Research On The Short-term Fluctuation Of China's Economy Based On Asynchronous Mixing Data

The study of economic fluctuations,especially the prediction of short-term economic fluctuations,has always been an important and hot topic in macroeconomics.Studying the economic fluctuations of china started in the 1980 s.In the early years,methods developed by foreign researchers were used to analyze and explore the economic cycle and economic fluctuations in China.According to the inertia of economic system and historical data from statistical survey,regression models were widely used to monitor and forecast economic fluctuations.With the standardization and improved management of statistical surveying in China,the quality of statistical survey data has been significantly improved.However,problems such as strong subjective bias of questionnaire design,high cost and delay in data acquisition,limited sample size,etc.still exist.Though the regression method has been constantly improved with different models proposed,it cannot capture the nonlinearity between the predictors and economic fluctuations.There is still large prediction errors when only historical data is used,and it is difficult to accurately predict the future economic fluctuations.The rapid development of Internet technology,computer science and artificial intelligence has brought new possibilities to the study of macroeconomic fluctuations.The use of large data sets obtained on the Internet and machine learning algorithms in macroeconomics are actively explored by researchers.Improved quality and timely publication of economic statistics by statistical investigation department provide reliable long-term data source for short-term economic fluctuation study,while Internet based big data provides high-frequency and real-time short-term data source.It is thus important to study the acquisition and processing of economic data from the Internet,and to uncover the relationship between the data generated by economic actors and economic fluctuations,as well as the mechanism governing such relationship.Under the background of big data,this study investigated the correlation between Internet search engine data and short-term economic fluctuations and proposed short-term factors of economic fluctuations,by taking advantage of the timeliness of Internet search data.In neural network models,the long-term and short-term factors that affect short-term economic fluctuations,time series data with different frequencies were combined to improve the accuracy.The timeliness and accuracy of short-term forecast of economic fluctuations have important theoretical and practical significance.The contents and methods of this study are summarized as follows.I)According to the theories of economic fluctuation,limited attention and behavioral decision-making,an analysis framework based on asynchronous web search data to predict short-term economic fluctuations of China was constructed.The analysis framework includes three parts: economic fluctuation,behavior decision and the relationship between web search and economic fluctuations.II)Using previous results,the theory of economic fluctuations,and the characteristics of the economic development of China,this study analyzed the relationship between the short-term economic fluctuations and macroeconomic indicators in China.The short-term fluctuations of the economy of China from the three industries,the three major demands and the external impact were investigated to uncovered the long-term factors influencing the short-term fluctuations,and proposed a long-term index.III)After the correlation between economic fluctuation and Internet search was clarified,a short-term attention index was constructed from investment,consumption,production and economic policy,using behavioral decision-making theory,limited attention theory and information asymmetry theory.IV)Prediction models were constructed,using synthesized time series data with different frequencies to combine the long-term and short-term factors.These models include AR,ARDL,MIDAS and AMP-LSTM.V)Finally,the models were used to analyze the short-term fluctuations in China.The performance of each model was evaluated,empirical results were analyzed,followed by the conclusion and future indication.The main results are summarized below.In first chapter,we proposed an analytic framework for short-term economic fluctuations by combining long-term and short-term factors,based on the theories of economic fluctuation,behavioral decision-making,limited attention and information asymmetry.Using both long-term and short-term factors,we systematically analyzed the long term trends that affect the economic fluctuations on basis of traditional statistical survey data.Meanwhile,following the current advancement in of big data analysis,we make use of the real-time,high-frequency and massive amount of Internet data,and included short-term factors based on web search into the analysis.The analytic framework provided a theoretical basis for the study of long-term and short-term factors on economic fluctuations in the future,and guided the building of models based on mixed frequency data.In the second chapter,the long-term factors of short-term economic fluctuations were investigated and a long-term index was proposed.We analyzed the mechanism of short-term economic fluctuation from three aspects: the composition of three industries,the influence of three major demands and external impact.Using the relative contribution of the three industries to GDP growth,we analyzed the relationship between the economic