ABSTRACT
Agents choose from a number of simple forecasting heuristics. The forecasting heuristics are similar to those obtained from estimating linear models on individual forecasting experimental data. Evolutionary selection or performance-based reinforcement learning based upon relative performance disciplines the individual choice of heuristics. Hence, the impact of each of the rules is evolving over time and agents tend to switch to more successful rules. The four forecasting heuristics are as follows:
ADA : pe1;tþ1 ¼ 0:65pt21 þ 0:35pe1;t; ð15Þ
WTR : pe2;tþ1 ¼ pt21 þ 0:4ðpt21 2 pt22Þ; ð16Þ
STR : pe3;tþ1 ¼ pt21 þ 1:3ðpt21 2 pt22Þ; ð17Þ
LAA : pe4;tþ1 ¼ pavt21 þ pt21
2 þ ðpt21 2 pt22Þ; ð18Þ
where pavt21 ¼ ð1=tÞ Pt21
j¼0pj is the sample average of past prices. Adaptive expectations (ADA) predicts that the price is a weighted average of the last observed price pt21 and the last price forecast pet . The trend-following rules extrapolate the last price change, either with a weak (WTR) or with a strong (STR) trend parameter. The fourth rule is an anchor and adjustment rule (Tversky & Kahneman, 1974), extrapolating a price change from a more flexible anchor.
