Time Series ARIMA model and LSTM model forecasting on Gold price comparison

Abstract

Gold price forecasting is a crucial task for investors and financial institutions as gold is considered a safe haven asset and an inflation hedge. This study compares the accuracy of two popular time series forecasting models, namely the autoregressive integrated moving average (ARIMA) model and the long short-term memory (LSTM) model, in predicting daily gold prices from 2000 to 2023. The ARIMA model is a traditional approach that relies on past values to forecast future values, while the LSTM model is a deep learning technique that captures long-term dependencies in time series data. The performance of both models was evaluated using standard metrics such as ME, RSME, MAE, MPE, MAPE, and MASE to assess their accuracy, precision, and goodness of fit. The results suggest that the LSTM model outperforms the ARIMA model in terms of forecasting accuracy, with lower values of ME, RSME, MAE, MPE, MAPE, and MASE. These findings highlight the potential benefits of deep learning techniques in capturing complex patterns in gold prices and improving the accuracy of forecasting models. The implications of this study are relevant for investors, financial analysts, and policy-makers who rely on gold price forecasts to make investment decisions, assess market risks, and monitor macroeconomic indicators. By comparing the performance of two popular forecasting models, this study contributes to the literature on time series analysis and provides insights into the effectiveness of different methods for gold price forecasting.