Abstract

This study examines the application of deep learning in forecasting Malta's Foreign Trade monthly statistics using historical time-series data. To do this, we built and tested several models based on recurrent neural networks by aiming to see whether they can increase the accuracy of short‑term economic forecasts.
Our work compares how each model performs and points out where these methods could fit into the analytical tools used by institutions that produce economic and financial statistics. The project examines both the accuracy and efficiency of different model designs and discusses what they could mean for improving the way economic indicators are produced.
Early findings are encouraging. Ongoing work continues to refine the models, evaluate robustness and explore their operational use in real statistical workflows. Overall, this research adds to the wider conversation about applying artificial intelligence in official statistics and central banking, with the goal of supporting more data‑driven decision making.