The Problem With The Google Trends Market Prediction Data


Leinweber, David

July 19, 2013

Greetings Social Media in Financial Prediction fans.  The attentive reader will recall that last time we became suspicious about the data underlying the basic Google Trends Market Predictor, based on the the simplified analysis of the the change in the search volume on the magic keyword “debt.”

The secret sauce (no, not mayo, ketchup and relish, though that is mighty good) was the trend in searches for “debt” over three weeks. But since we are still waiting for the “official Google Trends Application Program Interface” announced by Marissa Meyer about six years ago, we need to rely on bootleg API’s, like the one offered up by reader and ace nerd-on-the-street Josh Z.

Here are a couple of charts that show what we’re up against here.  Google seems to update the trends information in a vaguely erratic way, without normalizing by the time the data is updated. Take a look below to see what we mean. The first chart shows the 90 day trend in debt search, grabbed from Trends on Sunday evening (July 14).

The nose-dive on the far right, down to the under 10 in relative search volume is the problem. In the “big data” biz, we call that an artifact of reporting. It used to be called a glitch or a a bug, but the pay is better for “artifacts of reporting”. Better still for “zero day exploits”, but that’s another story.

Anyway, now take a look at the same 90 day chart for searches on “debt, just three days later, below.

What happened to that reading of 10 in relative search volume? Nowhere to be seen! We still see the cyclic pattern for weekends relative to week days (which, depending on the exact time one grabs this stuff, could make weekend readings unrealistically optimistic for the DJIA), but the drop down to 10 and below, seen just this weekend is gone!  ”Relative” is obviously a relative term when it comes to reporting of Google Trends.

So the Market Predictor remains up on the lift until we have enough data to accurately calibrate it. But all is not lost.

Clearly, the secular trend over the last 90 days is down on relative volume of search on “debt” and the secular trend on the DJIA is clearly up. Look at the 90-day chart below and score a big one for the trends predictor.

The weekly predictions, including the one for this week, are still subject to those pesky “reporting artifacts”, though if we pull out the magnifying glass, and gaze real hard at the “debt” trends, peak-to-peak, Monday-to-Monday, or take your pick we see pretty much a flat line.

So for this week, the Google Trends Market Predictor says “meh, not much happening”.  We’ll keep poking at this rascal and keep you up to date.

Great ! Thanks for your subscription !

You will soon receive the first Content Loop Newsletter