Blog chatter vs prediction markets
I was wondering about the extent to which blog chatter correlates with information from other sources such as prediction markets. This is pretty difficult to test, because the things that people write on blogs can be hard to decode and categorise. But I think I found one reasonable and straightforward test case: use of the word ‘recession’ in blogs versus the probability of a recession from a prediction market. The theory is that when a recession is more likely, people should blog about it more, and the level of recession chatter should be positively correlated with the probability of a recession predicted by a prediction market.
So I got the daily counts of the number of times ‘recession’ was used in blogs, according to Technorati, and the probability of a US recession in 2008 according to Intrade. The overlapping data for both series covered about the past 90 days. Of course, people could blog about recessions in any country, not just the US, and people could use the word ‘recession’ when they’re saying something like ‘a recession is unlikely in 2008′, so I expect the relationship to be far from perfect. Here’s the basic data that I got, with the count of the word ‘recession’ on the left, and the Intrade probability on the right:
Here’s a scatter plot of the two data series together, with a fitted logarithmic curve (it fits slightly better than a linear relationship):
For those econometricians reading this who may be thinking spurious regression, I did a Phillips-Ouliaris cointegration test on the fitted model above and the p-value was 0.079, so it’s borderline whether this is spurious or not, but I’m willing to believe there’s a meaningful relationship.
5 Comments
Cool. How do you download these data?
John: I got the Technorati data using their API (applications programming interface). Details are on their site here: http://www.technorati.com/developers/api/ … I used the ‘dailycounts’ query to get the recession data for the past 180 days. It is returned in XML format which can be easily imported into Excel. The Intrade data can be downloaded by clicking the ‘Advanced graphing’ link on the page that I linked to.
I write about recession probabilities, but not because I see one as likely in the next twelve months. It is the need to deal with the obsession of other bloggers — particularly non-economists, that drives the topic.
Having said this, it just means that I am reflecting the overall sentiment, even though I do not agree. I am still an indicator….
Facinating problem. I linked it and made comments at my blog (see http://learnecon.blogspot.com/2008/01/recession-blog-chatter-vs-prediction.html)
Thanks.
I’m not sure spurious is the right term for this one. I tend to think of spurious regression in cases of high R^2. (R^2 > D.W. is often cited in the literature.) The R^2 here is not particularly good.
Steven is on the right track in asking about an omitted third variable. Jeff is correct as well. But there’s more.
The blog chatter is a more volatile series (just by eyeballing the chart). Part of this is probably due to the fact that many blogs quiet down or post lighter topics on weekends. This may explain the fairly regular dips in the series throughout the sample.
There is also a weird lead/lag relationship in the data you post. It looks to me as if blog chatter was lagging the probability in late August/early September and then it was leading the probability back down in late September/early October. Again, I’m just eyeballing it, but take a closer look and see if you agree. It definitely looks like the blogs lag at the beginning. Whether they lead later on depends on how you squint while looking at the data perhaps.
My money is on economic news and mainstream media references as driving both. The lead/lag relationship could be explained by the bloggers responding for a few days after some arrival of news to which Intrade responds immediately (lagging on the way up). When no adverse news arrives, the blogs say little and over a period of days the Intrade probability goes down (leading on the way down).
Just a thought.