GPT Models or Econometric Models: A Comparative Analysis of Gold Price Determinants

The purpose of the article. The main objective of this article is to compare the results of data analysis regarding gold prices and their determinants using two approaches: a classical econometric model and Microsoft Copilot, which integrates advanced artificial intelligence technologies, including the GPT-4 language model (Generative Pre-trained Transformer 4). The secondary objective is to identify, based on the existing literature, the main factors influencing fluctuations in gold prices. These include: the price of crude oil, the USD/EUR exchange rate, the S&P 500 index, and the Consumer Price Index (CPI) in the United States. Methodology. The empirical study involves determining the descriptive statistics of the analyzed variables, the correlation matrix, and estimating the structural parameters of the model explaining the gold price. Results of the research. The best results were obtained for the logarithmic returns of the analyzed variables. In line with the stated hypotheses, there is a negative relationship between the gold price and changes in the S&P 500 index, a negative relationship between the gold price and changes in the US$/EUR exchange rate, and a positive relationship between the gold price and the CPI. The study shows that, during the analyzed period (02.2004–11.2023), changes in crude oil prices did not have a statistically significant impact on gold price changes. To obtain data analysis results using Microsoft Copilot, a "chat" session was conducted. The responses provided the following information: proposed determinants of gold prices, a list of scientific articles, and R code to perform the auto.arima procedure. A comparison was made between the model incorporating economic theory-based factors and the model from the auto.arima procedure suggested by Microsoft Copilot. Based on the conducted study, it can be concluded that the model incorporating both autoregressive factors and other gold price determinants better explains the analyzed variable.