Some Statistical Properties of Models of Transitory Earnings

Abstract

The fact that most of the financial ratios are mean-reverting, is well known. Due to the importance of earnings forecast accuracy, relevant scientific literature in this area concentrates on transitory earnings. The main models used for description of earnings and/or profitability time series are adaptive expectation, autoregressive and partial adjustment models. However, their construction implies severe drawbacks like assumption of intentional adjustment of earnings, sometimes even towards unknown target or towards company-specific target uninfluenced by market, instead of rather realistic assumption of random push of market forces, as we found earlier. This paper proposes a model of mechanical mean reversion of earnings (and/or other company financial data, including ratios). Simulation-based tests of accuracy in a cyclical environment and robustness to input variables non-normality show that the proposed model is more accurate and less biased in capturing the reversion of earnings to industry averages, compared to the most commonly used partial adjustment models.