Weibo Volume And Price Change In Financial Markets

With the exponential growth of information available and the development of ways to storage and analyze data,we find new opportunities for data analysis.While in the past much of the research revolved around creating theories that described our reality in most situations and refine as we find more information.In the present,through using massive amounts of datasets we are able to not just create a theory,but to do a pure empirical research models using artificial intelligence,machine learning,statistics and database systems to describe reality as it is.Many fields of research are now benefitting from the sudden increase in information gathered and stored,including science,social sciences,humanities and business.Social media,the target of our research combines most of the fields mentioned.First we are required knowledge of computer science in order to collect the data,second we need knowledge of social sciences to formulate the assumptions and filter the data properly,lastly we need business,or finance,to guide our data analysis and find useful information on financial trends.As preparation we signed up on Weibo as a developer user,in order to have access to more information and to be able to use commands on the API(Application Programming Interface).On the first step the research involved using a Google Spreadsheet combined with programming to access the Weibo API and gather data relevant to our research.This step found the 100 most influential finance accounts on Weibo,by follower count,and followed them on the developer account.After that on the API we gathered hundreds of daily Weibos from the 100 most influential financial Weibo accounts over a two to three month period,and separated in a 30 day and a 38day sample.The second step we made sense of the information,after looking through the raw data,we selected two words possible to better analyze the samples,'stock' and 'gold'(the words were in Chinese language'?' and'??'),these two words were chosen because they were found more often than other indexes in the Weibo posts and we are able to relate them to daily prices on public sources,respectively,Shanghai Composite Index and world gold market price on Bloomberg.We then developed a graph to show how and if the amount of Weibos on a subject reflects the absolute change in the market prices on both keywords and both samples.The third step,after noticing a relationship between the graphs of the price change predicted on Weibo and the actual price change in the price markets,we decided to perform a regression using the data.First we checked for stationarity,in this case we found that the vast majority of the unit root tests resulted in a very low p value,meaning our samples were linear and we were able to perform a linear regression on the samples.The fourth step involved the actual regression.First we performed Ordinary Least Squares,since it is a very common and recognized method of regression,with centuries of proven use and research.In the OLS regression,while we did find a regression,the p value was not as low as expected,so after trying different methods of regression,including Quantile Regression,VAR,GARCH,we noticed that the General Linear Model using Newey West Covariance Method lowered the p values considerably,we then presented this regression's results.On the fifth and last step,we discussed the limitations and conclusions of our work.Among them we can list that our research was absolute price change related,we did not do Sentiment Analysis or Picture Analysis that could improve the accuracy of our results,so our research can only predict price change quantity but not the direction in change.The second was that our research was not real-time,we collected the data and later predicted with an already complete sample,in a real life use we would need to collect past samples and relate them to fture predictions,in this case we essentially predicted,or established relationships for the past.The third limitation was the sample size,while we collected daily Weibos for over 2 months,an improve in sample size would be very useful in a more thorough research.While there were limitations,we did collect real-life samples,we were innovative in our research seeing as of the moment of the writing,we didn't find or compare to similar work published using the number of posts on Weibo and change in market prices,our research was not a simple re-do of post work.We obviously researched and we did look at past work done mostly on Twitter,all the regressions and graphs done on the samples were ours.Any picture not developed by the author was clearly given credit for.Given as this is a very popular topic recently,there are too many recommendations and possible improvements.Bigger sample size,different regression models,different types of analyzes on the same sample,different keywords,those are just some of the possible recommendations.The methods,technologies and opportunities in exploring social media are too dynamic and I am sure that in a few years the questions and models will be very different than as of the time of the writing.