关键词: 天然气需求, 概率预测, 混频数据, 动态因子, 分位数模型

Abstract: With the rapid growth of natural gas consumption demand, demand volatility has also been significantly intensified, resulting in seasonal and cyclical imbalances in natural gas supply and demand in some regions. In this study, a mixed-probability prediction model for natural gas demand is presented based on the mixed-frequency data sampling, quantile regression model, and kernel density estimation. The comprehensive dynamic factors system is also constructed covering weather conditions, energy market, capital market, and investment interest to predict the natural gas demand in China more accurately. The empirical results demonstrate that the predictive power of weather conditions, energy markets, and investment interest performs better than capital market. For the weather conditions, the daily temperature is the best indicator of monthly natural gas demand. For the energy market, the monthly natural gas demand in China has a significant auto-correlation, lasting 3 to 5 months, while the influence of oil and coal process last for fewer days, i.e. 11 days and 10 days respectively. In terms of forecast performance, most of the actual data on out-of-sample monthly natural gas demand appears near the highest point of the probability density curve. The model proposed in this paper can not only directly use the forward-looking information of the mixed-frequency dynamic factors, but also obtain a smooth natural gas demand probability density curve, which is 14.13%~29.15% higher than the prediction accuracy of the existing model. The research conclusions provide a useful reference for maintaining the natural gas market security in China and improving the “dual carbon” policy design.

Key words: Natural Gas Demand, Probability Forecast, Mixed-Frequency Data, Dynamic Factors, Quantile Model