A Study of forecasting of Exchange Rates Using Non Robust and Robust Estimators

Presence of outliers in exchange rates data is a common feature. In the present study we have tried to construct forecasting models for two exchange rates, that are less sensitive to data contamination by outliers through the Robust estimation techniques namely Least Median Squares (LMS) and Least Trimmed Squares (LTS). The built models are used to assess the predictability of two exchange rates at 1-, 3- and 6- month horizons. The predictive ability of the Robust Linear Autoregressive (RAR) models as compared to that of the Random Walk (RW) and Least Squares (LS) fitted linear autoregressive (AR) models are assessed in terms of forecast accuracies. Further using Diebold-Mariano test the equivalence of forecasts accuracy of two competing models are examined. Using the same criterion the RAR models are also compared. A study on Forecasting models for exchange rates is carried out by Preminger, A and Franck, R, (2007) using RW model and linear AR models fitted by the LS and S- methods of estimation. In the present study we observed that, in general, the performances of robust estimation techniques are better than the LS estimation technique and the overall performance of LTS is better than the LMS and S-estimation techniques.