Forecasting Malaysian Exchange Rate: Do Artificial Neural Networks Work?

Being a small and open economy, the stability and predictability of Malaysian foreign exchange are crucially important. However, despite the general failure of conventional monetary models, foreign exchange misalignments and authority intervention have both caused the forecasting process an uneasy task. The present paper employs the monetary-portfolio balance exchange rate model and its modified version in the analysis. We then compare two Artificial Neural Networks (ANNs) estimation procedures (MLFN and GRNN) with random walk (RW) in the modeling-prediction process of RM/USD during the post-Bretton Wood era (1990M1-2008M8). The out-of-sample forecasting assessment reveals that the ANNs have outperformed the RW, which in particular, the MLFNs outperform GRNNs where as the latter outperform the RW models with consistency in both the exchange rate models by all evaluation criteria. In addition, the findings also show that the modified model has superior forecasting performance than the first model. In brief, economic fundamentals are vital in forecasting and explaining the RM/USD exchange rate. The findings are beneficial in policy making, investment modeling as well as corporate planning.