Wednesday, March 04, 2009

Wanna bet some of that Nobel money?

Paul Krugman suggests that my skepticism about the administration's growth forecast over the next few years is somehow "evil." Well, Paul, if you are so confident in this forecast, would you like to place a wager on it and take advantage of my wickedness?

Team Obama says that real GDP in 2013 will be 15.6 percent above real GDP in 2008. (That number comes from compounding their predicted growth rates for these five years.) So, Paul, are you willing to wager that the economy will meet or exceed this benchmark? I am not much of a gambler, but that is a bet I would be happy to take the other side of (even as I hope to lose, for the sake of the economy).

Related comments for econ geeks

On the substantive question that Paul raises about the unit root literature and the distinction between cyclical and other fluctuations in output, it is an issue that Campbell and I addressed in a companion paper, where we decided that the conventional wisdom on this matter, which Paul still espouses, does not hold. I do not claim we had the last word on the subject, but it is just wrong to say we missed the obvious point that Paul raises. I don't blame Paul for not being aware of this paper. After all, he is an international trade theorist rather than an empirical macroeconomist, and it is hard for anyone to stay informed about all literatures in the field.

Paul also directs us to a Brad DeLong post that includes this intriguing graph:

Brad infers from this cloud of points that higher unemployment typically points to more rapid subsequent growth.

There is not enough information presented for me to know whether to agree with Brad's inference. My guess is that this regression line (at least I presume it is a regression line) is completely driven by the few observations in the upper right, which are probably all from the Reagan-era boom that followed the 1982 recession. It looks like if you take out that one episode, the relationship would largely disappear. I would be curious to see the statistical significance of the regression, using the relevant serial correlation corrected standard errors (I believe that 8 lags would be needed, given the overlapping data). If I am right that what we have here is an uncorrelated cloud plus the Reagan boom, then I would not expect a high level of statistical significance for this relationship. If I am wrong, and the relationship is highly statistically significant, then it might encourage Paul to take the bet.

Update: Phil Rothman of East Carolina University was nice enough to email me the regression results. For the entire sample, the regression yields an R-bar-squared of 11 percent, and a t-statistic of 3.5. For the sample leaving out 8 quarters of the Reagan boom, the coefficient is smaller, the R-bar-squared is 5 percent, and the t-statistic is 2.1. (See also the figure here.) I will leave it up to Paul to determine whether these results are robust enough to take to the bank, so to speak.

Further update: Here.