Abstract:
This study forecasts the quarterly GDP growth rates for Pakistan using leading macro-economic indicators covering a period from January 2002 to December 2020. We use Mixed Data Sampling (MIDAS) regressions, which directly links the quarterly observations with the highfrequency monthly indicators without any aggregation techniques and compare the forecasting performance of MIDAS with the conventional Autoregressive Distributed Lag (ARDL) model. The main purpose of this study is to test the power of leading indicators in providing early estimates of quarterly GDP growth rates. The forecast horizon covers all the 4 quarters of the annual year 2020. The forecast evaluation criteria to compare the forecast of these models are RMSE, MAE, MAPE and Theil Inequality Coefficient. Diebold Mariano Test is also conducted to statistically check the forecasting accuracy of two models. Our results show that, MIDAS performs better than the benchmark model, the ARDL. Among the MIDAS variants considered in this study, U-MIDAS turns out to be the best option for forecasting.
