An Experimental Study of Overconfidence in Accounting Numbers Predictions

This paper analyzes experimentally investors' overconfidence when making predictions of financial and accounting numbers and explores which factors drive this bias. In particular we analyze the extent to which familiarity with the variable predicted, the complexity of the forecasting task, and the amount of information available for the investors affect their overconfidence. We also compare the extent to which professional analysts differ from other investors in that respect. For this we conducted three experiments. Two experiments with advanced accounting students as subjects, where the experiments differed in the firm the subjects analyzed and the third with professional financial analysts. In each experiment we provided the subjects with past accounting reports and other financial data of a firm. Based on these data the subjects were asked to forecast Net Income, EPS, and Share Price. In all the experiments we found that the subjects exhibited a considerable degree of overconfidence. The professional analysts were slightly more overconfident than the students. Subjects showed more overconfidence in predicting share prices than when forecasting other lesser familiar variables. However we could not detect correlation between the amount of information and overconfidence, neither between success and overconfidence, nor between the complexity of the task and overconfidence.