Predictability of Earnings and Reversion of Profitability

1. INTRODUCTION, RESEARCH QUESTION AND CONTRIBUTION

In a competitive environment, economists say there is a mean reversion of profitability. Mean reversion of profitability infers that variation in profitability and earnings can be predicted. Although there are some literatures making an effort to find prediction in profitability and earnings, the findings somewhat cannot fully explain those variations. Early researches (Beaver 1970; Brooks and Buckmaster 1976; and Lookabill 1976) did not test the prediction formally. When there were formal tests, models were mostly time-series and identified only companies with long-earning histories (20 years). This approach causes the issue that long-term survivors might not represent all the firms. Furthermore, 20 years of data on earnings is an inaccurate estimation of the time-series model. Thus, the results found are statistically weak (Lev 1969; Freeman, Ohlson, and Penman 1982). There are some later researches attempting to identify those variations as well. Freeman et al (1982), Collins and Kothari (1989), Easton and Zmijewski (1989), Ou and Penman (1989), Elgers and Lo (1994) and Basu (1997) found that cross-sectional tests constructed more consistent evidence of predictability. However, Elgers and Lo (1994) found the unrealistic assumption that there is no correlation among companies due to changes in earnings and profitability. Moreover, most existing literatures do not investigate connection of the predictability of profitability and that of earnings. Contrastingly, like Freeman et al. (1982) and Lev (1983), this research paper is to answer the question: Is much of what is predictable about earnings due to the mean reversion of profitability? The result confirms the answer to question is yes. Those results are applicable to the real world. Therefore, the main contribution is the confirmation of economists’ presumption that there is a mean reversion of profitability in a competitive environment.

2. data and methodology

2.1 A First-Pass Partial Adjustment Model for Profitability

This test uses a simple cross-section partial adjustment regression in profitability changing for each year t from 1964 to 1995. This regression from t to t+1 is as followed

          (1a)

                           (1b)

where is total book assets of firm at the end of year t,is earnings before interest,is profitability measure, is expected value of profitability measure, is profitability change from year t to t+1 and is the profitability deviation from the expected value.

    The paper uses a two-step method to identify equation (1). After doing regression to investigate differences in expected profitability among companies, the fitted values from the first-step regression are used as the proxy for in the cross-section regression.

    In the first-stage regression, (dividends to book value of equity at the end of year t) is used as proxy for expected profitability, (dummy variable to capture nonlinear relationship of dividends and expected profitability) and (market-to-book ratio to find variation of expected profitability which cannot find by dividend determinants. In the cross-section regression, in (1) is the fitted value from the first-stage regression.

    (2)

    Due to the high regulation during the sample period (1964-1995), financial companies and utilities are omitted. This paper considers only the firms with more than $10 million assets and more than $5 million book equity. With these exclusions, 2,343 companies per year are taken into account.

The interpretation is based on the average slopes and the time-series standard errors of the average slopes. However, with only 33 slope observation from 1964 to 1995, the estimation of autocorrelation is inaccurate. Therefore, this paper uses a less strict approach with t-statistics requirement of about 2.8 rather than the common 2.0.

2.2 A Nonlinear Partial Adjustment Model for Profitability

This test is developed to investigate whether there is comparable nonlinearity in profitability characteristics with the hypothesis that the mean reversion of profitability results in the predictability of earnings. The nonlinear partial adjustment model equation is expanded from equation (1).

(3)

         where is the negative deviations of profitability from expected values, is the squared negative deviations, is the squared positive deviations, is negative changes in profitability, is squared negative changes and is squared positive changes. , and are to capture the nonlinearity in the mean reversion of profitability and , and are to capture the nonlinearity in the profitability changes autocorrelation.

2.3 Predicting Earnings

Freeman et al. (1982) and Lev (1983) argue that the competition causes mean reversion of profitability. This paper inspects the predictable changes in earnings and how much of the predictability brings the nonlinearity of mean reversion in profitability. The dependent variable is change in earnings, . The regression of change in earnings is

(4)

       where is negative changes in earnings, is squared negative changes in earnings and is squared positive changes in earnings. , and are to capture the nonlinearity in the earning changes autocorrelation.

3. results

3.1 A First-Pass Partial Adjustment Model for Profitability

The negative slope of implies that there is a nonlinear relationship of dividends and profitability. The significant positive slope of confirms the hypothesis that market-to-book ratio investigates variation of expected profitability which cannot find by dividend determinants.

3.2 A Nonlinear Partial Adjustment Model for Profitability

3.3 Predicting Earnings