs evenly throughout the year, then the monthly adjustments will often approximate the results that would be obtained by multiplying revenues and expenses by an average index for the year. For 30 percent inflation, the average index would be 1.15. The primary purpose of the restatement is to facilitate the analysis of year-to-year changes in operating results. (2) Since fundamental signals are calculated using changes in income statement amounts, assets, and liabilities as reported in the annual report (e.g., change in selling and administrative expenses, change in accounts receivable), the signals used in this study are calculated in constant, consumer purchasing power. This paper is the first to study fundamental signals based on RCPLA accounting data.
III. RESEARCH METHODOLOGY
Association of Earnings and Fundamental Signals with Contemporaneous Returns
The value relevance of earnings is assessed by estimating the following equation:
(1) [R.sub.it] = [[alpha].sub.0] + [[beta].sub.1][E.sub.it] + [v.sub.it]
where R represents a market-adjusted, 12-month return, ending three months following the fiscal year-end (which is December for all firms on the Bolsa); E is earnings-per-share deflated by price at the beginning of the return accumulation period; i and t are firm and time subscripts, respectively; and v is an error term assumed iid N(0, [[sigma].sup.2]). (3) Abnormal returns were calculated as the difference between the firm's raw return and a market-adjusted amount using the IPC (Prices and Quotations Index) from the Bolsa. Model (1) is estimated for the 1994 devaluation year and again for the five non-devaluation years.
As discussed further in the next section of the paper, we use the fundamental signals originally developed by LT (with the addition of a leverage signal). The fundamental signals model is expressed as follows:
(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Similar to LT, PTE is defined as pre-tax earnings multiplied by 1 minus the tax rate in year t-1 and deflated by price at the beginning of the return accumulation period. [X.sub.j] represents the jth fundamental signal that is included in the model. Model (2) is also estimated separately for the 1994 devaluation year and over the five non-devaluation years.
We investigate the value relevance of earnings and fundamental signals in three steps. First, we use Model (1) to determine whether earnings are value relevant in the 1994 devaluation-reporting period and in the non-devaluation-reporting periods. Second, we use Model (2) to determine whether the fundamental signals provide value relevance that is incremental to earnings in the devaluation and non-devaluation periods. Third, we examine whether the coefficients for earnings and the fundamental signals differ between the devaluation and non-devaluation periods.
Any increase in value relevance for the fundamental signals when comparing 1994 to the non-devaluation annual periods could be due to the December devaluation event or economic conditions during the year that triggered the devaluation. To investigate this issue, we use interim accounting data to fit Models (1) and (2) for two subperiods of 1994. The first subperiod uses deflated earnings for the first three calendar quarters of the year regressed on market-adjusted returns beginning April 1, 1994 (so that calendar 1993 earnings have been released) and extending through November 30, 1994 (so that third quarter 1994 earnings have been released). (4) The second subperiod uses fourth quarter earnings regressed on returns surrounding the devaluation event, covering December 1, 1994 through March 31, 1995 (see Figure 1, Panel B). We then investigate whether the results found in the comparison of Models (1) and (2) for 1994 is due to the fourth (devaluation) quarter, the first three quarters, or both.
Association of Fundamental Signals with Future Earnings Changes, Analysts' Forecast Revisions, and Analysts' Forecast Errors
The notion underlying fundamental analysis is that it provides insight concerning the quality of earnings and, therefore, provides information about the probability that current earnings will persist into the future. Although an association between fundamental signals and contemporaneous returns indicates that the fundamental signals are capturing information used by market participants, it does not necessarily mean that the signals themselves are useful in assessing earnings quality. To provide this information, we apply the methodology used by Abarbanell and Bushee (1997) (hereafter AB). First, we examine if an association exists between the fundamental signals and future earnings changes. A significant association indicates that the signals provide insight into earnings persistence. Second, we examine the relation between the fundamental signals and revisions in analysts' earnings forecasts to determine whether analysts use the information provided by the signals in revising their forecasts. Third, we examine the relation between the fundamental signals and analysts' earnings forecast errors to determine whether analysts completely use the signals in revising their forecasts.
We examine the usefulness of the fundamental signals in Mexico in predicting future earnings changes, analysts' forecast revisions, and analysts' forecast errors using the following equation:
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where Y represents the change in one-year-ahead earnings (denoted [DELTA][E.sub.it+1]), the revised analyst forecast for year t+1 after year t's earnings are announced (denoted RE[V.sub.it]), or the analyst forecast error for year t+1 (denoted AF[E.sub.t+1]). [DELTA][E.sub.it+1] is defined as earnings-per-share in year t+1 minus earnings-per-share in year t deflated by price at the beginning of the returns accumulation period. Similar to Abarbanell and Bushee (1997), RE[V.sub.it] is defined as follows:
(4) RE[V.sub.it] = [(Post[F.sub.t+1] - [E.sub.t]) - (Pre[F.sub.t+1] - [F.sub.t])]/[P.sub.t-1]
where Post[F.sub.t+1] is the first available mean consensus forecast for earnings in year t+1 issued subsequent to the year t earnings announcement, [E.sub.t] is realized earnings in year t, Pre[F.sub.t+1] is the first available mean consensus forecast in year t for earnings in year t+1, and [F.sub.t] is the first available mean consensus forecast in year t for year t earnings. (5) RE[V.sub.it] can be thought of as the forecast revision (Post[F.sub.t+1] - Pre[F.sub.t+1]) after removing the effect of the current period's forecast error on the revised forecast of future earnings (i.e., removing [E.sub.t] - [F.sub.1]). Finally, AF[E.sub.it+1] is defined as realized earnings-per-share in year t+1 minus the first forecasted earnings-per-share for year t+1 issued subsequent to year t's earnings announcement, deflated by price at the beginning of the returns accumulation period.
A significant coefficient on a fundamental signal indicates the signal is useful in predicting the dependent variable (either realized earnings changes, analysts' forecast revisions, or analysts' forecast errors). Comparing the association of a fundamental signal across the three equations can provide insight as to whether analysts fully use the information about future earnings contained in the signal. For example, if a fundamental signal is associated with future earnings changes and analysts' forecast errors but not with analysts' forecast revisions, then this finding would suggest that analysts do not use the information contained in the signals. Underutilization suggests that investors could take advantage of analysts' inefficiencies to earn abnormal profits.
IV. FUNDAMENTAL SIGNALS
We relied primarily on the fundamental signals chosen by LT, who based their choice on an examination of the types of accounting information used by financial analysts in the U.S. Before using the LT fundamental signals, however, we examined sell-side analyst reports for seven large Mexican firms. We found that the report format and types of analyses are very similar to that for U.S. companies. Sufficient information is available to calculate five fundamental signals that were statistically significant in the LT model. Although several signals cannot be used, the tests in this study include most of the fundamental signals that were found to be significant in the LT empirical analyses. (6) In a more recent study, Amir et al. (1999) also use only these five signals.
The five signals were designed by LT so that a negative sign reflects good news (and, therefore, positive returns) under the most common interpretation by analysts for U.S. firms. Under the economic conditions faced by firms in Mexico, the expected sign can differ from the U.S. Each of the five signals is discussed briefly in the following paragraphs, with comments on possible differences in applying the signals to Mexican companies. We also discuss a borrowing cost signal, which could be important to Mexican companies because they face volatile interest rates. (7)
1. Inventory: An increase in inventory relative to sales is generally interpreted by financial analysts as a negative signal for two reasons. First, such an occurrence indicates a greater chance that inventory will become obsolete. Second, holding costs are an increasing function of the amount of inventory on hand. On the other hand, a higher level of inventory might be viewed as a positive signal, since it reduces the chances of experiencing an inventory shortage and could signal that managers expect an increase in future sales. In an inflationary economy, an increase in inventory may also be a positive sign if the additional inventory has been obtained or produced in anticipation of cost increases. Further, inventory purchases, rather than cost of goods sold, are deducted for taxes in Mexico. As a result, while a negative sign is expected in the U.S., an increase in inventory could be either a positive or negative signal in Mexico. The signal is computed as follows:
Inv = [[Inventory.sub.t] - [Inventory.sub.t-1]]/[Inventory.sub.t-1] - [[Sales.sub.t] - [Sales.sub.t-1]]/[Sales.sub.t-1]
2. Accounts Receivable: Disproportionate increases in accounts receivable relative to sales are considered an unfavorable signal since this could indicate collections are later in relation to the firm's credit policy. This practice is more likely to occur in inflationary periods because customers benefit more by delaying payments. Higher receivables could also indicate that the company has relaxed credit terms in order to sustain a given level of sales. Last, a significant increase in accounts receivable might indicate the company has managed earnings by "stuffing the channel" by shipping goods near the end of a reporting period. In each case, consistent with the U.S., a negative association is expected for Mexico between this fundamental signal and returns. The accounts receivable signal is computed as:
AR = [[Acct.Rec..sub.t] - [Acct.Rec..sub.t-1]]/[Acct.Rec..sub.t-1] - [[Sales.sub.t] - [Sales.sub.t-1]]/[Sales.sub.t-1]
3. Gross Margin: A disproportionate (to sales) decrease in the gross margin amount (sales minus cost of goods sold) is viewed as a negative signal by analysts about the long-term performance of the firm. A decrease in the gross margin signal could be caused by a decrease in sales price or an increase in input costs. A decrease in sales price could reflect increased competition or customer resistance to prices. An increase in input costs could be due to fixed or variable costs. Cost increases are more likely when inflation is high, a more frequent condition in Mexico than in the U.S. In each case, the permanence of future earnings is likely to be impaired by a decline in gross margin that is not offset by an increase in sales, so we expect a negative sign. The signal is defined as follows:
GM = [[Sales.sub.t] - [Sales.sub.t-1]]/[Sales.sub.t-1] - [Gross [Margin.sub.t] - Gross [Margin.sub.t-1]]/Gross [Margin.sub.t-1]
4. Selling and Administrative Expenses: An increase in administrative expenses relative to sales can indicate a loss of control over fixed expenses that cannot be passed on to customers and will adversely affect future cash flows. In addition, an increase in selling expenses relative to sales can indicate that an increased effort to produce sales was not entirely effective. Recent research has shown administrative expenses to be "sticky," meaning they tend to increase more with economic upturns than the decrease with a downturn (Anderson et al. 2002). While LT found this signal to be important in the U.S., we expect it could be even more important in Mexico because the level of sales can change substantially from period to period. The S&A fundamental signal, which is expected to have a negative sign, is defined as:
S&A = [S&[A.sub.t] - S&[A.sub.t-1]]/S&[A.sub.t-1] - [[Sales.sub.t] - [Sales.sub.t-1]]/[Sales.sub.t-1]
where S&A represents selling and administrative expenses.
