Forecasting Turkish Lira Via Combination of ANN and Econometrics Approaches -An Emerging Market Case Study

This study aims to forecast the Turkish Lira to US Dollar exchange rate from January 2011 to December 2024, withholding the 2024 data for forecasting. The study utilizes four econometric models (ARIMA, Naïve, exponential smoothing, and NARDL) along with one Artificial Neural Network model (ANN). It heeds the recommendation of Poon and Granger (2003) to combine these models and assess their predictive accuracy using two methods: equal-weights and variance-covariance. The study finds that the combined model outperforms individual models in predicting the exchange rate. The fusion of ANN and NARDL emerges as particularly effective in forecasting the Turkish Lira's performance. This highlights the significance of macro-economic fundamentals and asymmetric lag values in enhancing forecasting efficiency, surpassing other individual and combined models. Notably, the combined ANN-NARDL model achieves the lowest MAPE value, at 0.014, underscoring its robustness. The research outcomes are of practical importance to various stakeholders, including policymakers, economists, academics, traders, and more. They offer valuable insights for making informed investment decisions and managing exposure to exchange rate risk. The research helps to comprehend fluctuations in exchange rates and the utility of combining different models for more accurate predictions.