| Risk and return coexist in the financial market.As an absolute index to measure the risk,volatility itself is not good or bad.Its existence means both opportunities and challenges.Volatility can affect the pricing of financial derivatives,the management of asset portfolio and the hedging of futures and spot market.So it is particularly important in modern financial market.In most risk management strategies,the effective measurement or prediction of risk plays a connecting role.As an important varieties of commodity market,the prediction of the volatility of gold has attracted more and more attention.On the basis of previous studies,this paper explores the volatility characteristics of China's gold price,innovatively combines GARCH model with artificial intelligence algorithm,introduces artificial neural network model ANN and long-short-term memory model LSTM.At the same time,we also explore two deep learning models LSTM-ANN hybrid model and LSTM-E ensemble model.In order to improve the learning efficiency of the model,this paper also adds the financial data of exchange rate market,stock market and other macro indicators.The purpose of this paper is to explore the characteristics and forecasting effect of gold price volatility in China through different time window dimensions of input data,different model dimensions(five empirical models)and different target dimensions(gold futures and spot).According to the final empirical results,we can see that the machine learning model can achieve better prediction results than GARCH model when forecasting the volatility of gold price in the short and medium term,whether it is gold futures or gold spot.Compared with ANN,the deep learning model has better fitting and prediction effect and better performance,both in training set and test set.Because the hybrid model LSTM-ANN has better fitting and prediction ability than the basic model ANN,but it is not as good as the basic model LSTM,so the promotion rate of the hybrid model relative to the single basic model depends on the situation.The ensemble model has a strong fitting ability,and has a good fitting effect in the training set,but the ensemble model is prone to over fitting.From the perspective of stability,the prediction performance of gold futures and gold spot is still lower than that of LSTM model.When the data input time window is 10 days,the prediction effect of the model will be better,which shows that in China's gold market,the medium-short-term factors to predict the medium-short-term volatility effect will be better.In the selection of single optimal model,the optimal prediction model of gold futures market is the LSTM model with the input time window of 10 days,and the optimal prediction model of gold spot market is the LSTM-E integrated model with the time window of 20 days.But generally speaking,the prediction performance of LSTM model is better,and the prediction effect of 10 days' data in different window periods is better.The main contribution of this paper is to provide a new way to predict the price volatility for the domestic gold market,and enrich the modeling of machine learning model.By analyzing the error between the predicted value and the actual value of each model,it is found that the predicted result of the model can basically achieve the same direction fluctuation with the actual value.In the empirical application,we apply the prediction results of some models to the value-at-risk model of risk management.We find that the accuracy of deep learning model is higher than that of GARCH model and ANN model,whether it is gold futures or gold spot.The accuracy level of gold futures is generally higher than that of gold spot,which is probably because the distribution of the return rate of gold spot price is more peak and thick tail than that of gold futures.So it is more susceptible to the influence of extreme value.The results of this paper hope to provide help for the risk management of gold market price volatility in the future. |
PDF Full Text Request