5. Effective Tax Rate: A change in the effective tax rate that is due to reasons other than a statutory tax rate change is considered by analysts to be transitory in nature. The effective-tax-rate fundamental signal is calculated by splitting the change in earnings into two components. The first component is the current year's after-tax earnings that would have occurred if the prior year's tax rate had been in effect. The second component, the tax signal component, is the effect of the tax rate change on the current level of earnings. Using mathematical notation, the expression is:
[E.sub.t] = PTE(1 - [T.sub.t-1]) + PTE([T.sub.t-1] - [T.sub.t]).
LT predict and find a negative relation between this fundamental signal and returns. The expected sign is more complicated for Mexican companies since a change in effective tax rate is more likely to have a permanent effect on earnings. Mexican businesses pay the greater of an income tax or an asset tax, resulting in taxes due for most loss years. If the tax rate increases because a firm paid an asset-based tax in the prior year, but now owes an income tax in the current year, then the signal could capture a permanent effect on earnings. (8) As a result, the tax signal could be positive or negative for Mexican companies.
6. Borrowing Costs: Leverage is a widely used ratio in fundamental analysis (Penman 2001), although LT did not include it among their set of fundamental signals. A firm that earns a higher (lower) rate of return on invested capital than the rate of interest paid on debt provides a higher (lower) return to stockholders. However, borrowing costs are highly volatile in Mexico, regardless of whether a firm borrows domestically or from outside the country. If a company has debt denominated in pesos, then domestic interest rates can increase dramatically to reflect higher anticipated inflation. If a Mexican firm has debt denominated in the U.S. dollar or another foreign currency, then the amount of pesos required for interest and principal payments increase immediately after a currency devaluation. Because borrowing costs are so volatile in Mexico, the fundamental signal is expected to have a negative sign. It is defined as:
Lev = Total Liabilities/Total Assets
V. RESULTS
Descriptive Statistics and Benchmark Model
The Economatica database provides a total of 444 firm-year observations with the requisite financial statement and returns data, with the number of observations increasing over time. Given the volatile economic environment of Mexico during our sample period, it is not surprising that a preliminary examination of the data revealed some extreme fluctuations in returns, earnings, and other variables employed in our regression models. To control for influential observations, we winsorized all regression variables at the 5 and 95 percent tails of their respective distributions.
Table 1, Panel A, presents descriptive statistics for each year from 1993 to 1998, with amounts that are statistically significant at the .05 level presented in boldface. Since the focus of the paper is on the effects of the 1994 devaluation, the following discussion considers how 1994 differs from the other years. For the 1994 devaluation year, the means and medians for pretax earnings deflated by price (PTE) are not significantly different from zero. (9) In contrast, PTE is significantly positive in the other years with the exception of the mean for 1998, a year in which the peso exchange rate declined due to uncertainty generated by the Asian currency crisis (see Figure 1, Panel B). With regard to the fundamental signals, the inventory (Inv) and accounts receivable (AR) signals in 1994 are both significantly positive and the magnitudes are unusually large, indicating that inventory and accounts receivable grew faster than sales (i.e., downbeat signals). In 1995, the signs change to significantly negative as Mexican companies responded to the crises by reducing the level of inventory and accounts receivable. A similar reduction continued in 1996, which was then followed by increases in 1997. Leverage (Lev), which is always significantly different from zero, rose in the 1994 devaluation year and remained high in 1995 before declining in the following years. The higher leverage likely results from an increase in the number of pesos required to settle debt denominated in the U.S. dollar (or another foreign currency).
The magnitude of the correlations in the matrix in Table 1, Panel B, suggests that multicollinearity among the fundamental signals will not pose a significant problem. The top line of Table 1, Panel B, provides a univariate test of associations with returns for each of the explanatory variables using data pooled over the 1993 to 1998 period. As expected, pretax earnings (PTE) has a significant, positive association with returns (R). Three of the fundamental signals, gross margin (GM), selling and administrative expense (S&A), and leverage (Lev), are significantly negatively correlated with returns (as hypothesized).
Table 1, Panel C presents descriptive evidence about the properties of analysts' forecasts in Mexico. Analysts overestimated earnings (FE) by a median percentage of a little over 4 percent in 1994 and 1998. Analysts may not have anticipated the decline in the peso exchange rate in 1994 and 1998 (see Figure 1, Panel B) and the resultant negative effect on reported earnings. In general, negative signs indicate an overestimation of earnings, and research has found this occurs in several countries (e.g., Brown et al. 1985; O'Brien 1988; Francis and Philbrick 1993; Capstaff et al. 1998). Analysts' consensus forecast revisions (REV) measure the forecasted change in earnings from year t to t+1 after controlling for the current period's earnings surprise. The significant positive amounts for REV in 1994 are influenced by large negative forecast errors (FE) in that year (see Equation (4)).
The dispersion in analysts' forecasts is measured as the standard deviation of all available forecasts that comprise [F.sub.t] divided by [F.sub.t]. Dispersion increases significantly in 1995 and 1996 and then gradually returns to more normal levels, indicating that the 1994 devaluation reduced the predictability of earnings in subsequent years. The mean number of analysts following a company, which also corresponds to the number of forecasts that comprise [F.sub.t], ranges from 8.5 in 1998 to 13.4 in 1997. The number of forecasts that changed either up or down from the previous period is shown in the last column of Panel C of Table 1. Revisions were not unusually frequent after the devaluation.
We also examined whether analysts' forecasts are updated on a timely basis. This information is of interest primarily because we later examine the effects of the fundamental signals on REV and want to know if the findings could be influenced by stale forecasts. In untabulated analyses, we found that, subsequent to the announcement of year t earnings, the first available forecast for year t+1 earnings (Post[F.sub.t-1]) during non-devaluation years is available by March (April) of year t+1 for 94.8 percent (97.1 percent) of firms. The corresponding amounts after 1994 earnings are released are slower at 39.1 percent (100 percent), which means that Post[F.sub.t+1] (the forecast for 1995) was issued one month later for about half the firms (i.e., 94.8 percent less 39.1 percent).
This delay by analysts in providing Post[F.sub.t+1] might be due to the following reasons: Mexican companies were slow in reporting their earnings for 1994, and/or Mexican analysts required additional time in formulating a forecast of 1995 earnings after the announcement of 1994 earnings. We investigate these explanations by accessing the annual earnings announcement date from I/B/E/S and examining the timeliness of earnings reports across the sample period. For each year, we calculate the number of days from fiscal year-end to the earnings announcement date (not tabulated). Excluding 1994, the median ranges from 61 to 68 days. However, for 1994 the median value is 93 days. Thus, the delay by analysts in providing Post[F.sub.t+1] in 1994 is due to Mexican companies taking additional time to announce their 1994 earnings. While forecast staleness is not a serious problem, this delay brings up the question of whether our returns period for 1994 should be extended beyond March. As a result, we reran all analyses for 1994 using a return period ending with April 1995. None of our analyses changed substantively, so we present results using the March return period in order to be consistent with other years.
