Research On COVID-19 Based On ARIMA Model And ARDL Model

The COVID-19,named by the World Health Organization as "Coronavirus Disease 2019" broke out in 2019.It swept the world rapidly with its extremely high infectivity and variability.Although a vaccine against COVID-19 has been developed in China,the epidemic situation is still serious.How to predict the arrival of the COVID-19 in advance is worth thinking deeply.This paper explores the relationship between Baidu index of keywords and the newly confirmed cases from the search volume of relevant keywords on the Internet,providing a way to monitor when and where the epidemic occurs.Firstly,considering the transmission characteristics of COVID-19,this paper used a univariate time series ARIMA model to predict the new confirmed cases.Secondly,in order to study the search volume of relevant keywords on the Internet before the outbreak of the COVID-19,the Baidu index of each keyword was used to calculate the Spearman correlation coefficient between Baidu index and the newly diagnosed number of people,and the Kendall concordance coefficient test showed that the Internet search volume of related terms and the newly confirmed cases were consistent.Then take the keyword search volume of Hubei and Hebei provinces before the outbreak of the COVID-19 as an example to verify the regional correlation between Baidu index and the newly confirmed cases.Finally,the network search volume of each keyword was added into the univariate time series model to establish the ARDL model of multiple time series variables,to determine the lag order of different keywords,and to clarify the dynamic development process of Baidu index and COVID-19.Through the Baidu migration index,the relationship between the time when COVID-19 first arrived in each province,the geographical distance and the migration rate was explained,so as to prepare for the prevention and control of the epidemic in each region.The results showed that the analysis of COVID-19 using univariate time series model can achieve a good effect in short-term prediction,and the method was simple.In addition,Pearson correlation coefficient and Spearman correlation coefficient showed that Baidu index and the newly confirmed cases had significant positive correlation in both time dimension and space dimensions.Kendall concordance coefficient test verified the consistency between Baidu index with different lag periods and COVID-19 from a statistical perspective.Therefore,the Baidu index reflected the development and severity of the epidemic in a way.On the one hand,the ARDL model with multiple temporal series variables can well predict the development of COVID-19.On the other hand,the prediction results had passed KS test,and the prediction effect was improved compared with the ARIMA model and the regression model.Through this study,it revealed that Internet-based search information can predict the outbreak of COVID-19 in a way,which provided a theoretical basis and guiding significance for the Internet data in the early warning and prevention of COVID-19 in China.