A Study Of RMB Exchange Rate Forecast Based On Taylor‘s Rule And Gradient Boosting Regression

Since the foreign exchange regime reform in 2005,the RMB exchange rate regime has been reforming towards foreign exchange marketization.As the RMB exchange rate forming system changes over time,the volatility of the RMB exchange rate is more frequent and significant.How to effectively forecast the RMB exchange rate becomes an important research subject both in the academia and in the industry.Currently,there are two types of researches on the RMB exchange rate forecast: the first ones focus on fundamental analysis,construct relations between the exchange rate and the macroeconomic variables,and forecast the exchange rate through econometric techniques;the second ones believe that there are no structural relations between the exchange rate and macroeconomic variables.Based on the information provided by the exchange rate itself,these researches forecast exchange rate by parameter and non-parameter methods.This paper combines fundamental analysis and non-parameter method,deducts two exchange rate models based on Taylor‘s rule--homogeneous finite difference Taylor‘s model(FDT-hom)and heterogeneous finite difference Taylor‘s model(FDT-het),and uses gradient boosting regression to forecast exchange rates based on both models.The samples of this paper are exchange rate data of RMB against USD,EUR,JPY and GBP after the exchange rate regime reform.Empirical test shows that: through gradient boosting regression,both models produce higher average forecast accuracies than that of the random walk model in different forecast step sizes.The FDT-hom model can produce higher average forecast accuracies than that of the FDT-het model under the same forecast step size.As the forecast step size increases,the average forecast accuracies of models increase first but then decline.The highest forecast accuracies of the RMB exchange rate can be achieved when the forecast step size is 12 periods.To compare this paper‘s method with other exchange rate forecast methods,this paper conducts a series of comparative experiments,and the results show that: Taylor‘s models can produce higher forecast accuracies than other fundamental models under the same regression method;the gradient boosting regression can achieve higher forecast accuracies than other regression methods under the same theoretical model.To analyze the ‘predictability' of the RMB exchange rate,this paper conducts Diebold-Mariano-West test and Theil‘s U test.The test results show that: through gradient boosting regression,two Taylor‘s models both have significant better overall forecast capacities than that of the random walk model;FDT-hom model has better overall forecast capacity than that of the na?ve model.The tests prove the RMB exchange rate is ‘predictable'.