Research On High-Dimensional Mixed-Frequency Short-Term Forecasting And Its Accuracy Improvement Mechanism Of China’s Economic Growth

China has been contending with a complex and volatile external economic environment due to the endless stream of global trade disputes and turmoil in emerging economies since the global financial crisis in 2008.At the same time,the Chinese economy is experiencing increasing downward pressure caused by insufficient demand,increasing production costs and bottlenecks in scientific and technological innovation.However,the unexpected outbreak of the Corona Virus Disease 2019(COVID-19)in early 2020 has adversely affected the Chinese economy.Therefore,this study aims to provide accurate and timely short-term forecasts of China’s economic growth,thus providing a reliable reference for the reasonable regulation of macroeconomic policy in China.Since accurate predictions depend on reliable modelling techniques and abundant data information,this study has been gradually expanded from two aspects of the methodology and data.The research framework of this paper consists of four distinct parts.Firstly,we develop a real-time nowcasting paradigm of China’s economic growth.Secondly,the internal mechanism of economic forecasting are explored.Thirdly,the economic growth forecasting models based on the complex economic environment are developed.Lastly,high-frequency and high-dimensional web search query data are introduced to generate real-time economic growth forecasts and corresponding forecasting models that can incorporate high-dimensional mixed-frequency data are proposed.The research is presented as follows:First,this study systematically analyzes the economic growth theory,macroeconomic forecasting theory and the factors that influence economic growth,and constructs a large set of macroeconomic variables that include leading,consistent and financial indicators.Secondly,this paper details the release dates of the macroeconomic data set,and develops a pseudo-real-time forecasting research paradigm based on the data-release dates to generate real-time forecasts of China’s economic growth.Factor Mixed-frequency Data Sampling models(FA-MIDAS and FA-U-MIDAS)that can incorporate high-dimensional mixed-frequency data are utilized to test whether the forecast accuracy will be gradually improved with the addition of updated data.Empirical results demonstrate that the constructed pseudo-real-time forecast paradigm can capture more real-time data information,resulting in improved forecasting accuracy for China’s economic growth.In particular,the introduction of new data leads to a gradual improvement in forecasting performance,especially in the areas of transportation industry,investment,and consumption indicators.Moreover,the robustness tests demonstrate that the nowcast accuracy will be improved by adding more monthly updated real-time information,regardless of the length of the out-of-sample nowcast window and data update interval.Secondly,the lack of practical significance in the common factor extracted from the FA-MIDAS-type models makes it challenging to explore the accuracy improvement mechanism of the model.Therefore,this paper introduces a novel Group Penalized Unrestricted MIDAS(GP-U-MIDAS)model for forecasting China’s economic growth.The GP-U-MIDAS model can identify a series of core indicators that affect China’s economic growth,resulting in an improvement in the predictive and explanatory capabilities of the model.The empirical results show that,first,the GP-U-MIDAS model improves the nowcasting and short-term forecasting accuracy of China’s economic growth by achieving the function of group variable selection,parameter estimation and mixed-frequency data analysis simultaneously.Second,industrial-added value is didentified as a key indicator for forecasting China’s economic growth.Third,a few financial indicators have important predictive powers across different forecasting horizons.The negative impact of adverse shocks on the Chinese economy can be mitigated through implementing effective supervision of unforeseen shocks that may arise in the stock and bond markets.Thirdly,the outbreak of the COVID-19 pandemic since early 2020 has had a severe adverse impact on the Chinese economy.In a crisis context,the relationships between economic variables may differ.Therefore,this study develops a Threshold Group Penalized U-MIDAS(T-GP-U-MIDAS)model capable of identifying possible change points in economic operations by introducing a threshold function,thus being applicable to nowcast and forecast economic growth,and identify key indicators that affect China’s economic growth in a complex economic environment.The empirical results show that,the forecast performance of the T-GP-U-MIDAS model is superior to benchmark models,and the forecasting performance can be improved when more updated data become available.The industrial added value,investment,consumption,import and export are identified as core indicators driving economic growth,while a few financial variables play a crucial role in short-term economic growth forecasting.The T-GP-U-MIDAS model selects more variables to ensure forecasting performance for longer forecast horizons.The variables with a threshold effect are mainly concentrated during COVID-19 and are primarily related to investment,consumption,import,export,transportation and industrial enterprises.The variable selection results provide a reliable reference and direction for real-time monitoring of China’s economic growth in a complex economic environment.Finally,accurate forecasting depends on reliable modelling techniques and abundant data information.However,the availability of macroeconomic data has been reduced by time delats in data release and the stagnation of economic activity caused by the COVID-19 outbreak,this paper thus introduces daily updated high-frequency search query data to generate short-term forecasts of China’s economic growth in a complex economic environment.Web search query data are selected from five aspects based on the demand theory: consumption,investment,import and export,government purchases,and employment.However,the T-GP-U-MIDAS model proposed in this study cannot incorporate mixed-frequency data with a big frequency mismatch.To address this limitation,a novel Threshold Group Penalized MIDAS(T-GP-MIDAS)model is developed by introducing a polynomial weight function.Meanwhile,the potential reasons for improving forecasting accuracy are analyzed in-depth.The empirical results show that,the addition of real-time updated daily data provides effective data information for nowcasting and short-term forecasting of China’s economic growth.Compared with benchmark models,the T-GP-MIDAS model can accurately capture the negative impact of COVID-19 on China’s economic growth when the information within the quarter is available and the forecasting accuracy is substantially improved with the increase of available data information.Variable selection results show that consumption-related data have important predictive powers for forecasting China’s economic growth during COVID-19 and the recovery period,followed by search query data related to import and export,government purchases,and employment.However,the predictive power of investment-related data for economic growth is weaker than before COVID-19.These empirical results can provide leading information for the real-time direction and decision-making of China’s macroeconomic policies.In this paper,systematic sorting and innovative expansion for Chinese macroeconomic forecasting research are carried out from two aspects of method and data,which enhances the forecasting and interpretation capabilities of the forecasting model.This study effectively supplements and extends the existing forecasting and nowcasting methods.