Association of Earnings and Fundamental Signals with Contemporaneous Returns
Table 2, Panel A, presents information about the association of annual earnings with returns for Model (1), which serves as a benchmark in subsequent tests of the incremental value relevance of the fundamental signals. As expected, earnings lose value relevance in 1994 due to the peso devaluation in the last week of the year. In contrast, when observations from the five non-devaluation years are pooled, earnings have considerable value relevance with an [R.sup.2] of 21.7 percent and the earnings coefficient of .694 has a t-value of 10.5. This result is robust after controlling for cross-sectional dependence, since the five-year mean of 0.596 is significant with an intertemporal t-statistic of 6.16. To provide a direct comparison between the devaluation and non-devaluation periods, we formally test whether the earnings coefficients differ between 1994 and the non-devaluation years. Pooling data for all six years and including a dummy variable for the 1994 observations, the t-value of 2.52 allows rejection of equality at the .05 significance level.
To determine whether the loss of value relevance in 1994 is due to economic conditions existing throughout the year or to the economic shock of the December devaluation, Table 2, Panel B, reports separate results for the first three quarters of 1994 and for the devaluation quarter. Since earnings become insignificant only in the devaluation quarter, we conclude that the loss of value relevance for 1994 annual earnings is due primarily to the economic shock of the devaluation.
Table 3, Panel A, examines whether the fundamental signals provide incremental explanatory power in comparison to annual earnings. For the 1994 devaluation year, the model containing both the fundamental signals and earnings has considerable explanatory power with an adjusted [R.sup.2] of 25.0 percent. This is substantially more than the adjusted [R.sup.2] of -0.2 percent for the earnings-only benchmark model, and the partial F-statistic of 3.94 shows that the improvement in explanatory power from including the six fundamental signals is statistically significant. The S&A signal is highly significant in 1994 (t = -3.95) and provides most of the improvement.
In the non-devaluation years, the fundamental signals again provide incremental explanatory power in comparison to annual earnings. The adjusted [R.sup.2] of 26.2 percent for the full model exceeds the adjusted [R.sup.2] of 21.7 percent for the benchmark earnings-only model, and the partial F-statistic of 4.95 indicates the difference is significant. The GM and S&A signals are statistically significant with t-statistics of -2.37 and -5.22, respectively. The corresponding intertemporal t-statistics are -2.55 and -1.96, and both signals have coefficients with the hypothesized negative sign in all five years. (10) In addition, the tests for a difference in coefficients indicate that the PTE and S&A coefficients for the devaluation year are significantly different than for non-devaluation years.
To determine whether the increased importance of the fundamental signals in 1994 is due to economic conditions prior to the devaluation or to the shock of the devaluation, Table 3, Panel B, reports separate results for the first three quarters of 1994 and for the fourth (devaluation) quarter. Examining the fourth quarter results, the adjusted [R.sup.2] of 24.7 percent exceeds the adjusted [R.sup.2] of -1.9 percent for the earning-only benchmark model (and is similar to the adjusted [R.sup.2] of 25 percent for the annual period). This increase in explanatory power is significant based on the partial F-statistic of 3.62. Also consistent with the annual results, the contribution of the fundamental signals is driven primarily by the S&A signal, which is significant with a t-value of -1.99. In contrast, the partial F-statistic for the first three quarters of 1994 shows that the fundamental signals as a group do not have significant explanatory power--although the S&A signal is significant with a t-value of -1.67. Based on this result, we conclude that the increased importance of the fundamental signals in the annual analyses is due to the economic shock of the devaluation event, as opposed to economic conditions that triggered the devaluation.
Usefulness of Fundamental Signals in Predicting Future Earnings, Forecast Revisions, and Forecast Errors
Table 4 reports results from estimating Equation (3) with three different dependent variables: one-year-ahead earnings change ([DELTA][E.sub.t+1]), analysts' forecast revisions (RE[V.sub.t]), and analysts' forecast errors (AF[E.sub.t+1]). Examining the [DELTA][E.sub.t+1] equation for both the devaluation year and the pooled non-devaluation years, the coefficient on PTE is significantly negative in both instances, which indicates reversion toward the mean. When revising their earnings forecasts (RE[V.sub.t]), analysts anticipate mean reversion, as indicated by a significant negative coefficient on PTE in both the devaluation and non-devaluation time periods. The amount of the anticipation is sufficient so that PTE is not significant in explaining one-year-ahead forecast errors (AF[E.sub.t+1]). (11)
With respect to the fundamental signals in the devaluation year, we find that the 1994 GM and S&A signals have a significantly negative association with the change in earnings, and analysts' revisions are significantly influenced by both signals. However, the predictive information provided by the signals is underutilized because both signals are significant in explaining 1995 forecast errors (AF[E.sub.t+1]). This underutilization of information suggests that investors may be able to earn abnormal returns by trading on the information provided by these signals.
For the pooled non-devaluation years, we find that the GM and S&A signals again have a significant negative association with future earnings changes. Analysts do not use this information in their forecast revisions (whereas they used the information but not fully in the devaluation year). As a result, both signals are significantly associated with forecast errors. The underutilization of the information in those two signals indicates that investors may be able to earn abnormal returns even during non-devaluation years. Examining the other signals, the tax signal is also associated with future earnings changes, but analysts use this information in their forecast revisions, so it is not significant in explaining forecast errors. One other signal, leverage, is associated with forecast errors.
The bottom three rows of Table 4 test whether the coefficients differ significantly between devaluation and non-devaluation years on any of the variables found to be significant in explaining earnings changes, forecast revisions, or forecast errors. For [DELTA][E.sub.t+1], the tests show that PTE is significantly different with a greater extent of mean reversion in the devaluation year than in the non-devaluation years (t-value of -2.18). However, the effect of PTE on forecast revisions and forecast errors does not differ between the periods. Another variable with a significant difference is the S&A signal, which has a higher negative association with future earnings changes ([DELTA][E.sub.t+1]) in the devaluation year than in nondevaluation years (t-value of -2.49). Analysts recognize this relationship in revising their forecasts, since their use of the S&A signal is significantly greater in the devaluation year compared to other years (t-value of -2.65). As a consequence, the forecast error is not significantly different between the periods. None of the other fundamental signals have a statistically different effect between the devaluation year and the non-devaluation years in predicting earnings changes, forecast revisions, or forecast errors.
The forecast revisions (RE[V.sub.t]) reported in Table 4, Panel A, use information about operations throughout 1994. To determine whether the revisions are due to the December devaluation, we next consider analysts' forecast revisions during the two months immediately surrounding the devaluation. Table 4, Panel B, compares the last monthly consensus forecast issued prior to the devaluation (in December 1994) (12) to the last consensus forecast issued before actual earnings are released (in February 1995). (13) We then regress fundamental signals measured using third quarter 1994 financial statement amounts on this forecast revision to determine whether the fundamental signals are associated with the information used by analysts. Corresponding to the results for RE[V.sub.t] reported on Panel A, analysts place significant weight on two components of operating performance, namely, a company's past ability to earn high profit margins (GM) and maintain control over operating expenses (S&A). Finding similar results indicates that the devaluation event, rather than economic conditions earlier during 1994, are driving the significance of the GM and S&A signals in RE[V.sub.t]
Hedged Portfolio Trading Strategy
Our final analysis investigates the possibility that abnormal returns can be earned using a trading strategy that exploits information in the fundamental signals following the December 1994 devaluation. The trading strategy involves construction of hedged portfolios. The methodology has been previously used by Abarbanell and Bushee (1998) to investigate the profitability of trading on the LT signals in U.S. markets. This section provides only a brief description of the hedged portfolio methodology. For further details, the reader should refer to Abarbanell and Bushee (1998) and Fame and Macbeth (1973).
We add to our fundamental analysis model two risk proxies: firm size ([SIZE.sub.t]), measured as the natural log of the market value of equity at the end of year t, and the market-to-book ratio (M[B.sub.t]), calculated as the end-of-year market value of equity divided by the book value of equity. These control variables have been used in prior studies (e.g., Fama and French 1992; Fama 1998) to reduce the chance that any significant returns from the hedged portfolios are due to omitted risk factors. (14)
For every year, we rank each fundamental signal, earnings, and risk proxy into deciles (0 to 9) and divide each decile rank by 9 to create variables with a range from 0 to 1. Next, we change the sign of each variable so that its expected association with future returns is positive. (15) To determine the hedged portfolio return that would be realized by optimally weighting the portfolio based on the fundamental signals, we estimate the following equation:
(5) [R.sub.it+1] = [[alpha].sub.0] + [[beta].sub.1]PT[E.sub.it] + [[beta].sub.2][SIZE.sub.it] + [[beta].sub.3]M[B.sub.it] + [[beta].sub.4][Inv.sub.it] + [[beta].sub.5]A[R.sub.it] + [[beta].sub.6]G[M.sub.it] + [[beta].sub.7]S&[A.sub.it] + [[beta].sub.8][Tax.sub.it] + [[beta].sub.9][Lev.sub.it] + [[omega].sub.it],
where [R.sub.it+1] is a market-adjusted return where the accumulation period begins in a month after the information in the fundamental signals is disclosed. The coefficients from Equation (5) represent the hedged abnormal return from a zero-investment portfolio that is optimally formed to exploit the information contained in that signal. The sum of [[beta].sub.4] through [[beta].sub.9] represents the net portfolio return from exploiting the information contained in all the signals, while controlling for year t earnings, size, and the market-to-book ratio.
