An Empirical Study Of RMB Exchange Rate Forecast Based On ARCH Model And XGBoost Model

With the development of China’s economy,China’s status in the world economy continues to improve,and the Chinese economy is increasingly important to the development of the world economy.Under such circumstances,China has proposed a strategy for the internationalization of RMB in order to enhance RMB’s status in world trade and break the obstacle of the dollar’s hegemony to world trade.In order to cope with the strategy of RMB internationalization,Chinese government departments have continuously strengthened the process of RMB exchange rate reform and accelerated the pace of RMB’s integration with world currencies.Under such circumstances,the fluctuation of RMB exchange rate has attracted more and more attention from the world,especially when it involves cross-border trade and investment.This paper uses an empirical analysis method to study exchange rate forecasts.The data used is from the public database published on the official website of the China Foreign Exchange Trading Center.A total of 974 sample data from March 1,2017 to March 1,2021 are selected for research.First,perform simple descriptive statistics on the daily granularity data of the RMB exchange rate against the US dollar,and perform stationarity test analysis on the original series.It is concluded that the daily data of RMB exchange rate against the US dollar has a certain degree of volatility,and the time sequence diagram shows a basically stable trend after the first-order difference.Further ADF test analysis shows that the data after the first-order difference is stable.At the same time,the correlation coefficient and the partial correlation coefficient are determined by establishing the autocorrelation graph and the partial autocorrelation graph,and based on the first-order difference The data is researched and predicted,the ARMA(1,0)model is established,and the residual sequence of the model is tested for white noise and stationarity.The analysis shows that the residual sequence of ARMA(1,0)is a non-white noise sequence and is stable Then based on the ARMA(1,0)model for ARCH model modeling,the p value of the ARCH effect test result of the residual error of the data obtained through the ARCH effect test analysis is much less than 0.01,that is,the data is significant at the level of reliability of 1%.Therefore,the data has an ARCH effect,and the ARCH model is further established.The residuals basically fluctuate around 0,and the model fitting effect is relatively good.At the same time,the XGBoost model in machine learning is used to predict the daily data of the RMB exchange rate against the US dollar.Through the visualization of the real value and the predicted value,it is preliminarily found that the XGBoost basic model prediction has a certain error.After adjusting the hyperparameters to optimize the model,the predicted value is relatively fit the real value.The MAE of the two models is subsequently calculated and the coefficients of the combined model are obtained by the reciprocal variance method.The predicted value calculated by the combined model is relatively close to the true value than the two single models.The MAE is lower than the other two models,so the prediction effect of the combined model is better than that of the single modelIn general,this paper uses the ARCH model,XGBoost model and ARCH-XGBoost combined model to predict RMB exchange rate.The combined model has a relatively better forecasting effect than the single model,so it can be used in the study of exchange rate forecasting.