Predicting bourse trends from Twitter moods

The subject dates from 2010 and was published on Rue89, but rnews

Even the world, PCWorld lesechos or resumed in the new choir, as usual by copying and pasting the information from the AFP without actually reading the study .. (sunny days are better too)


To help you understand, here are the key quotes

Between February and December 2008, three researchers from the University of Bloomington, Indiana, studied 9853498 « tweets » – messages of 140 characters traded on the micro-blogging site – written by 2 7 million users.

The researchers used two programs: Opinion Finder, which lists the messages « positive » and « negative » on the Internet and Google Profile of Mood States, whose approach is a little more subtle. This software differs in effect six different mindsets, « calm » to « happy. »
Twitter Wall Street ahead of two to six days

Result, the proportion of messages expressing « calm » would predict the movements of the Dow Jones in two or six days, with an accuracy of 87.6%.

You perplexed? The three researchers also. First, they wonder why only the « calm » would such predictions: the tweets expressing « happiness » have not been followed by a surge in Wall Street.

Importantly, researchers readily admit that their study « provides no information on causal mechanisms that may link the public moods and changes in the Dow Jones. »

In trading rooms, the son of financial information agency Reuters and Bloomberg will they soon replaced by Twitter? Some traders may be tempted: be sure to purchase 87.6% of shares that will rise or fall, is to ensure nice gains.

Incidentally, the number of prediction becomes 90% in some articles .. but none of the items appear to have actually read the study. Comments from readers of these news are very critical, denouncing the fact that known in statistics, two facts may seem related and not be at all

Here jumble why this pseudo study is not worth much:
– What about the meaning provided by each « mood »? These figures are balanced mean absolutely nothing: What predicts what? calm predicted increase or decrease? With what certainty?

– What about the test protocol? Over an interval short enough, I can correlate almost everything with everything. For example correlate the ignorance of the journalist who took up this information, the length of his article.

– What about interest? Interpretation. We play the little chemist ‘like it’s funny poop by mixing journalist and trust me, I get an item, there must be an explanation.  »

– The conclusion? Idiot. A financial arbitrage disappears as soon as it is known. Because this information inevitably fall into the hands of the highest bidder, and more influencers, the prediction would be self-fulfilling. Its predictive value in two days would be zero, because the realization is almost immediate. It was a bit like the journalist spoke of trader to look pretty, they are also looking for keywords for their site?

– What have we learned? Many things: it is time to take a course of intellectual self-defense, the scientist has a bright future ahead of it, the job of a journalist is to « pass » something you read without not understand that U.S. researchers can sometimes work on topics morons.

The problem is that the researcher backtests going through the power of the computer and its data base to try to ‘bring to life’ the theory by changing the settings until it works but forgetting that he works on series existing (it tests, it does not work then change the settings until it works). Then he says: test on 2 years running, I’ll take a third .. and then wham the success rate to 3 days but the prediction diminishes to 2j increases, so now I say hop Twitter predicts the stock market with two days in advance .. and so on

In fact, you can read the original study on the relationship between the DJIA and Twitter. It is much less simplistic than newspaper articles, which are in TMZ and here. Including « Le Monde » and « Les échos ». That’s why your sources of information should be serious sites, not the mainstream media.

Laisser un commentaire

Votre adresse de messagerie ne sera pas publiée. Les champs obligatoires sont indiqués avec *