Research on RMB Exchange Rate Prediction Based on LSTM Model
Abstract
Accurate exchange rate analysis and forecasting has a profound impact on the country's formulation of corresponding fiscal policies, and can also help enterprises to properly avoid financial risks and individuals to make reasonable financial investments. However, exchange rates are characterized by high volatility, nonlinearity and high noise, making it difficult to forecast exchange rates accurately and stably. Due to the long memory nature of exchange rate price series, traditional forecasting models cannot predict the effectiveness of exchange rates in the market. lSTM, i.e. long and short term memory network, is particularly suitable for handling and forecasting time series with relatively long intervals and delays. In this context, the exchange rate forecasting is based on the long short-term memory deep learning network model LSTM. In this paper, the trend of RMB-USD exchange rate data for the past ten years from February 2011 to November 2021 is selected for analysis and prediction. The prediction results show that the model has good prediction accuracy for the RMB exchange rate test data. the LSTM neural network can effectively control the fluctuation of RMB exchange rate model prediction error and has strong stability.
Keywords
exchange rate prediction; deep learning; LSTM neural network
1 Introduction
The foreign exchange market is a multivariate nonlinear system, and the factors affecting foreign exchange rates are numerous and complex. The tools that people use in the process of analyzing and forecasting foreign exchange rates have undergone a series of changes. At the beginning, people tried to find patterns in the foreign exchange rates through linear models, but the research process did not find many valuable conclusions. Later, people noticed the non-linear characteristics of foreign exchange rates and tried to predict foreign exchange rates through some time series models, but due to the complexity of foreign exchange rate changes and fluctuations, the general time series models are not very good at fitting and predicting the results. Later, with further research, people realized that the past for the exchange rate time series forecasting are only single model forecasting method, this method only focus on the time series local effective information, cannot ensure that at all points in time have a relatively high accuracy . In order to avoid the drawbacks of this method of single model, people proposed to combine multiple models to predict time series, using different models to predict different components of time series respectively, and then combine these models according to reasonable weights to predict time series. Later, people noticed the advantages of neural network models in dealing with systems with nonlinear characteristics, so some scholars began to apply neural networks to the field of foreign exchange and other similar fields with nonlinear characteristics.
With the continuous development of human history, the phenomenon of economic integration among countries in the world has become more and more remarkable, and the level of integration has been deepening. After the 1970s, the world economic integration has risen to a new level: financial globalization. As a new component of economic globalization, financial globalization makes the financial activities of countries or regions in the world more and more integrated, the financial activities between countries are more and more closely related, and the money market and financial markets between countries are also increasingly closely linked. And changes in the foreign exchange market and foreign exchange rates can not only affect the global financial environment, but also deeply affect the development of many areas of the economies of the countries concerned.
Since the world's main exchange rate system was changed from a fixed exchange rate system based on the Bretton Woods system to a floating exchange rate system in 1973, changes in foreign exchange rates have become more frequent and their impact factors more complex. As the changes in foreign exchange rates can significantly affect the financial policies formulated by the government and the hedging of foreign exchange rates