Predicting the probability of recession in Croatia: Is economic sentiment the missing link?*1

This paper aims to assess the possibility of predicting Croatian recessionary episodes using probit models. The authors first estimate a baseline static model using four leading indicators of recession (monetary base, unemployment, industrial production, and CROBEX stock market index). Lag lengths of up to 6 months are examined for each of the observed variables in the probit specification, and several important conclusions arise from the estimated models. First, the stock market and money supply exhibit the most pronounced leading characteristics in the Croatian economy (a 3-month lag length is selected by the information criteria). Second, the dynamic model (including a lagged dependent dummy variable) significantly outperforms the baseline static model. Third, the authors augment the probit model by the Economic Sentiment Indicator, which significantly contributes to the model accuracy. The latter confirms the main hypothesis of the paper, going in line with the assertion that psychological factors largely govern the economic cycles, growing in significance in times of economic hardship.