| 讲座简介： | Exchange rate forecast is of great importance to economic agents, including households, businesses, and policy makers. However, predicting exchange rates is a challenging task. The Meese-Rogoff puzzle claims that economic models of exchange rates seem unable to outperform a simple Random Walk model in out-of-sample forecasting. This paper uses a novel econometric micro approach to forecast exchange rate based on a large panel of individual stock prices, which is expected contain valuable information about future inflation according to economic theory. Our results demonstrate that this micro forecasting method significantly enhances the accuracy of exchange rate forecasting when compared to macro-econometric models such as Random Walk and AR time series models. By employing machine learning algorithms, we can effectively aggregate the information contained in individual stock prices, which leads to more precise exchange forecasts. Moreover, the improvement achieved through the micro forecasting method becomes increasingly prominent as the forecast horizon increases. |
