China’s Business Cycle Forecasting: A Machine Learning Approach

Business cycle refers to the cyclical phenomenon of economic expansion and economic contraction in a country or region,which is the economic climate of a country.The establishment of an effective business cycle prediction model to predict the future economic development trend of China will help the state to timely introduce corresponding macro-policies to regulate the economy.Traditional prediction methods have poor effect on nonlinear data time series data processing,while machine learning algorithm can handle nonlinear features and various types of data well,and has obvious advantages in building high-precision prediction models.Based on the above background,this paper selects five machine learning models,which are widely used,such as extreme gradient lifting,support vector machine,logistic regression model,decision tree and random forest,to predict China’s business cycle.In the construction of indicators,63 related indicators such as money supply,fiscal revenue and expenditure,manufacturing,import and export,real estate investment,etc.are comprehensively selected as explanatory variables,and the consistency index of macroeconomic prosperity index is selected as the explained variable.In data set processing,because data can’t have both time interval length and index quantity richness,this paper divides independent variables into two data sets according to time interval and index quantity,among which data set 1 contains comprehensive indexes but short time interval,and data set 2 has a long time interval but contains few prediction indexes.In the application of the model,the moving average method is used to predict the data of the next six months with the data of the past 36 months,which effectively eliminates random fluctuations and shows the development trend of events.The sample data is divided into training set,verification set and prediction set.Five kinds of machine learning models are used to train the data in the sample.Through the data self-learning and automatic parameter adjustment,the model is continuously optimized,and then the data outside the sample is predicted.Statistical evaluation indexes are used to evaluate the prediction ability of five machine learning models in two data sets.The evaluation results show that these five machine learning techniques can predict China’s economic cycle more robustly,and the prediction accuracy of support vector machine and logistic regression model is better than that of the other three machine learning models.Moreover,the prediction effects of different machine learning models in the same data set are significantly different,and the same machine learning model will show different accuracy in different data sets.The classification prediction performance of the five models in data set 1is better than that in data set 2.The prediction results show that the five machine learning models are consistent in the prediction of China’s business cycle in the next six months.Except for the prediction of April2022 by support vector machine,the prediction results of other models from January to June2022 are all expansion.Based on the recent three years’ China’s economic fluctuation situation and macroeconomic leading index,the results show that the economic downturn is a small fluctuation in the stage of economic recovery,short-term economic fluctuations will not affect the long-term economic trend,and China’s economy has certain resistance to negative impacts.After revision,it can be seen that with the support and guidance of government policies,the economy has shown a steady and positive trend,and it is now in a long rising cycle.In order to further promote China’s economic forecasting research and promote China’s macro-economic development,based on the research process and results of this paper,some suggestions are put forward on index construction,machine learning model and result test.