China's business cycle forecasting: a machine learning approach

Forecasting the business cycle can help policymakers implement economic policies more effectively. This paper selects 62 macroeconomic and financial indicators and divides them into two data sets to forecast China's business cycle. The data of the past 36 months is used to predict China’s business cycle for the next month by the simple rolling window method. For testing the training set and determining model parameters, five machine learning models are used: XGBoost, SVM, Logistic Regression, Decision Tree, and Random Forest. The statistical evaluation indicators of the confusion matrix show that these five machine learning algorithms can reliably anticipate China's economy cycle, with Logistic Regression outperforming the others. At the same time, the paper compares the model predictions with the actual values and discusses the differences between them.