fluctuations in China and the three industries.We also analyzed the relationship between the economic growth and the contribution of the three major demands –consumption,investment and net export,with a special focus on the consumption and investment of China in major economies.In addition,we analyzed the relationship between economic fluctuation and economic policy,by studying the implementation pathway of economic policies in China.The impact of external impact on the economic fluctuations was also investigated.The main findings of this chapter are as follows.I)China's economy has entered a "new normal" from 2013 to 2019,and the economic development has entered a period of great easing,before COVID-19 breaking this balance.II)The economic growth momentum is evolving gradually.Before 2013,the economic growth in China was mainly driven by investment;after that,consumption become the "ballast" and "stabilizer".III)Compared with other major economies,China's consumption rate is low,while the investment rate is high.This deviation is even more prominent when comparing with developed countries.IV)Economic policies play an important role in China's economic fluctuations and they are important when coping with major external shocks.Through detailed analysis of the long-term factors of economic short-term fluctuations,we determined the most important influencing factors to economic fluctuations.We used factor analysis to reduce and weigh the dimensions and synthesized the long-term factor index(LTFI),which is the basis for later chapters to verify role of long-term factors in economic fluctuations.In Chapter 3,we analyzed the short-term factors of economic short-term fluctuation and proposed the attention index.Long-term factors are usually of low frequency and delayed in time.To overcome these problems,we used behavioral decision-making theory,limited attention theory and information asymmetry theory to study the attention behavior of micro subjects,taking advantages the highly frequent,real-time and large amount of Internet data.This study used text mining technology to extract keywords from the Internet search behavior of economic agents to depict the Internet attention behavior of economic agents,and uses Baidu search index to construct four short-term factor attention indexes: IAI,CAI,PAI and EPAI from four dimensions of investors,consumers,producers and economic policies.They were then used to test the ability of short-term factors on forecasting economic short-term fluctuation.The results show that the above mentioned four indexes are highly correlated with the corresponding economic indicators.They demonstrated cointegrating and Granger causality with the macroeconomic indicators.Thus,they can be used in short-term economic forecasting,and provide data support for the verification.In Chapter 4,mixed frequency data were used for short-term economic fluctuation prediction.In order to test if the long-term and short-term factor indexes constructed can be used for short-term fluctuation prediction,and show the advantages of using asynchronous mixed frequency data in the prediction,we studies China's economic fluctuations from January 1,2011 to June 30,2020.Based on AR(the benchmark model),ARDL,MIDAS and AMP-LSTM models were used.The data was divided into training and test set,and the performance of each model was evaluated.Ours results showed that by incorporating long-term,short term indexes or both,the prediction performance of the three models increased by 11.4%,3,4% or 18.9% from the benchmark model,respectively.This indicated that the indexes improved the prediction performance,especially long term indexes.In Chapter 5,we explored the application of neural network,in combination with statistical methods,in forecasting short-term economic fluctuations.We combined statistical metrology and LSTM neural network,and constructed a asynchronous mixing phased LSTM(AMP-LSTM)model.Our study showed that: I)AMP-LSTM model had smaller prediction error.Compared with the benchmark AR model,the prediction error of AMP-LSTM was reduced by 25.8%.The use of AMP-LSTM model reflected the advantages of mixed frequency data and using neural networks in short-term economic fluctuation prediction.II)The AMP-LSTM model had the ability to predict in advance.The data of the first 30 days and the first 60 days of the quarter can be used to forecast the GDP growth rate.By comparing the in sample training error and out of sample prediction error of 30 days,60 days and 90 days,using first 60 days achieved highest accuracy.Therefore,the forecast results can be published 60 days in advance,and the more accurate forecast results of the current quarter can be published 30 days in advance.III)AMP-LSTM can forecast the inflection point.When predicting the occurrence of a inflection point,the F1 was 7.7%,31.3%,50% or 50% respectively,for AR,ARDL,MIDAS,or AMP-LSTM,and only the AMP-LSTM achieved an out-of-sample F1 of50%,comparing to the other models of 0.Overall,AMP-LSTM model has the best performances.In conclusion,this study provided broader directions for macroeconomic research.By studying the short-term fluctuations of economy,we proved that the Internet search data and machine learning methods were effective and should be considered in macroeconomics studies.Secondly,we should make use of Internet big data for macroeconomics.Compared with traditional statistical survey data,the Internet data is real-time and of massive size and high frequency.Internet data should be combined with conventional statistical survey data in future macroeconomic studies.