Research On Short-Term Forecast Of Hog Price Based On Combination Forecasting Strategy

Our country is a big country in pig production and consumption,which is important to our country’s national economy.In recent years,the abnormal fluctuation of hog prices has seriously affected the healthy development of our country’s animal husbandry and the daily life of the people.The judgment of the future trend of hog price provides important reference information for the practitioners of the hog industry and relevant authorities to make decisions.Therefore,it is of great practical significance to make a scientific and accurate prediction of the price of hog in our country.This paper establishes machine learning prediction models for hog prices from the causal relationship model and the time series model respectively,so as to avoid the dependence on a single model and improve the scientificity of the prediction.In terms of the causal relationship prediction models,this paper selects 14 variables from the aspects of pig supply and demand,epidemic diseases and our country’s macroeconomic environment,and then constructs ridge regression model,Bayesian generalized linear model,extreme learning machine,partial least squares regression model,neural network model,and support vector regression model.In time series forecasting models,this paper selects input data for different step lengths of multi-step prediction,and builds naive prediction model,random walk prediction model,Holt-Winters with additive seasonality model,Holt-Winters with multiplicative seasonality model,exponential smoothing state space model and autoregressive integrated moving average model.From the perspective of combined forecasting,five combined forecasting strategies are used to combine the predicted values of the above machine learning forecasting models to improve the forecasting accuracy.The results show that the prediction models based on machine learning can make an accurate prediction on the price of hog.In the causal relationship model,the neural network model has the highest prediction accuracy,followed by support vector regression,extreme learning machine,Bayesian model,partial least squares and ridge regression model.The prediction accuracy of the combined model is generally higher than that of the single model.Compared with the optimal single model(i.e.,the neural network model),the prediction accuracy of the five combined prediction strategies is improved by 2%-10%.Among them,OLS combined forecasting strategy has the highest forecasting accuracy.Among the time series models,the forecast performance deteriorates with increasing forecast step size,and the exponentially smoothed state space model(ETS)performs the best overall.The combined forecasting strategy has a1%-7% improvement in accuracy relative to the best performing single forecasting model.When the prediction step size is 3,the Robust combination strategy performs the best.When the prediction step size is 6,the CLS combination strategy performs the best.The research work in this paper will help the hog industry management departments and practitioners to grasp the price operation law of our country’s pig market,and then improve the foresight and scientificity of relevant decision-making.