关键词: 通货膨胀, 机器学习, SHAP值解释性方法, 通胀预期

Abstract: Identifying the influencing factors of inflation can help policymakers proactively take measures to prevent inflation risk. Traditional methods such as SVAR are adopted by a lot of literature, but with limited types of factors and non-linear relationships. To solve the limitation, this paper innovatively employs SHapley Additive exPlanations (SHAP) method and several non-linear machine learning methods like SVAR in analyzing the five rounds of inflation that happened in 2001—2019 in China. We have found three main results. Firstly, inflation expectation and food price rise are the driving factors in almost every round of inflation in the past 20 years. Secondly, the impact of demand factors such as consumption and investment gradually diminish, while the influence of cost-driven factors especially labor cost has considerably increased. Thirdly, monetary policy is always a major factor influencing inflation through various channels. Given these results, we can prevent inflation risks by guiding inflation expectation more effectively, stabilizing the production and supply of foods, creating an enabling business environment to mitigate the rising cost of labor. Moreover, monetary policy should not be constraint by inflation pressure. Under the premise of macro-prudential supervision of financial and real estate markets, the monetary policy should appropriately enhance its support to the real economy to relieve the downward economic pressure in China.

Key words: Inflation, Machine Learning, SHAP Method, Inflation Expectation