A while ago I read an interesting quote from Google saying that because they can see searches in Real Time they are able to effectively see market trends long before the public consciously recognizes them. This struck me as really cool, so when I came across a Boston startup called Compete (www.compete.com), and saw that you could download data in regards to web analytics, I became pretty excited.  If page hits is covariant with web trends and consumer habits then Compete tells me that I should be able to predict the next web trends based on their data. I thus became curious if we could predict stock prices.

Therefore I took a sampling of 3 companies where I thought web hits would directly relate to economic success: Apple, Ebay, and Amazon.  The more people who visit these sites, the more transactions these companies are going to generate…or so the theory goes.  Further, one would hope that there is some sort of delay between site hits and stock price such that one could set up a stock portfolio and trade upon our non-public information (semi-strong theory of price efficiency).  Unfortunately, I’m slightly limited in the data that Compete provides (for non-paying, and or, cheap-skate, users), as they only have aggregated monthly numbers for 1 year. Thus, for a comparison, I averaged daily stock prices to get a monthly average.  This makes it increasingly difficult to separate noise out of visitor numbers and stock price, but hopefully the numbers can provide us a “trend” of sorts. Later on, to add robustness, I might look at the “Engagement” metrics that are provided to see if this can tell us anything…but for now, I’m lazy.

First: The basic overlay between stock price and site volumes.

From the graphs above its hard to tell if there is any causation, but there definitely seems to be a trend occurring and possibly correlation (Amazon at .39, Apple at .28 and Ebay at a negligible .10).  Although, the T-Stats, p-values and R squareds of each basic regression of price to volumes indicate that we can’t make any conclusion from a basic overlay.

However offsetting the price by 2 months increases the correlation pretty significantly by the following: Amazon at .72, Apple at .57 and Ebay at .75. (Offsetting means that Volumes from T-2 months prior are matched with Prices @ T 0)

Here are the results of a basic regression on the Offset Data:

You can see that our T-Stats are now meaningful, p-values are all great with the exception of Apple which is slightly higher than you would normally like, and R^2 show that half of the movement in price can be explained by site volumes.

Obviously, a much more rigorous analysis would need to be completed in order to prove anything, hopefully with much more data on many many more companies, but the preliminary results are kind of cool.  Maybe our next hedge fund will come from Compete. Oh, and btw, no secondary research was completed so its entirely possible that this study has been done before. Let me know if you’ve seen anything like it.