Nate Silver invades our turf

Robert Waldmann

Having demonstrated an ability to find amazing patterns in numbers from baseball and polls, Nate Silver turns his attention to the stock market.

He finds mean reversion in day to day prices since 2000 ! That is evidence of over-reaction and a correction. Amazing !

more after the jump.

Now Silver knows he is taking a chance talking about the stock market

“Wading into a discussion about the stock market is dangerous business. In contrast to politics, where there are relatively few people willing to do serious quantitative analysis, there are entire industries full of such geeks in the financial sector.”

Now I remember the time that my boss and advisor (Larry Summers) asked me if daily changes in S&P 500 closing prices had more mean reversion when the variance of the price was higher. This would be great evidence that noise traders were causing high volatility.

I thought I found evidence, but there was a problem. In the late 40s and 50s there were days during which no shares of a given company were traded. No one bought or sold a share. Their price is recorded as the price at which they were traded the day before. This automatically creates a bit of positive correlation in day to day returns. This “non traded stock effect” appears in diluted form in the S&P500.

Now Silver uses the Dow Jones Industrial Average made up of shares which are traded frequently, so, I’m not sure this is a problem for him. That is, for all I know, at least one share of every Dow index company was traded every trading day from 1946 on. Still I worry.

The non traded stock effect suggests that some of the evidence of positive correlation in the past might be spurious.

Silver mixes in data from the extremely low trading volume 40s and 50s with data from the 60s and 70s. I would drop years before 1960 or look at correlations between the change today and the change the day after tomorrow. So I am not convinced by this part of his post.

What’s interesting, however, is the nature of this correlation has changed over time. From 1946 through 1979, the market posted a gain about 57.7 percent of the time following a gain on the previous day — but just 45.6 percent of the time following a loss. This suggests that the market was slow to react to new information, with rallies or slumps sometimes taking several days to play out.

However, I am convinced by the comparison of the 80s and 90s to the noughties

Between 1980 and 1999, he market posted a gain 51.8 percent of the time following a prior day’s gain, versus 52.7 percent of the time following a loss — essentially no difference.


Since then, however, something funny has happened. The market has developed a tendency toward inverse serial correlation; it is more likely to follow a gain with a loss, and vice versa. Since 2000, the market has gained ground just 47.7 percent of the time following a prior day’s gain — but 54.2 percent of the time following a loss. !!!!!!

(emphasis added).

Damn why didn’t I see that ? Oh yeah Summers asked me in 1987.

Bottom line ? Silver is golden. I think he can step into a crowded competitive field, do some simple calculations which should have been done years ago, and hit it out of the park.