Adaptive Sg-LASSO-(U)MIDAS Model And Its Application On Mixed Frequency Forecast Of China’s Economic Growth

GDP is the macro variable that can best describe the operation of the macro economy.Forecasting GDP can provide consumers,investors and governments with the information needed to effectively allocate resources.Therefore,forecasting GDP has become the focus of domestic and foreign scholars.The report of the 20 th National Congress of the Communist Party of China pointed out that the overall goal of China’s development is to significantly increase its economic strength and raise the per capita GDP to a new level.At the same time,the epidemic of the 21 st century has a far-reaching impact,the trend of anti-globalization has risen,unilateralism and protectionism have risen significantly,the world economic recovery is weak,local conflicts and turbulence have occurred frequently,global problems have intensified,and the world has entered a new period of turbulence and change.How to make the macro economy run smoothly under this great change is the most concerned issue of the government.This paper analyzes and forecasts GDP based on macroeconomic,financial markets and economic news,which has important practical significance.On the basis of sparse group penalty(sparse group LASSO)and unconstrained mixed data sampling model(R-MIDAS),considering the importance of different variables and the lag structure of high-frequency variables,this paper establishes an adaptive sg-LASSO-MIDAS model,and uses Monte Carlo analysis to demonstrate that the model has good variable selection ability and prediction performance.At present,the research of frequency mixing model is often based on applying polynomial constraints on the lag term of high-frequency variables to reduce the estimated parameters and increase the degree of freedom.In this paper,the mixed frequency data is used to enhance the accuracy and timeliness of the model prediction.While the polynomial constraints of high-frequency variables are considered to be relaxed,so as to realize the direct selection of high-frequency lag terms,and use the high-frequency data itself rather than the low-frequency information subject to polynomial constraints.It further increases the accuracy of model prediction.At the same time,it introduces the attention information of the news of the Economic Daily constructed by using the LDA model,and compares the effects of various regularized mixed frequency models and parameter-driven mixed data models on GDP prediction.The empirical results show that the model constructed in this paper has achieved the best prediction effect in terms of prediction accuracy.The empirical conclusions of this paper can be summarized as follows:First,the adaptive SGL method can realize the adaptive two-level selection of different high-frequency variables and high-frequency variable lag terms,and make full use of the structural information of the mixed data,which can significantly increase the fitting results of the insample data and the prediction results of the outsample data.Second,China’s economic news can provide rich and timely information for China’s macroeconomic forecasting and improve the forecasting accuracy of the model.In summary,the adaptive sg-LASSO-MIDAS model has many advantages in predicting China’s GDP.This paper studies the impact of macroeconomic and financial markets on GDP using mixed-frequency data,and forecasts the growth rate of China’s GDP by structuring news focus data based on text information,which improves the accuracy and timeliness of the prediction,and provides guidance and suggestions for the government to predict and control the growth rate of GDP and stabilize macroeconomic operation.