A model for predicting Finnish household loan stocks

We propose a new Bayesian VAR model for forecasting household loan stocks in Finland. The model is designed to work as a satellite model of a larger DSGE model for the Finnish economy, the Aino 2.0 model. The forecasts produced with the BVAR model can be conditioned on projections of several macro variables obtained from the Aino 2.0 model. We study several specifications for the set of variables and lags included in the BVAR, and evaluate their out-of-sample forecast accuracy with root mean squared forecasting errors (RMSFEs). We then select a preferred specification that performs best in predicting the loan stocks over forecast horizons ranging from one to twelve quarters ahead. The model adds to the existing toolkit of forecast models currently in use at the Bank of Finland and improves our understanding of household debt trends in Finland.