Table 5 presents the results from estimating Equation (5). To provide a benchmark, we also provide results from a trading strategy that is based on actual realized earnings in year t+1, which is entitled the "perfect foresight strategy." As expected, knowing future earnings levels results in a significantly positive abnormal return in most periods. The only exception is for 1993, which is consistent with the insignificant association between earnings and returns in 1994 (i.e., having perfect foresight of 1994 earnings in 1993 does not produce an abnormal return). (16)
The first set of results in Table 5 measure the abnormal returns during periods immediately following the devaluation. (17) We find that using a fundamental analysis strategy that exploits the information contained in the fundamental signals results in a significant abnormal return following the devaluation. Consistent with results in Table 4, the signals with significant coefficients in Equation (5) for this return window are the same that are significant in predicting future forecast errors, GM and S&A. Given that GM and S&A have been consistently the strongest signals throughout our analyses, the last columns present returns when the hedge portfolios are formed using only information in these two signals. The hedged returns are again highly significant in the extended windows following the devaluation. The last part of Table 5 investigates the success of the fundamental analysis strategy in the other years of our sample period. None of the fundamental analysis strategy returns are significant, and we do not find any individual fundamental signal coefficients to be significant. (18)
Although the method used to assess the hedged portfolio strategy in Table 5 is desirable in that it controls for risk factors while simultaneously considering the combined effect of all fundamental signals, we cannot determine how much of the abnormal returns are driven by the buy-side (long) portfolio versus the sell-side (short) portfolio. If the returns are driven by the sell-side portfolio, this would suggest that the strategy is less implementable since transaction costs would be more substantial. In Table 6, we present returns from an alternative hedged trading strategy. This strategy groups the sample firms by year into four portfolios based on individual fundamental signals that were derived from the 1994 annual report. The most (least) favorable change in the associated signal was labeled the long (short) portfolio. The mean and median returns to the portfolios assume that every security within each portfolio is equally weighted. Table 6 presents abnormal returns for the May 1, 1995, through April 30, 1996 return window for the GM and S&A signals. Consistent with the results in Table 5, statistical significance is found for each signal, and more importantly, we document significant buy-side returns for both signals.
To summarize, while the GM and S&A fundamental signals have a statistically significant association with contemporary returns, analysts underutilize the information in those signals, providing an opportunity for investors to earn abnormal returns after the 1994 devaluation. The market therefore does not fully impound the additional information provided by fundamental analysis after the economic shock.
Supplemental Robustness Tests
Adding a Signal for Negative Earnings
Research in the U.S. shows that the relationship between earnings and returns is attenuated when earnings are negative (Hayn 1995). For 1994, we find that 39.6 percent of our sample firms have negative earnings (E) and 52.8 percent have negative earnings using the prior year's tax rate (PTE). To investigate the impact on our results of controlling for negative earnings, we set a dummy variable equal to 1 when earnings are negative (0 otherwise) and include it in our benchmark regression, Model (1), along with its interaction with earnings. The dummy variable main effect is insignificant in every year. However, the interaction with earnings (E) is significantly negative in 1994 and 1998 at the .05 level using a one-tailed test, and E becomes significantly positive in 1994. (19) The interaction term is insignificant over the 1993 to 1998 period in the cross-temporal means test.
Because research indicates that the association of earnings with returns is reduced by negative earnings, we investigate whether including an additional fundamental variable to indicate the sign of earnings could further enhance our ability to explain contemporaneous returns. This possibility is investigated by including a dummy variable (negative earnings equal to 1) as both an intercept and an interaction term by multiplying it times PTE in Equation (2). The primary results of our analyses are qualitatively unchanged, and inclusion of the dummy interaction term does not improve upon the hedged portfolio returns.
Constant Sample over Time
Data availability in Economatica allows us to increase the number of firms in our sample over time. We examine how our results are affected by retaining only those companies that have the requisite data for the entire sample period. Although the results with this smaller sample of 36 firms are somewhat weaker across years, the overall findings are generally consistent with the results reported in the tables. We still find that GM and S&A are consistently associated with contemporaneous returns (especially in the devaluation year where the adjusted [R.sup.2] is -2.2 percent for the benchmark model versus 30.4 percent for the fundamental signals model) and analysts tend to underutilize the information provided by these signals. When only GM and S&A are included in the model, the hedged portfolio provides a significantly positive abnormal return for 1994. Using the full model, however, the hedged return is not significant.
VI. CONCLUSION
Investors' need for forward-looking accounting information is greatly increased when an economic shock occurs during a reporting period. This is particularly true when the shock occurs late in the reporting period, so that current earnings information cannot be extrapolated to the future. Under these conditions, earnings lose value relevance and investors are forced to rely on other information. Currency devaluations provide a rich, natural experimental setting in which to investigate the effects of severe economic change on the usefulness of accounting information. The primary insight from research on the effects of a currency devaluation on accounting information is that earnings lose value relevance (Graham et al. 2000; Ho et al. 2001). This paper differs from prior research by investigating whether the detailed performance information provided in financial statements retains value relevance after a currency devaluation. This is accomplished using fundamental analysis, relying primarily on the signals developed by Lev and Thiagarajan (1993) (LT). More specifically, we test whether various fundamental signals (e.g., changes in gross margin or in selling and administrative expenses) provide forward-looking information to investors after the 1994 peso devaluation in Mexico.
For a sample of companies traded on the Mexican Bolsa, we find that earnings are not value relevant immediately following the 1994 currency devaluation. In contrast, the fundamental signals, especially the selling and administrative expense signal, provide value relevant accounting information ([R.sup.2] of 25 percent). Additional tests using interim accounting disclosures show that earnings remain value relevant during the first three quarters of 1994, but in the fourth (devaluation) quarter, earnings lose value relevance. In contrast, fundamental analysis, primarily the selling and administrative expense signal, results in significant value relevance in the devaluation quarter ([R.sup.2] of 24.7 percent). This finding indicates the increased contribution of fundamental analysis is due to the devaluation event, rather than to a gradual decline in economic conditions throughout 1994.
We also provide analyses that do not rely on an assumption that contemporaneous market reactions capture the full extent to which accounting information may be of value to investors. The primary purpose of those analyses is to determine (1) whether the fundamental signals provide forward-looking information after the 1994 devaluation and (2) whether analysts fully capture this information. To answer those questions, we examine associations of the fundamental signals with one-year-ahead earnings changes, analysts' revisions of their earnings forecasts, and analyst forecast errors. Two fundamental signals, gross margin and the selling and administrative expense signals, are significant in several of those analyses. Since the evidence indicates that analysts underutilize the information in 1994, an opportunity may exist in 1994 to use those signals to profit using a zero-investment trading strategy. Our tests show that a substantial profit can be earned in this manner. This test indicates that the market does not fully understand the increased value of fundamental analysis after an economic shock.
Our overall conclusion is that fundamental analysis assumes additional importance in the wake of the economic shock from a devaluation, but the market does not fully impound this fact. This finding presents the intriguing possibility that the more detailed accounting information used in fundamental analysis may provide value relevant information in the aftermath of other types of economic shocks (e.g., a terrorist attack, an oil embargo, or a labor strike).
Editorial Data
The following table contains information about turnaround time for manuscripts (including revisions) on which editorial decisions were made in the 12-month period ended February 28, 2003. Turnaround time is the number of days between the date that the manuscript was received and the date of the editor's letter to the author(s):
Number of
Manu- Cumulative Cumulative
scripts Number Percent
0 [less than or equal to] Days 70 70 18.45
[less than or equal to] 30
31 [less than or equal to] Days 137 207 54.62
[less than or equal to] 60
61 [less than or equal to] Days 121 328 86.54
[less than or equal to] 90
91 [less than or equal to] Days 41 369 97.36
[less than or equal to] 120
121 [less than or equal to] Days 10 379 100.00
The mean review time was 56.8 days; the median review time was 57 days.
