Trend Prediction and Empirical Analysis of Pork Futures Prices Based on Machine Learning

Abstract

The live hog futures market has faced abnormal price cycles and severe fluctuations in recent years, making traditional forecasting methods less effective. Accurate price prediction is crucial for market participants to make informed decisions. This study employs the XGBoost model to predict live hog futures price trends. It uses price indices from 1 to 28 days prior as independent variables and closing prices as dependent variables to analyze feature importance. The significant variables identified are then used to develop a prediction model. The study finds that prices from 1 to 3 days prior, 5 days prior, and 22 days prior are significant for prediction. The developed model achieves a Mean Absolute Percentage Error (MAPE) of 0.11% on the training set and 1.71% on the test set. It forecasts a volatile trend for the next 20 days. The study innovatively integrates the findings with traditional trading indicators (e.g., "5-day moving average" and "20-day moving average") and demonstrates their relevance in the live hog futures market. This approach provides a new perspective for predicting price trends in complex and dynamic markets.