On the Use of Naive Expectations of Earnings per Share as Experimental Benchmarks

Both martingale and submartingale processes have been used as experimental benchmarks for corporate earnings forecasts or expectations in a large number of studies. This paper investigates the relative accuracy of forecasts from these two processes for a sample of Australian'firms. Submartingale models incorporating drift factors estimated over a large number of past observations consistently outperformed those using fewer observations, while the martingale model always outperformed the submartingale with drift factors estimaied over less than six prior earnings changes. The relative superiority of the martingale versus the best submartingale models used depended on whether a quadratic or linear error metric was deemed appropriate.