TABLE 1
Descriptive Statistics for Mexican Firms
Panel A: Regression Variables (a) (means
(medians) reported in first (second) row)
Year R PTE Inv AR GM
1993 12.8% 25.6% -4.8% 10.5% 1.6%
n = 38 7.1 12.5 -3.1 10.0 -0.0
1994 6.2 0.7 21.5 20.4 2.2
n = 53 0.3 4.6 20.7 17.4 1.4
1995 -1.0 13.1 -5.8 -5.6 0.8
n = 70 -1.3 9.5 -6.4 -11.4 2.9
1996 2.0 14.7 -5.6 -4.5 4.0
n = 74 0.2 11.9 -6.0 -7.6 1.5
1997 -1.7 11.8 8.0 6.8 1.4
n = 91 -0.5 8.1 8.0 5.3 0.0
1998 -20.0 2.4 -0.8 0.4 0.1
n = 118 -16.6 3.5 -0.1 -2.9 0.0
Year S&A Tax Lev
1993 0.6% -1.8% 40.6%
n = 38 -0.6 -0.2 41.1
1994 0.6 -1.9 47.1
n = 53 1.7 -0.3 46.2
1995 -2.1 -1.7 45.7
n = 70 -1.3 -0.2 47.4
1996 -4.6 -0.8 42.4
n = 74 -3.9 -0.0 43.2
1997 -2.1 -1.8 41.9
n = 91 -1.4 -0.3 43.5
1998 3.5 -1.4 43.3
n = 118 3.4 -0.2 43.6
Panel B: Correlation Analysis (a)
PTE Inv AR GM S&A Tax Lev
R .39 .05 .07 -.14 -.30 .05 -.14
PTE -.04 .01 -.14 -.17 .01 -.21
Inv .30 .10 .10 .01 .00
AR -.07 .14 .01 -.04
GM -.15 -.04 .08
S&A .06 .01
Tax -.22
Panel C: Properties of Analysts' Forecasts in Mexico (b)
(means (medians) reported in first (second) row)
No. of
Dis- Analyst Forecast
Year REV FE persion (c) Following Revisions
1993 -0.68% 1.91% 20.6% 9.9 1.6
n = 35 -0.37 -1.04 17.8 10.0 2.0
1994 7.37 -9.18 33.4 10.5 2.8
n = 46 3.01 -4.25 15.8 11.0 3.0
1995 4.72 -2.45 115.9 12.5 2.9
n = 59 2.17 0.67 41.8 11.0 2.0
1996 -2.26 0.15 367.7 13.0 3.2
n = 62 -2.20 1.10 28.6 12.5 3.0
1997 -1.30 -0.59 22.0 13.4 4.2
n = 72 -0.66 -0.57 22.6 13.0 3.0
1998 3.08 -5.85 19.1 8.5 3.2
n = 80 1.52 -4.15 16.8 8.0 2.0
(a) The sample consists of 444 firm-year observations from
1993-1998 with the requisite financial statement and price
data from the Economatica database. To control for outliers,
all variables in the model are winsorized at the 5 percent
and 95 percent tails of their respective distributions.
Numbers reported in bold indicate significance at the
[alpha] = .05 level (two-tailed) using parametric t-tests
or the nonparametric Wilcoxon test for medians.
R = market-adjusted 12-month return ending three months
after the fiscal-year-end; PTE = pre-tax operating-earnings-per-share
multiplied by 1 minus the tax rate in the preceding year,
deflated by price at the beginning of the return accumulation
period; Inv = percentage change in inventory minus the percentage
change in sales; AR = percentage change in accounts receivable
minus the percentage change in sales; GM = percentage change in
sales minus the percentage change in gross margin;
S&A = percentage change in selling and administrative
expenses minus the percentage change in sales; Tax = pre-tax
earnings-per-share multiplied by the difference in effective
tax rates for the preceding year and the current year; and
Lev = total liabilities divided by total assets.
(b) Analysts' forecasts were obtained from the I/B/E/S
International Summary database. REV = [(Post[F.sub.t+1] - [E.sub.t])
- (Pre[F.sub.t+1] - [F.sub.t])]/[P.sub.t-1] where Post[F.sub.t+1]
is the first available mean consensus forecast for earnings in year
t+1 subsequent to the year t earnings announcement; [E.sub.t] is
realized earnings in year t; Pre[F.sub.t+1] is the first available
mean consensus forecast in year t for earnings in year t+1;
[F.sub.t] is the first available mean consensus forecast in year
t for year t earnings; and [P.sub.t-1] is price at the beginning
of the returns accumulation period in year t. FE = ([E.sub.t]
- [F.sub.t])/[P.sub.t-1]. Dispersion is measured as the standard
deviation of all available forecasts that comprise [F.sub.t]
divided by [F.sub.t]. Analyst Following is equal to the number
of forecasts that comprise [F.sub.t]. No. of Forecast Revisions
represents the number of analysts who changed their forecasts
of [F.sub.t] either up or down from their last forecast.
(c) Dispersion was calculated for all companies where the analyst
following is greater than 1. The n for each year for this variable
is 1993 = 31; 1994 = 45; 1995 = 58; 1996 = 62; 1997 = 69; 1998 = 75.
TABLE 2
Benchmark Model: Earnings/Return Relation
for Sample of Mexican Firms (a)
[R.sub.it] = [[alpha].sub.0] + [[beta].sub.1][E.sub.it] + [v.sub.it]
Panel A: Tests for Annual Periods
Adj.
Intercept E [R.sup.2]
Devaluation Year: 1994 (n = 53)
Coefficient 0.065 0.213 -0.2%
(t-statistic) (1.24) (0.95)
1993; 1995-1998 (n = 391)
Pooled Coefficient -0.115 0.694 21.7%
(t-statistic) (-7.12) (10.5)
Mean Annual Coefficient -0.085 0.596
[No. Positive] [1] [5]
(Inter-temporal t-statistic) (-2.40) (6.16)
Difference in Earnings Coefficients
Ho: [[beta].sub.1;94 --
[[beta].sub.1;93,95-98] = 0
t-statistic -2.52
p-value (.012)
Panel B: Tests For 1994 Subperiods
Adj.
Intercept E [R.sup.2]
1st, 2nd, 3rd Quarters (n = 45)
Coefficient -0.030 0.464 7.2%
(t-statistic) (-0.92) (2.10)
4th Quarter (n = 45)
Coefficient 0.019 -0.109 1.9%
(t-statistic) (0.27) (-0.45)
(a) The sample consists of 444 firm-year observations from 1993-1998
with the requisite financial statement and price data from the
Economatica database. To control for outliers, all variables in the
model are winsorized at the 5 percent and 95 percent tails of their
respective distributions. Numbers reported in bold indicate
statistical significance at the [alpha] = .05 level (one-tailed).
[R.sub.it] = market-adjusted 12-month return for firm i ending three
months following the end of fiscal year t; and [E.sub.it] =
earnings-per-share for firm i in year t, deflated by price
at the beginning of the return accumulation period.
TABLE 3
Results of Earnings and Fundamental Analysis Signals
Regressed on Contemporaneous Returns (a)
[R.sub.it] = [[alpha].sub.0] + [[beta].sub.1]PT[E.sub.it]
+ [[beta].sub.2][Inv.sub.it] + [[beta].sub.3]A[R.sub.it]
+ [[beta].sub.4]G[M.sub.it] + [[beta].sub.5]S&[A.sub.it]
+ [[beta].sub.6][Tax.sub.it] + [[beta].sub.7][Lev.sub.it]
+ [[omega].sub.it]
Panel A: Tests for Annual Periods
Intercept PTE Inv
(+/-) (+) (+/-)
Devaluation Year: 1994 (n = 53)
Coefficient 0.316 -0.164 -0.014
(t-statistic) (1.69) (-0.63) (-0.08)
1993; 1995-1998 (n = 391)
Pooled Coefficient -0.064 0.600 0.069
(t-statistic) (-1.61) (8.77) (1.09)
Mean Annual Coefficient -0.042 0.553 0.087
[No. Positive] [2] [5] [3]
(Inter-temporal t-statistic) (-0.57) (4.21) (1.01)
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94 =
[[beta].sub.i;93,95-98]
t-statistic -3.21
p-value (.001)
Panel B: Tests for 1994 Subperiods (b)
Intercept PTE Inv
(+/-) (+) (+/-)
1st, 2nd, 3rd Quarters (n = 45)
Coefficient 0.039 0.384 -0.124
(t-statistic) (0.42) (1.59) (-0.88)
4th Quarter (n = 45)
Coefficient 0.162 -0.449 -0.077
(t-statistic) (0.72) (-2.08) (-0.49)
Panel A: Tests for Annual Periods
AR GM S&A
(-) (-) (-)
Devaluation Year: 1994 (n = 53)
Coefficient 0.037 -0.423 -0.985
(t-statistic) (0.23) (-1.23) (-3.95)
1993; 1995-1998 (n = 391)
Pooled Coefficient 0.016 -0.287 -0.499
(t-statistic) (0.28) (-2.37) (-5.22)
Mean Annual Coefficient -0.057 -0.253 -0.423
[No. Positive] [2] [0] [0]
(Inter-temporal t-statistic) (-0.50) (-2.55) (-1.96)
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94 =
[[beta].sub.i;93,95-98]
t-statistic -0.42 -2.03
p-value (0.675) (.043)
Panel B: Tests for 1994 Subperiods (b)
AR GM S&A
(-) (-) (-)
1st, 2nd, 3rd Quarters (n = 45)
Coefficient 0.072 -0.300 -0.409
(t-statistic) (0.82) (-0.75) (-1.67)
4th Quarter (n = 45)
Coefficient 0.204 0.547 -1.420
(t-statistic) (1.83) (0.47) (-1.99)
Panel A: Tests for Annual Periods
Tax Lev Adj.
