Jan

29

 Seems like a good idea, and so I bought some GREK at 10.71 this morning.

I wrote to the professor to ask if the model that called for the buy on Greek stocks is publicly available. Haven't heard back yet.

I became a Shiller skeptic after reading an article of his that had what I think was a serious statistical error. His thesis was that market valuations fluctuate "too much". To prove it, he compared each year's valuation with the then present value of the forward earnings, known retrospectively over the following 50 or so years–I'll call that number the "retrospectively known present value" or "RKPV". Sure enough, he found that market valuations fluctuated much more than the RKPVs. He correctly pointed out that if a predictor (the market valuation) has a bigger variance than what's predicted (the RKPV), then the predictor clearly has room for improvement.

The problem though is that he only had on the order of 100 years of data to work with. Since each RKPV involved about 50 years, he effectively had a non-overlapping sample size of around 2 — so small as to be meaningless.

In 1932, people thought the RKPV would be very low. It turns out that they were wrong. But if we had 1000 years of data to play with, it might turn out that the next "1932" event will indeed be followed by a super-low RKPV, perhaps even lower than the market's valuation. So we don't know enough to conclude that the market is irrational.

The Chair has pointed out that Shiller made a similar mistake in his "CAPE" analysis. He'd have roughly 100 years of data, which sounds good, but his CAPE predictor looked back 10 years and predicted forward 10 years, so the non-overlapping sample size, again, was only around 10. Shiller would claim the CAPE was good at predicting the 10-year forward return, The Chair, however, did a little snooping and showed that it was in fact bad at predicting the forward 1-year return, getting the sign wrong even. It's kind of implausible to claim that one has a predictor for the year 2024 return, but it can say nothing about 2015. So that further suggests the CAPE results are an artifact of small sample size.

It's possible I've missed something, but these seem like novice errors. It made me suspect that it's a little easier to get the Nobel Prize in Economics than in Physics or Chemistry or Biomedical Sciences.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search