Interdum stultus opportuna loquitur...

Thursday, August 26, 2004

US Data Methods - a Brilliant Review

Note - from June 24th 2009, this blog has migrated from Blogger to a self-hosted version. Click here to go straight there.

Those of you who are slightly mystified at my constant carping about the chicanery behind "official" U.S. Statistics, I implore you to read the following:

A Primer on Government Economic Reports

Further to my previous comments about the balderdash involved in saying "it's a Presidential election year, so the markets are likely to rise", a bit of a look at the data shows a couple of things: First, it appears that since 1872, markets are a little under three times as likely to rise during an election year as they are to fall. This contrasts with the "all years" ratio of 1.74 - that is, markets rise, on average, a little under twice as often as they fall.
The problem of sample size rears its head; with so few observations for election years, it turns out that 2.67 is statistically "the same" as 1.74.
No Restriction on PEUpDownRatio
Election Years2492.67
All Years82471.74

When we introduce a constraint to try to account for valuation (at the beginning of the year), the Presidential year ratio doesn't change very much. I've used a value of 15x 12-month trailing earnings as the hurdle, which is being pretty kind, since 15x is close to the average PE over the period. The "who cares what year it is" data also shows no significant change.

Trailing PE > 15UpDownRatio
Election Years933
All Years32201.60
Turns out that with so few data points (only 33 election years, and only 12 which showed even modest overvaluation), there is absolutely no statistical difference between the Presidential-election and non-Presidential-election years.
There have only been 2 Presidential Election years with a PE above 20; 2000 and 1992... as such even trying to perform an anlaysis of what happens in a Presidential election year with a valuation as high as we have now, is fruitless.