(+/-) (-) [R.sup.2]
Devaluation Year: 1994 (n = 53)
Coefficient -0.608 -0.541 25.0%
(t-statistic) (-0.61) (-1.38)
1993; 1995-1998 (n = 391)
Pooled Coefficient 0.556 -0.098 26.2%
(t-statistic) (1.72) (-1.18)
Mean Annual Coefficient 1.241 -0.077
[No. Positive] [4] [2]
(Inter-temporal t-statistic) (2.19) (-0.61)
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94 =
[[beta].sub.i;93,95-98]
t-statistic
p-value
Panel B: Tests for 1994 Subperiods (b)
Tax Lev Adj.
(+/-) (-) [R.sup.2]
1st, 2nd, 3rd Quarters (n = 45)
Coefficient 0.179 -0.169 9.4%
(t-statistic) (0.10) (-0.81)
4th Quarter (n = 45)
Coefficient 1.004 -0.600 24.7%
(t-statistic) (0.71) (-1.38)
Panel A: Tests for Annual Periods
Benchmark
Adj. Partial
[R.sup.2] F-statistic
Devaluation Year: 1994 (n = 53)
Coefficient -0.2% 3.94
(t-statistic)
1993; 1995-1998 (n = 391)
Pooled Coefficient 21.7% 4.95
(t-statistic)
Mean Annual Coefficient
[No. Positive]
(Inter-temporal t-statistic)
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94 =
[[beta].sub.i;93,95-98]
t-statistic
p-value
Panel B: Tests for 1994 Subperiods (b)
Benchmark
Adj. Partial
[R.sup.2] F-statistic
1st, 2nd, 3rd Quarters (n = 45)
Coefficient 7.2% 1.21
(t-statistic)
4th Quarter (n = 45)
Coefficient -1.9% 3.62
(t-statistic)
(a) The sample consists of 444 firm-year observations from 1993-1998
with the requisite financial statement and price data from the
Economatica database. To control for outliers, all variables in the
model are winsorized at the 5 percent and 95 percent tails of their
respective distributions. Numbers reported in bold indicate
significance consistent with the direction prediction at the
[alpha] = .05 level (one-tailed or two-tailed as indicated in the
column heading). The partial F-statistic tests whether the
fundamental variables provide a statistically significant explanatory
power for returns over the benchmark earnings model.
R = market-adjusted 12-month return for firm i ending three months
following the end of fiscal year t;
PTE = pre-tax operating-earnings-per-share multiplied by 1 minus the
tax rate in the preceding year, deflated by price at the beginning
of the return accumulation period;
Inv = percentage change in inventory minus the percentage change in
sales;
AR = percentage change in accounts receivable minus the percentage
change in sales;
GM = percentage change in sales minus the percentage change in
gross margin;
S&A = percentage change in selling and administrative expenses minus
the percentage change in sales;
Tax = pre-tax earnings-per-share multiplied by the difference in
effective tax rates for the preceding year and the current year; and
Lev = total liabilities divided by total assets.
(b) The first subperiod uses deflated earnings for the first three
quarters of the year regressed on market-adjusted returns beginning
April 1, 1994 (so that calendar 1993 earnings have been released)
and extending through November 30, 1994 (so that third quarter
earnings have been released). The second subperiod uses fourth
quarter earnings regressed on returns surrounding the devaluation
event, covering December 1, 1994 through March 31, 1995
(see Figure 1, Panel B).
TABLE 4
Results of Earnings and Fundamental Analysis Signals Regressed
on Future Earnings Changes, Analysts' Forecast Revisions, and
Analysts' Forecast Errors
[Y.sub.it] = [[alpha].sub.0] + [[beta].sub.1]PT[E.sub.it]
+ [[beta].sub.2][Inv.sub.it] + [[beta].sub.3]A[R.sub.it]
+ [[beta].sub.4]G[M.sub.it] + [[beta].sub.5]S&[A.sub.it]
+ [[beta].sub.6][Tax.sub.it] + [[beta].sub.7][Lev.sub.it]
+ [[omega].sub.it]
Panel A: Time Period
Intercept PTE Inv
(+/-) (-) (+/-)
Devaluation Year: 1994 (n = 46)
Y = [DELTA][E.sub.t+1] 0.137 -0.508 ** 0.006
Y = RE[V.sub.t] -0.039 -0.169 * -0.021
Y = AF[E.sub.t+1] 0.120 -0.126 0.001
1993, 1995-1998 (n = 308)
Y = [DELTA][E.sub.t+1]
Pooled Coefficient 0.039 -0.172 ** -0.057
Mean Annual Coefficient 0.057 * -0.088 0.007
Y = RE[V.sub.t]
Pooled Coefficient -0.012 -0.111 ** -0.021
Mean Annual Coefficient -0.008 -0.078 -0.035
Y = AF[E.sub.t+1]
Pooled Coefficient 0.031 -0.025 0.001
Mean Annual Coefficient 0.043 0.037 0.078
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94] =
[[beta].sub.i;93,95-98]
Y = [DELTA][E.sub.t+1] t-statistic -2.18 *
Y = RE[V.sub.t] t-statistic -0.74
Y = AF[E.sub.t+1] t-statistic -0.72
Panel B: Dependent Variable
Intercept PTE Inv
(+/-) (-) (+/-)
Forecast Revisions Surrounding 0.005 -0.040 -0.005
December 1994 Devaluation for (0.45) (-1.54) (-0.48)
1995 Annual Earnings
Panel A: Time Period
AR GM S&A
(-) (-) (-)
Devaluation Year: 1994 (n = 46)
Y = [DELTA][E.sub.t+1] -0.030 -0.282 ** -0.517 **
Y = RE[V.sub.t] 0.012 -0.150 * -0.140 **
Y = AF[E.sub.t+1] -0.035 -0.225 * -0.337 **
1993, 1995-1998 (n = 308)
Y = [DELTA][E.sub.t+1]
Pooled Coefficient 0.025 -0.307 ** -0.194 **
Mean Annual Coefficient 0.042 -0.251 * -0.179
Y = RE[V.sub.t]
Pooled Coefficient 0.002 -0.057 0.032
Mean Annual Coefficient 0.019 0.035 0.022
Y = AF[E.sub.t+1]
Pooled Coefficient 0.019 -0.266 ** -0.240 **
Mean Annual Coefficient 0.028 -0.267 * -0.198
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94] =
[[beta].sub.i;93,95-98]
Y = [DELTA][E.sub.t+1] 0.14 -2.49 *
Y = RE[V.sub.t] -1.05 -2.65 **
Y = AF[E.sub.t+1] 0.25 -0.82
Panel B: Dependent Variable
AR GM S&A
(-) (-) (-)
Forecast Revisions Surrounding 0.009 -0.170 -0.050
December 1994 Devaluation for (0.95) (-4.38) (-2.19)
1995 Annual Earnings
Panel A: Time Period
Tax Lev Adj.
(+/-) (-) [R.sup.2]
Devaluation Year: 1994 (n = 46)
Y = [DELTA][E.sub.t+1] 0.193 -0.053 57.9%
Y = RE[V.sub.t] -0.361 0.241 51.4%
Y = AF[E.sub.t+1] 0.085 -0.228 26.7%
1993, 1995-1998 (n = 308)
Y = [DELTA][E.sub.t+1]
Pooled Coefficient -0.467 * -0.053 9.9%
Mean Annual Coefficient -0.182 -0.146
Y = RE[V.sub.t]
Pooled Coefficient -0.320 ** 0.072 15.9%
Mean Annual Coefficient -0.201 * 0.042
Y = AF[E.sub.t+1]
Pooled Coefficient 0.002 -0.168 ** 11.0%
Mean Annual Coefficient 0.085 -0.242
Difference in Earnings Coefficients
Ho: [[beta].sub.i;94] =
[[beta].sub.i;93,95-98]
Y = [DELTA][E.sub.t+1] 1.27 0.00
Y = RE[V.sub.t] -0.16 1.55
Y = AF[E.sub.t+1] 0.18 -0.30
Panel B: Dependent Variable
Tax Lev Adj.
(+/-) (-) [R.sup.2]
Forecast Revisions Surrounding 0.115 0.011 44.7%
December 1994 Devaluation for (1.25) (0.48)
1995 Annual Earnings
*, ** Indicate significance at the [alpha] = .05 and .01 levels,
respectively, using a one-tailed test where directional predictions
are provided; otherwise, two-tailed. All significant amounts are
reported in bold type.
