Using Macro-Financial Variables to Forecast Recessions : An Analysis of Canada, 1957-2002

We employ artificial neural networks using macro-financial variables to predict recessions. We model the relationship between indicator variables and recessions 1 to 10 periods into the future and employ a procedure that penalizes a misclassified recession more than a misclassified non-recession. Our results reveal that among 16 models that we constructed from 9 indicator variables and their combinations, the indicator variables Spread, 3-year bond rates, 10-year bond rates, monetary base, industrial production are candidate variables for predicting recessions ranging 2 to 10 periods in the future. However, most indicator variables become candidate for predicting recessions when misclassified recessions are penalized heavily than misclassified non-recessions