What Influences Overprecision in Judgmental Forecasting

Previous studies (Krawczyk, 2011; Mannes and Moore, 2013) showed that asymmetric reward functions can be used to get the information on estimated relevant percentiles of the distribution (upper and lower bounds of the confidence intervals, respectively) and thus to analyze the overconfidence level. The estimations obtained by using this indirect method were different than when the participants were directly asked about the value of upper and lower bounds of the relevant confidence intervals. In this article, we consider the problem if these observed differences are permanent and independent of the learning process. In the experiment students provided direct point forecasts and classical lower and upper bounds of the confidence interval and the probability distribution of forecasted weekly rate of returns for WIG and DAX indexes. Based on the reward (loss) functions, indirect estimates of the median (symmetric reward functions) and lower and upper bounds of the confidence interval (asymmetric reward functions) were also derived. There were no significant differences between directly and indirectly provided confidence intervals, implying that the level of overprecision measured by these two methods do not differ if participants are given enough trials to learn the reward function. This suggests studying other than a reward function shape sources of illusion of control. The results have also practical implications, for example for options markets, where the volatility can be estimated directly or indirectly by setting the price of an option (implied volatility).