Surprisingly, the Taylor rule was referred to more frequently than ever in 2009. According to Google Scholar more articles referred to it than in any year since 1993 when John Lipsky, now First Deputy Managing Director at the IMF, first called the rule by that name. Many more pieces appeared in blogs or in the news media.
The increased commentary is surprising because the Fed did not change its interest rate target once during 2009. Most likely the reasons for the attention are that: (1) the rule is cited as evidence that interest rates were “too low for too long” in 2003-2005 thereby helping to cause the financial crisis, (2) the rule can be used to help determine when the Fed should or will increase its interest rate target above zero, (3) the rule is used by some to determine how much quantitative easing is needed.
Many excellent pieces were written, in my view, including several by Michael Woodford of Columbia and Vasco Curdia of the New York Fed on adjusting policy rules during financial crises. 2009 also saw the release of the 2003 FOMC transcripts with telling references to the Taylor rule by Ben Bernanke. Columns by Michael McKee of Bloomberg and Gene Epstein of Barron’s were clear and insightful.
At the other end of the spectrum was Brad DeLong’s recent Taylor rule blog post, which unfortunately contains serious errors. For starters, he asserts that “John Taylor in the long run wants ‘Taylor Rule’ to mean any statistically-fitted reaction function in which interest rates respond to inflation and the output gap and not to the one rule he fitted over 1987-1992.” In my original paper I did not “statistically fit” a reaction function over 1987-1992 or any other period for that matter. The coefficients of the rule in that paper were derived from monetary theory and models developed during the 1980s. As I explained in the paper, using a variety of quantitative economic models, I found, through stochastic simulations, that such a rule worked well in stabilizing inflation and real GDP. Statistically fitting such a rule over a short five-year span would make little sense and adding earlier years would have made even less sense because Fed policy during the 1970s was terrible. To illustrate how the rule would work in practice I pointed to episodes when the rule was similar and also to episodes when it was different from what the Greenspan Fed was doing. There are no reaction function regressions in that paper.
De Long also tries to make the case that a policy rule which was statistically fit by San Francisco Fed economist Glenn Rudebush is an improvement over the rule I proposed. That estimated rule has a larger coefficient on the output gap and therefore gives lower interest rate settings now. It implies that the interest rate will remain at zero for a very long period which is what DeLong advocates. He likes that estimated version, but curve fitting without theory is dangerous. In the case of policy rules, it can perpetuate mistakes: the higher coefficient on the gap may be due to periods when the funds rate was too low for too long. Also interpretation of the lagged interest rate in fitted regressions is very difficult.
DeLong provides no demonstration that a higher coefficient on the output gap is an improvement over my original proposal. The recent review paper by me and John Williams, Director of Research at the San Francisco Fed, reviews the debate over the size of that coefficient and shows why a higher coefficient is not robust. Moreover, others, such as Bob Hall, argue that the coefficient on the output gap should be lower, not higher, than in the Taylor rule because of uncertainty of measuring the output gap. DeLong does not mention such alternatives.
DeLong is wrong about what he claims “John Taylor in the long run wants.” Don’t we all want good monetary policy? If there is a better policy rule that improves economic performance, then I am all for it, and I don’t much care what you call it. In 2008 I proposed adjusting the Taylor rule with the Libor-OIS spread to deal with the turbulence in the financial markets. The Curdia and Woodford papers analyzed that proposal and improved on it by changing the coefficient on the Libor-OIS spread. Their analysis was not based on statistical curve fitting, but rather on good monetary economics, which is what we need more of right now.