The sample consists of 354 firm-year observations from 1993-1998 with
the requisite financial statement and price data from the Economatica
database and analysts' forecast data from I/B/E/S. To control for
outliers, all variables in the model are winsorized at the 5 percent
and 95 percent tails of their respective distributions.
Panel A: [DELTA][E.sub.t+1] = actual earnings-per-share in year
t+1 minus actual earnings-per-share in year t, deflated by price
at the beginning of the returns accumulation period (all data obtained
from I/B/E/S);
RE[V.sub.t] = analyst earnings forecast revision for year t+1 using
data obtained from I/B/E/S:
[(Post[F.sub.t+1] - [E.sub.t]) - (Pre[F.sub.t+1] - [F.sub.t])]/
[P.sub./t-1]
where:
Post[F.sub.t+1] = the first forecast of t+1 earnings available
subsequent to the announcement of time period t earnings;
Pre[F.sub.t+1] = the first forecast of t+1 earnings available in
time period t;
[F.sub.t] = the first forecast of t earnings available in time
period t;
[E.sub.t] = actual earnings for time period t;
[P.sub.t-1] = price at the beginning of the returns accumulation
period.
AF[E.sub.it+1] = actual earnings-per-share in year t+1 minus the
first forecasted earnings-per-share for year t+1 subsequent to year
t's earnings announcement, deflated by price at the beginning of the
returns accumulation period;
PTE = pre-tax operating-earnings-per-share multiplied by 1 minus the
tax rate in the preceding year, deflated by price at the beginning of
the return accumulation period;
Inv = percentage change in inventory minus the percentage change in
sales;
AR = percentage change in accounts receivable minus the percentage
change in sales;
GM = percentage change in sales minus the percentage change in gross
margin;
S&A = percentage change in selling and administrative expenses minus
the percentage change in sales;
Tax = pre-tax earnings-per-share multiplied by the difference in
effective tax rates for the preceding year and the current year; and
Lev = total liabilities divided by total assets.
Panel B: The February 1995 mean consensus forecast minus the December
1994 mean consensus forecast for 1995 earnings deflated by price as of
April 1, 1994. The fundamental signals are calculated using accounting
balances reported in the third quarter of 1994. Coefficients reported
with t-values in parentheses.
TABLE 5
Mean Hedged Portfolio Return from Fundamental Analysis and Perfect
Foresight Strategies (a)
Fundamental Signal Equation:
[R.sub.it+1] = [[alpha].sub.0] + [[beta].sub.1[PTE].sub.it] +
[[beta].sub.2][SIZE.sub.it] + [[beta].sub.3][MB.sub.it] +
[[beta].sub.4][Inv.sub.it] + [[beta].sub.5][AR.sub.it] +
[[beta].sub.6][GM.sub.it] + [[beta].sub.7]S&[A.sub.it] +
[[beta].sub.8][Tax.sub.it] + [[beta].sub.9][Lev.sub.it] +
[[omega].sub.it]
Perfect Foresight Equation:
[R.sub.it+1] = [[gamma].sub.0] + [[gamma].sub.1][E.sub.it+1] +
[[gamma].sub.2][SIZE.sub.it] + [[gamma].sub.3]M[B.sub.it] +
[[epsilon].sub.it]
Perfect
Foresight
Stragtegy
Signals Derived [[gamma] p-
Return Window From n .sub.1] value
Hedged Portfolio Returns
Based on 1994 Data
Mar. 1, 1995-Apr. 30, 1996 1994 3rd Q Data 44 0.805 <.01
May 1, 1995-Apr. 30, 1996 1994 Annual Data 51 0.928 <.01
Hedged Portfolio Returns
Based on Non-1994 Data
Apr. 1, 1994-Mar. 31, 1995 1993 Annual Data 38 0.202 .18
Apr. 1, 1996-Mar. 31, 1997 1995 Annual Data 69 0.483 <.01
Apr. 1, 1997-Mar. 31, 1998 1996 Annual Data 74 0.382 .01
Apr. 1, 1998-Mar. 31, 1999 1997 Annual Data 89 0.465 <.01
Apr. 1, 1999-Mar. 31, 2000 1998 Annual Data 97 0.447 <.01
Fundamental Analysis Strategy
Mean Hedged
Return =
[MATHEMATICAL
EXPRESSION NOT Signals with
REPRODUCIBLE Significant
Return Window IN ASCII] p-value Coefficients
Hedged Portfolio Returns
Based on 1994 Data
Mar. 1, 1995-Apr. 30, 1996 0.973 .07 GM, S&A
May 1, 1995-Apr. 30, 1996 0.585 .10 GM, S&A
Hedged Portfolio Returns
Based on Non-1994 Data
Apr. 1, 1994-Mar. 31, 1995 -0.241 .62 *
Apr. 1, 1996-Mar. 31, 1997 0.211 .42 *
Apr. 1, 1997-Mar. 31, 1998 0.112 .76 *
Apr. 1, 1998-Mar. 31, 1999 0.131 .62 *
Apr. 1, 1999-Mar. 31, 2000 -0.128 .42 *
Fundamental Analysis
Strategy Using Only
GM and S&A Signals
Mean Hedged
Return =
[MATHEMATICAL
EXPRESSION NOT
REPRODUCIBLE
Return Window IN ASCII] p-value
Hedged Portfolio Returns
Based on 1994 Data
Mar. 1, 1995-Apr. 30, 1996 0.907 <.01
May 1, 1995-Apr. 30, 1996 0.904 <.01
Hedged Portfolio Returns
Based on Non-1994 Data
Apr. 1, 1994-Mar. 31, 1995 0.099 .76
Apr. 1, 1996-Mar. 31, 1997 -0.060 .71
Apr. 1, 1997-Mar. 31, 1998 0.274 .13
Apr. 1, 1998-Mar. 31, 1999 -0.121 .42
Apr. 1, 1999-Mar. 31, 2000 -0.040 .85
(a) The sample consists of all available observations from 1993-1998
with the requisite financial statement and price data from the
Economatica database. To control for outliers, all variables in the
model are winsorized at the 5 percent and 95 percent tails of their
respective distributions.
[R.sub.it+1] = future (relative to when the fundamental signals are
calculated) market-adjusted return corresponding to
the return window as indicated in the table;
PTE = pre-tax operating-earnings-per-share multiplied by 1
minus the tax rate in the preceding year, deflated
by price at the beginning of the return accumulation
period;
Inv = percentage change in inventory minus the percentage
change in sales;
AR = percentage change in accounts receivable minus the
percentage change in sales;
GM = percentage change in sales minus the percentage
change in gross margin;
S&A = percentage change in selling and administrative
expenses minus the percentage change in sales;
Tax = pre-tax earnings-per-share multiplied by the
difference in effective tax rates for the preceding
year and the current year;
Lev = total liabilities divided by total assets;
SIZE = the natural log of the market value of equity;
MB = market value of equity divided by the book value
of equity; and
E = net income, deflated by price at the beginning of
the return accumulation period.
TABLE 6
Portfolio Returns to Long and Short Positions in Mexican Companies
Based on Individual Fundamental Signals (a)
Funamental Signal = GM
Long Portfolio Short Portfolio
Year n Mean Median n Mean Median
1994 12 0.215 * 0.190 * 13 -0.198 -0.363
Funamental Signal = S&A
Long Portfolio Short Portfolio
Year n Mean Median n Mean Median
1994 12 0.246 * 0.221 * 13 -0.327 -0.409
* The long (short) portfolio return is significantly greater than
the short (long) portfolio return at the [alpha] = .01 level using
a two-tailed test.
(a) Each fundamental signal is defined as before. The entire sample
is ranked into four portfolios based on the associated fundamental
signal. The long (short) portfolio is comprised of firms with the
most (least) favorable signal. The returns period runs from
May 1, 1995 to April 30, 1996.
FIGURE 1
Inflation and Exchange Rate Statistics (per Banco de Mexico)
Panel A: Annual Inflation Rates (Mexican CPI)
1992 11.94
1993 8.01
1994 7.05
1995 51.97
1996 27.7
1997 15.72
1998 18.61
Note: Table made from bar graph.
[GRAPHIC OMITTED]
We acknowledge Brian Bushee, Bill Cready, Mahendra Gujarathi, Don Herrmann, Mary Lea McAnally, Gary Meek, Grace Pownall, Mike Wilkins, two anonymous reviewers, and participants at workshops at Oklahoma State University, Texas A&M University, and University of North Texas for helpful comments on earlier drafts. We also thank C. P. Francisco A. Castor del Valle of KPMG Cardenas Dosal, S. C. Peat Marwick, and C. P. Javier Cocina Martinex of Instituto Mexicano de Contadores Publicos, A.C., for meeting with the authors to answer questions about the Mexican accounting system. The Center for International Business Studies at Texas A&M University provided summer funding. We gratefully acknowledge the contribution of I/B/E/S International Inc. for providing earnings per share forecast data, available through the Institutional Brokers Estimate System. This data has been provided as part of a broad academic program to encourage earnings expectations research.
