Information content of survey data: applications of entropy and dissimilarity measures

Abstract

This paper evaluates information content of survey data by the means of entropy and dissimilarity measures. Similarities between a priori information (expectations) and a posteriori information (realizations) are assessed for four variables originating from Polish business tendency survey and analyzed across two ownership sectors and four size categories. The measures employed include Shannon empirical entropy, Kullback-Leibler relative entropy, and Chomatowski-Soko³owski dissimilarity coefficient. Results of empirical analysis allow to conclude that enterprise size does not significantly affect entropy values even though some size effects are observed in case of production and general business conditions variables. Public and private enterprises are not differentiated by interpretation of expectations horizon and therefore can be studied on the aggregated level without loss of important information on forecast horizon patterns. Additionally, previous findings (see E. Tomczyk, 2011) are generally confirmed on longer sample. Production time series are found to be characterized by the highest entropy, and prices data – the lowest; entropy of production is also found to be the least variable. In public enterprises, concentration of answers to the survey questions is higher and also more variable than in private sector.