October 18, 2010 at 17:14Twitter Psychometry Predicts DJIA for 2-6 Days
Johan Bollen (School of Computer Science at Indiana State University in Bloomington) and a group of his colleagues nevertheless published the sensational work on the statistical correlation of sentiment on Twitter and the Dow Jones Index (DJIA). They made the announcement of their research earlier, and the named figures on the accuracy of matches surprised many colleagues. October 14, 2010 published their scientific work in its entirety ( PDF ).
Johan Bollen and colleagues, using the Granger causality test , analyzed the Dow Jones index time series and twitter sentiment from 02/28/2008 to 11/03/2008 on weekdays. To identify emotions in tweets used the system of linguistic analysis OpinionFinderand GPOMS, a Profile of Mood States (POMS-bi) psychometric system, expanded by 4- and 5-letter words, compiled by Google in 2006.
The researchers tried different options with a time shift and found that according to one of the parameters, the accuracy of the matches leaves no chance.
The table shows the statistical significance (p-values) of the two-dimensional correlation according to the Granger causality test between moods and DJIA from 02/28/2008 to 11/03/2008.
As stated in the scientific work, “based on the values of the Granger causality test, we can with a high degree of confidence reject the null hypothesis,” that is, the hypothesis that there is no relationship between the two data sets. One of the parameters of psychometry (calm, calmness) shows a very high correlation with DJIA with a shift of 2-6 days, and here the results can be called statistically significant.
The value “calm” predicts the movement of the index with a probability of 87.6%. If you superimpose them on one chart, you will notice intervals when the mood on Twitter does not affect the index at all. According to scientists, these days the market is affected by unpredictable factors. For example, an unexpected Fed statement on a change in the discount rate or quantitative easing, etc.
The research methodology is shown in the diagram.
Johan Bollen and colleagues, using the Granger causality test , analyzed the Dow Jones index time series and twitter sentiment from 02/28/2008 to 11/03/2008 on weekdays. To identify emotions in tweets used the system of linguistic analysis OpinionFinderand GPOMS, a Profile of Mood States (POMS-bi) psychometric system, expanded by 4- and 5-letter words, compiled by Google in 2006.
The researchers tried different options with a time shift and found that according to one of the parameters, the accuracy of the matches leaves no chance.
The table shows the statistical significance (p-values) of the two-dimensional correlation according to the Granger causality test between moods and DJIA from 02/28/2008 to 11/03/2008.
As stated in the scientific work, “based on the values of the Granger causality test, we can with a high degree of confidence reject the null hypothesis,” that is, the hypothesis that there is no relationship between the two data sets. One of the parameters of psychometry (calm, calmness) shows a very high correlation with DJIA with a shift of 2-6 days, and here the results can be called statistically significant.
The value “calm” predicts the movement of the index with a probability of 87.6%. If you superimpose them on one chart, you will notice intervals when the mood on Twitter does not affect the index at all. According to scientists, these days the market is affected by unpredictable factors. For example, an unexpected Fed statement on a change in the discount rate or quantitative easing, etc.
The research methodology is shown in the diagram.