关键词: 汇率预测, 结构性突变, 经济基本面, 自适应可变窗算法

Abstract: To develop an outstanding Renminbi exchange rate forecasting model in the background of  Renminbi internationalization and the increased two-way exchange rate volatility becomes much more important. The forecasting ability of a parametric model depends not only on whether it is correctly specified but also on the efficiency of  detecting the structure changes to utilize the effective observations to estimate the parameters. This paper has developed the Multivariable Adaptive Rolling Window Algorithm (MARWA) which can not only detect the structure changes automatically at every time point but can also detect the parameter homogenous subintervals and identify the longest homogeneous subintervals which are used as the best subinterval to estimate the parameters. In the out-of-sample forecasting for the exchange rates of the USD, EUR, GBP, JPY against Renminbi, MARWA can manage to capture the parameter time-varying properties and significantly outperform the Random Walk, Purchasing Power Parity Model, Flexible Price Monetary Model, Interest Rate Parity Model, Taylor Rule Model, Taylor Rule Differential Model, Optimal Window Selection Algorithm and the Optimal Auto Regression Model in the middle and the long run (3 to 24 months ahead) forecasting. In the USD against RMB exchange rate regime, MARWA can manage to reduce the MAE forecasting error by 25% to 50% compared with the second-best model. The MARWA can also predicate the direction changes of RMB better than the other models. In the direction prediction, the MARWA can reduce the forecasting error by 16% to 40%. According to the fundamental dynamic changes around the break points, after the “811” exchange rate reform, the economic fundamentals will greatly improve its significance in determining the exchange rate expectations and the market risk preference has also changed. After the "811" exchange rate reform, the exchange rates become more sensitive to the exogenous shocks. The managed floating exchange rate system combined with policies which can support the interest rate marketization, the low inflation rates and the domestic income booming is crucial to the exchange rate marketization and prevention of the exchange rate risk.

Key words: Exchange Rate Forecasting, Structure Breaks, Economic Fundamentals, Multivariable Adaptive Rolling Window Algorithm