« Triple-Levered ETFs: Basement Bomb-Building Kits? | Main | Top 20 CDS by Net Exposure: Italy/Spain Rule »

Latest Stories

• Palm Pre, Plus Roger and Me
• You Say “Recession”, I Say “Recovery”, etc.
• Quote du Jour: Marc Rich
• South Korea Arrests Barry Ritholtz
• Afternoon Reading 01/08/09

November 4, 2008

Data Mining Presidential Elections: Bad for the Grass

It has become more or less predictable now in presidential cycles (and I've been guilty of playing along now and then too). We mine presidential election cycles to come up with historical predictors of which presidential choice will be best for the markets over the next four years, next few months, next ten minutes, etc.

We are told the best presidential returns come during Democratic administrations. Or we see that the Dow usually turns in positive numbers from the election to the end of the year. Or that when a Democrat president is elected the Dow, on average, declines through the end of the year. Or that when a Republican is elected the Dow usually gains through the rest of the year. And then there are more specific variants, like what happens to year-end in a Democratic sweep of the Congress and the Executive Office, or a split, with a Republican presidency and a Democratic sweep of Congress.

Such are the perils of our data-drenched times, and the entire exercise is mostly a waste. While there is no denying presidents can affect markets, and that electing, say, Mao or maybe Mussolini would really irritate capital markets (I think), for the most part markets simply want the entire exercise over with so that they can get on with doing their usual fluctuating thing.

But what about the data?! Isn't it true that Democratic presidents have delivered the best returns? Isn't it true that returns to end-of-year are better under new Republican presidents than new Democratic presidents?

All true. But the empirical issues are myriad. For starters, we have a teensy sample size. There have been 28 presidential administrations in the U.S since 1900, 12 of which were Democratic and 15 Republican. That is already giving us small-ish numbers from which to extrapolate anything. The situation gets worse, however, if we make it all post Depression and post-WWII, which is arguably a good idea. Then we're down to 15 administrations, six of which were Democratic and nine of which were Republican. That is, to a first approximation, a data set of size zero -– especially if you then begin slicing things up further to see what happens when you have a Democratic sweep (which, if you're curious, takes you down to a sample size of four since WWII.

If teensy sample sizes were the only issue, that might be almost forgivable. But we also have variance problems. Consider that the average return from election day to end of year under incoming Republican presidential administrations is 4.1%, while –1.0% under Democrats. That's impressive, right? Not really. The standard deviation of returns is 8.3% under Republicans and 5.3% under Democrats over the post-election period. In other words, it is entirely possible, statistically speaking, the returns are actually the same –- or that maybe Democrats do better. We just don't know enough to know.

Bottom line: Extrapolating much of anything useful from presidential data is entertaining, but a waste of time for everyone but the pundits who get to fill some air time with pseudo-scientific yapping.

[The data in this post came from a host of sources, including Yahoo Finance, Wikipedia, Bespoke Premium, and my own prior posts on the subject.]

Sphere It   |   Digg it   |   Bookmark it   |   Stumble it   |   Facebook it