Pattern recognition and subjective belief learning in a repeated constant-sum game

► Action choices and elicited beliefs in a repeated constant-sum game are analyzed. ► The weighted fictitious play belief model is extended to detect two-period patterns. ► Pattern recognition is discovered in both elicited beliefs and action choice data. ► The use of pattern recognition is increasing in the degree of exploitable patterns. ► Implicit learning of patterns, not measurable by belief elicitation, is significant.