Abstract

This paper estimates a DSGE model and three versions of VAR models (VARX, BVARX and BVAR) to analyze forecasting performance of these models in context of Pakistan. VAR models and a medium-scale DSGE model are estimated using quarterly data (1980Q4-2017Q2). Expanding window recursive out-of-sample forecasts for GDP growth, call money rate, CPI inflation and percent change in exchange rate are generated and compared over the period 2009Q1-2017Q2. Forecasting performance is analyzed by the comparison of bias and root mean squared errors (RMSE). Analysis of forecasting performance over 1-8 quarters forecast horizon reveals that BVAR model provides relatively better forecast in case of GDP growth, interest rate and inflation while BVARX provides more accurate forecast in case of exchange rate. In case of GDP growth, inflation and exchange rate, forecasting performance of DSGE model considerably improves as forecasting horizon expands. For longer forecast horizons, divergence between DSGE and Bayesian VAR forecasts tends to disappear. This implies that DSGE model is more relevant for medium term forecasting rather than short term forecasting. Structural interpretation of DSGE forecast errors reveals that there has been unutilized growth potential in economic activity. This slack in economic activity might be attributable to unnecessarily high interest rate and overvalued exchange rate.

Keywords: Forecasting · DSGE models · VAR models · Pakistan.

>> Full Article