Forecasting KRWUSD Exchange Rate Changes Using a Reduced Relative Nelson-Siegel Model with Deep Learning Techniques 축약된 Relative Nelson-Siegel 모형과 딥러닝 기법을 이용한 원달러 환율의 변화율 예측원문보기
This study applies Chen and Tsang's (2013) Relative Nelson-Siegel (RNS) model, which is based on the differences between the interest rate factors (level, slope, and curvature) of two countries, to forecast the KRWUSD exchange rate and shows that a reduced RNS model with only the curvature or the cu...
This study applies Chen and Tsang's (2013) Relative Nelson-Siegel (RNS) model, which is based on the differences between the interest rate factors (level, slope, and curvature) of two countries, to forecast the KRWUSD exchange rate and shows that a reduced RNS model with only the curvature or the curvature and level factors can avoid overfitting. Additionally, the study investigates the extent to which deep learning techniques, which consider nonlinearities, can enhance the forecasting performance. The empirical analysis uses monthly average data from January 2010 to December 2022 to forecast the 1-, 3-, 6-, and 12-month exchange rate changes. The main findings are threefold. First, the RNS model tends to suffer from the overfitting, while the reduced RNS model, including a relative curvature factor that reflects cross-country differences in monetary policy stance and unexpected changes in future short-term interest rates, demonstrates a relatively lower RMSE. Second, although no statistically significant model differences were found for the one-month forecast, the forecasting performance for the remaining forecast horizons was improved by deep learning techniques. Therefore, it is necessary to pay attention to cross-country differences in curvature factors when predicting KRWUSD exchange rate changes in the medium to long term and to use deep learning techniques to further improve the forecasting performance.
This study applies Chen and Tsang's (2013) Relative Nelson-Siegel (RNS) model, which is based on the differences between the interest rate factors (level, slope, and curvature) of two countries, to forecast the KRWUSD exchange rate and shows that a reduced RNS model with only the curvature or the curvature and level factors can avoid overfitting. Additionally, the study investigates the extent to which deep learning techniques, which consider nonlinearities, can enhance the forecasting performance. The empirical analysis uses monthly average data from January 2010 to December 2022 to forecast the 1-, 3-, 6-, and 12-month exchange rate changes. The main findings are threefold. First, the RNS model tends to suffer from the overfitting, while the reduced RNS model, including a relative curvature factor that reflects cross-country differences in monetary policy stance and unexpected changes in future short-term interest rates, demonstrates a relatively lower RMSE. Second, although no statistically significant model differences were found for the one-month forecast, the forecasting performance for the remaining forecast horizons was improved by deep learning techniques. Therefore, it is necessary to pay attention to cross-country differences in curvature factors when predicting KRWUSD exchange rate changes in the medium to long term and to use deep learning techniques to further improve the forecasting performance.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.