Author
Syed Ateeb Akhter SHAH*, Humaira KIRAN** and Sumbal QURESHI***
Abstract
This study forecasts imports and exports of Pakistan at a disaggregated level. Both in and outof-
sample forecasts are produced using conventional econometric time-series models and the
Artificial Neural Network (ANN), a machine learning approach. The forecast performance is
reported using the Root Mean Squared Error (RMSE). Given improved forecasts by the ‘iterative
optimization’ nature of the long and short-term method, ANN outperforms other model
in-sample. For the out-of-sample period, the autoregressive (AR) and ANN model outperforms
other models for import groups, while all univariate approaches outperform each other for
two out of six subgroups in out-of-sample forecasts. Hence, performing equivalently well for
export groups.
Keywords: Pakistan, Imports, Exports, Forecast, ANN.