Editor's note: This paper was accepted by Terry Shevlin, Senior Editor.
(1) In using the Economatica database, researchers need to make two types of adjustments to the data. First, the signs of certain earnings components are inconsistent. For example, a positive number for net interest expense sometimes represents an expense that is subtracted from net income and other times represents net interest income that is added to net income. We adjusted for this inconsistency by downloading from Economatica various income subtotals, which allowed us to determine the tree sign of an income statement component. Second, the inflation adjustment factor applied to financial statement items is sometimes inconsistent. In particular, we discovered that for most (but not all) of the companies in our sample, the 1994 data were adjusted using the 1995 inflation-adjustment factor; this problem exists to a much lesser extent in other years (we are grateful to an anonymous reviewer for pointing this problem out to us). We made corrections by downloading from Economatica both the reported and inflation-adjusted financial statement numbers and comparing them to determine the inflation-adjusted factor that was actually used by Economatica. We adjusted the data using a consistent factor. In a few instances, we determined the nature of the amounts reported by referring to the Worldscope database.
(2) Readers interested in an example of RCPLA accounting should see Gordon (2001, Appendix). Figure 1, Panel A, summarizes the magnitude of price changes over our test period.
(3) We also examined earnings/returns models that included as independent variables only deflated earnings change and deflated earnings change together with deflated earnings level. Consistent with Gordon (2001, 193), we find that earnings changes add little explanatory power.
(4) We obtained the 1994 third-quarter earnings announcement date for all companies where this variable is available on I/B/E/S. Although earnings announcement dates are available in this quarter for only 15 companies, every announcement was made prior to November 30, 1994.
(5) All data to construct [DELTA][E.sub.it+1], AF[E.sub.it+1], and RE[V.sub.it] were obtained from I/B/E/S to ensure that we use the realized earnings construct that analysts are trying to forecast. We do not adjust these dependent variables to a constant purchasing power. Deflating by stock price at the end of March reduces the need to do so. In addition, analysts' earnings forecasts would include anticipated inflation, so an additional adjustment would double-count inflation.
(6) LT report their results using the average for the last two years as a measure for the expected level of each variable. LT (1993, 194) state that expectation models using just the prior year yield very similar results. This study uses the single year (random walk) formulation to avoid losing another year of results. This practice allows us to include new firms in the second year, rather than the third year, in which they are included in the Economatica database.
(7) In earlier analyses, several other variables that are unique to Mexico and/or might be expected to be important to investors during a currency devaluation were included in the full model. These variables included variations of foreign debt as a percentage of total debt, foreign sales as a percentage of total sales, holding gains/losses (which are reported by Mexican companies in stockholders' equity), and integral cost of financing (which is an income statement item that consists of the sum of purchasing power gains/losses, exchange rate gains/losses, and net interest). None of these variables provided a consistent improvement for our tests and separate reporting of the integral cost of financing resulted in multicollinearity problems in several tests. Thus, these variables are not reported in our final models.
(8) Interpretation of the tax signal is even more complicated because of the existence of a provision that allows recovery under some circumstances of the amount by which an asset tax exceeded the amount that would have been paid under an income tax that year. And during our sample period, the carry-forward recovery period was changed from five to ten years. The asset tax is roughly 2 percent of assets, although some liabilities can be deducted from the asset base (but not debts owed to foreign entities).
(9) We use the expression "pretax earnings" to be consistent with LT's terminology, but this earnings measure has been reduced by taxes calculated using the tax rate from the prior year.
(10) Note that, with a sample size of five years, the t-value of 1.96 for S&A is not statistically significant.
(11) This discussion relies on the pooled data, although we also present mean annual coefficients for the nondevaluation years. The latter are often statistically insignificant because only five annual observations are available for the statistical tests.
(12) I/B/E/S indicates that the mean consensus forecasts are as of the Thursday before the third Friday of every month, which would be as of the 15th for the December 1994 forecast period--just four business days prior to the beginning of the devaluation.
(13) For every company in our sample, 1994 earnings had not yet been reported at the time the I/B/E/S February forecasts were published. Further, of the 40 companies with sufficient quarterly financial statement and analyst forecast data to be included in our tests, every one had a non-zero forecast revision, indicating the analysts were reacting to the devaluation. Also, 60 percent of the forecast revisions were negative.
(14) Abarbanell and Bushee (1998) use equity beta to control for risk. Many of the firms in our sample are thinly traded, which causes problems in measuring beta. Moreover, beta has been criticized as a poor measure of risk (Fama and French 1992).
(15) This step is executed by dividing all the fundamental signals by -9 except for Inv.
(16) Size is significant in both the fundamental signal and perfect foresight equation for years 1994 and 1997. MB is significant in the fundamental signal (perfect foresight) equation in 1993 and 1994 (1997).
(17) Given the delay of companies reporting 1994 financial results, the return windows based on 1994 data are extended one month.
(18) This result differs from Abarbanell and Bushee (1998) who found an average 12-month cumulative size-adjusted abnormal return of 13.2 percent for a sample of U.S. firms over the period 1974-1988.
(19) Graham et al. (2000) and Davis-Friday and Gordon (2002) perform similar analyses and also find that, after controlling for negative earnings in the devaluation year, positive earnings has a significant association with firm value.
REFERENCES
Abarbanell, J. S., and B. J. Bushee. 1997. Fundamental analysis, future earnings, and stock prices. Journal of Accounting Research 35 (Spring): 1-24.
--, and --. 1998. Abnormal returns to a fundamental analysis strategy. The Accounting Review 73 (January): 19-45.
Amir E., B. Lev, and T. Sougiannis. 1999. What value analysts? Working paper, New York University, New York, NY.
Anderson, M. C., D. Banker, and S. Janakiraman. 2000. Are selling, general, and administrative costs sticky? Journal of Accounting Research 41 (March): 47-63.
Brown, P., G. Foster, and E. Noreen. 1985. Security Analyst Multi-Year Earnings Forecasts and the Capital Market. Sarasota, FL: American Accounting Association.
Capstaff, J., K. Paudyal, and W. Rees. 1998. Analysts' forecasts of German firms' earnings: A comparative analysis. Journal of International Financial Management and Accounting 9: 83-116.
Davis-Friday, P., and E. Gordon. 2002. The effect of macroeconomic changes on the value-relevance of accounting information: The case of Mexico and the 1995 financial crisis. Working paper, University of Notre Dame, Notre Dame, IN.
Fama, E., and J. Macbeth. 1973. Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81 (May-June): 607-636.
--, and K. French. 1992. The cross-section of expected stock returns. Journal of Finance 47 (June): 427-465.
1998. Market efficiency, long-term returns, and behavioral finance. Journal of Financial Economics 49 (September): 283-306.
Francis, J., and D. Philbrick. 1993. Analysts' decisions as products of multi-task environment. Journal of Accounting Research 31 (Autumn): 216-230.
Gordon, E. A. 2001. Accounting for changing prices: The value relevance of historical cost, price level, and replacement cost accounting in Mexico. Journal of Accounting Research 39 (June): 177-200.
Graham, R., R. King, and J. Bailes. 2000. The value relevance of accounting information during a financial crisis: Thailand and the 1997 decline in the value of the baht. Journal of International Financial Management and Accounting 11 (Summer): 84-107.
Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20 (September): 125-153.
Ho, L. J., C. Liu, and P. S. Sohn. 2001. The value relevance of accounting information around the 1997 Asian financial crisis--The case of South Korea. Asia-Pacific Journal of Accounting & Economics 8: 83-107.
Instituto Mexicano de Contadores Publicos (IMCP). 1984. Principios de Contabilidad Generalmente Aceptados. Bulletin B-10, Reconocimiento de los Efectos de la Inflacion en la Informacion Financiera. Mexico City, Mexico: IMCP.
Lev, B., and S. R. Thiagarajan. 1993. Fundamental information analysis. Journal of Accounting Research 31 (Autumn): 190-215.
O'Brien, P. C. 1988. Analysts' forecasts as earnings expectations. Journal of Accounting and Economics 10: 53-83.
Penman, S. H. 2001. Financial Statement Analysis & Security Valuation. New York, NY: McGraw-Hill/Irwin.
Submitted January 2002
Accepted January 2003
Edward P. Swanson
Lynn Rees
Texas A&M University
Luis Felipe Juarez-Valdes
Universidad de las Americas-Puebla
In addition, make sure to read these articles:
- Five Advantages to Leasing Office and Technical Equipment
- Using Quickbooks to Develop Financials Part I
- eBay Q3 Earnings and eBay Express
- Valuable goods.
- Surviving the game.
- Schwager on Futures: Fundamental Analysis.
- Built to last.
- A Test of a Methodological Explanation For the Value Irrelevance of Losses.