A Comparative Study on Gold Future Forecasting Based on Time Series Large Models and Classic Models

Abstract

In the realm of economic research, the application of large models has become increasingly prevalent. Given the substantial reliance on time series data in finance-related studies, it is essential to evaluate the suitability of large time series models by comparing them with traditional models. This comparative approach is crucial for validating the efficacy and relevance of these models in the context of financial data analysis. In this paper, the closing price of COMEX gold futures from January 1, 2008 to March 8, 2024 is selected for empirical research. The price of COMEX gold futures is predicted vertically and compared with the results of the forecast data after further introducing the dollar index, crude oil prices, and the stock market as exogenous variables. At the same time, the traditional time series model and the machine learning model are compared in terms of model quality and efficiency. It was found that the Time-GPT model performed better in prediction accuracy after introducing exogenous variables. In terms of horizontal comparison, the quality results of the model are generally better than those of the traditional results. The Time-GPT model is strong in capturing long-term dependencies, but the large number of parameters makes the model run inefficiently.