A Security Analyst's Action Recommendations: An Application of Recursive Partitioning to Modelling Judgement

This paper illustrates an approach to modelling judgement. Mathematically, derived sequential decision tree models of judgement are constructed using the Classification And Regression Tree (CART) algorithm developed by Breiman et al. (1984). This technique was selected because it attempts to combine the predictive power of regression based models with the descriptive appeal of decision tree models based on verbal protocols.CART was used in the context of a study which attempted to model the decision processes used by the security analysts of a managed fund to make action recommendations. The CART model revealed that the decision processes of security analysts were complicated and that infor Mation was processed configurally. In general, the CART model achieved low error rates and had a predictive power that was at least equivalent if not better than multinomial LOGIT.