“没有免费午餐定理”与农业政策研究中的算法选择:以机器学习预测生猪价格变动为例

“No Free Lunch Theorem” and Algorithm Selection in Policy Research:Predicting Hog Price with Machine Learning

农产品价格政策是农业政策的核心之一,而准确预测价格变动又是价格政策制定的根本。人工智能的突破给农产品价格分析和预测提供了新的强大的工具,而如何选择相对较优的算法来分析和预测价格就成为了一个研究课题。本研究比较运用了四个机器学习模型:传统的ARIMA模型、卷积神经网络(CNN)、循环神经网络(RNN),以及长短期记忆网络(LSTM),来分析和预测中国生猪价格的变化。本文发现传统的ARIMA模型和LSTM模型的性能不相上下,这两者都远远优于CNN和RNN;而从价格的极端变化角度出发,LSTM模型较优越于ARIMA模型。这符合机器学习中的“没有免费午餐定理”(No Free Lunch Theorem):对于所有可能的研究问题,没有一个算法绝对优于其他算法。这要求研究者在政策研究中采用多种算法从而找出相对较好的算法。另外,“群体稳定性指数(PSI)”显示中国猪肉价格在2018年前后发生了显著的结构变化,由此,本文认为在数据结构发生变化的时候,要不断更新模型,与时俱进。

Agricultural product pricing policy is a key element of agricultural policy, where accurate prediction of price fluctuations is the crucial foundation for policy formulation.Breakthroughs in Artificial Intelligence(AI)have provided new and powerful tools for agricultural price analysis and prediction, and how to choose appropriate model algorithms to analyze and predict prices has become a research topic.This study employs and compares four machine learning models: the traditional Autoregressive Integrated Moving Average(ARIMA)model, Convolution Neural Network(CNN),Recurrent Neural Network(RNN),and Long Short-Term Memory(LSTM)model, for analyzing and forecasting the changes in China's pork prices.We find that the performance of the traditional ARIMA model does not show a significant difference from that of the LSTM model, but both significantly outperform the CNN and RNN models.Considering extreme price variations, the LSTM model may slightly outperform the ARIMA model.This result supports, to some extent, the “No Free Lunch Theorem” in machine learning, which suggests that no single algorithm is superior for all types of research problems.Therefore, it is imperative to employ a variety of algorithms in empirical research to find the most effective one.Furthermore, the “Population Stability Index(PSI)” indicates significant structure change in China's pork prices around 2018,necessitating continual updates to the models in response to changes in data structures.