机器学习型货币政策模式与传统规则型模式比较
Comparison between Machine Learning Monetary Policy Model and Traditional Rule-Based Model

随着机器学习算法的广泛应用，机器学习型货币政策模式相较于传统规则型模式是否更具优势是一个重要的研究课题。本文对比了泰勒规则模式与机器学习模式，首先预测出利率的泰勒规则值和机器学习值；其次使用多种指标评价利率预测的精确度；最后构建多元回归模型，来评价货币政策模式。结果表明，相对于泰勒规则模型，机器学习算法所预测的利率在多种不同的评价指标下都拥有更小的误差，能够更加精准地预测利率，并且机器学习算法预测的利率所对应的关键经济变量即GDP和CPI与实际值的差异也更小，能够更好地应用到实际中，因此机器学习型货币政策模式更有优势。
With the wide application of machine learning algorithms, the question of whether the machine learning model of monetary policy is more advantageous than the traditional rule-based model is an important research topic. This paper compares the Taylor rule model and the machine learning model, firstly, predicts the Taylor rule value and the machine learning value of the interest rate; secondly, evaluates the accuracy of the interest rate prediction by using a variety of indicators; finally, constructs a multiple regression model to evaluate the monetary policy model. The results show that, compared with the Taylor rule model, the machine learning algorithm predicts interest rates with smaller errors under a variety of different evaluation indexes, which can predict interest rates more accurately, and the key economic variables corresponding to the interest rates predicted by the machine learning algorithm, i.e. GDP and CPI, have smaller differences from the actual values, which can be better applied to the real world, and therefore, the machine learning monetary policy model is better.