Earnings Forecast Errors: A Decomposition Analysis

Theil decomposition method can help to determine the cause of error...the method is applied to the I/B/E/S' earnings per share forecast errors of nine money center banks...the quality of forecasts improves as forecast horizon shortens. The Theil decomposition method helps to analyze forecast error by breaking it down the into three components--mean bias proportion (how average predicted values systematically over-or under-predict the actual values), regression bias proportion (how well the predictions track turning points in the actual values) and random proportion (proportion of errors occurring randomly). The first two components (mean bias and regression bias) are called systematic components and the last one--random--is called non-systematic component. The forecasts are the best if the value of non-systematic component comes close to one. The study of the I/B/E/S' earnings per share forecasts of nine banks reveals that their forecasts are fairly good as their non-systematic components are reasonably high. The study also concludes that the quality of forecasts improves as the forecast horizon shortens. Accuracy of analyst forecasts of bank earnings plays an important role in assessing the future performance of bank stocks. If forecasts of earnings per share (EPS) of bank stocks are inaccurate, investors and shareholders will be misled regarding the true value of the stock. What is needed is a framework which will minimize mistakes made by forecasters. While this necessity has been recognized by the practicing analysts, the record shows that forecast reliability has generally fallen short of what is demanded. When analysts' forecasts of earnings are consistently off the mark, the credibility of the analyst and his/her firm becomes questionable. To the extent that forecasts of EPS are wrong, the firms will lose business. Competition among analysts can ensure some degree of accuracy. Ackert and Hunter (1993) conclude that while overly optimistic longer term forecasts may be made for the purpose of generating trades, short-term forecasts must be accurate for analysts to maintain credibility. We have examined the 6 and 12 month ahead EPS consensus forecasts of nine money center banks generated by the I/B/E/S using the Theil decomposition method. Generally, we find that our sample behaves well. That is, in each case, whether 6 month ahead or 12 month ahead forecasts are examined, we find that the two systematic components (mean bias and regression bias) of the forecast errors are smaller. The random component is the largest of the three components. Additionally, we find that in eight of the nine cases the random component falls as the forecasting horizon shortens from 12 to 6 months. NINE MONEY CENTER BANKS' EPS FORECAST ERRORS We here apply the Theil decomposition method to the EPS forecast errors of nine money center banks for the 1976-1992 period. Forecasts used were developed by I/B/E/S (Institutional Brokers Estimate System). Our focus is on money center banks because they are widely held and actively traded on the NYSE. The I/B/E/S EPS forecasts are updated every month. Each forecast is the average of forecasts provided by different analysts. Zarnowitz (1984) found that, "the probabilities are high that on average the consensus (forecast) will be less wrong than one person's forecast, and almost certainly less wrong than ARIMA projections and other mechanical extrapolations of the recent past." Likewise, Makridakis and Winkler (1983) found evidence that combining forecasts leads to increased accuracy relative to individual forecasts. Thus, consensus (average) forecasts, like those generated in our sample, are more accurate than forecasts from any single source. We analyze the accuracy of I/B/E/S's EPS forecasts for the 6 and 12 month ahead forecasts. Since short term forecasts are generally more accurate, we expect the six month ahead forecasts to be more accurate, as measured by the Theil decomposition, than the twelve month ahead forecasts. …