OUT-OF-SAMPLE FORECASTING PERFORMANCE OF A ROBUST NEURAL EXCHANGE RATE MODEL OF RON/USD

This paper aims to explore the forecasting accuracy of RON/USD exchange rate structural models with monetary fundamentals. I used robust regression approach for constructing robust neural models less sensitive to contamination with outliers and I studied its predictability on 1 to 6-month horizon against nonrobust linear and nonlinear regressions and, especially, random walk. The results show that robust model with low breakdown point improve the forecast accuracy of RW and AR models on 1- and 4-month horizon and performs better than RW at all time horizons.