Monetary policy rules can be used both for prescriptive and descriptive purposes, but it’s important to be clear about which purpose one has in mind. A policy rule estimated over a period which included the Great Inflation of the 1970s, for example, might be a good description of policy during that period, but it would be a terrible prescription for policy today. Similarly, a policy rule estimated over a period which included the 2003-2005 period, when rates are unusually low, would not be a good prescription, in my view, for policy today.
Misunderstandings about whether a particular rule is meant to be descriptive or prescriptive can thus lead to policy mistakes. Consider this passage from the March 18, 2011 issue of JP Morgan’s Global Data Watch (p 16): “The original Taylor rule was descriptive and meant to match how Fed policy was set in the 1987-1992 period. Subsequently, some researchers found that variants of Taylor rules were optimal in certain economic models, and so could also be prescriptive. However, those classes of models were usually quite restrictive, such as assuming policy was not at the zero bound, or that the central bank only tried to influence the short end of the curve. In fact, very few optimal policy models even incorporate multi-period interest rates.”
But the Taylor rule was not meant to be descriptive as I made clear in my original paper. Rather it was very explicitly meant to be prescriptive. I derived it by experimenting with different types of rules in stochastic simulations of different monetary models, including my multi-country model at Stanford, and by studying the results of other people’s simulations. This pinned down the left-hand side variable and the right-hand side variables, and led to simple functional forms and coefficients.
The models used to derive this rule were not restrictive in the sense used in the quoted paragraph. My multi-country model has a term structure of interest rates, with long rates affecting aggregate demand. Exchange rates are part of the monetary transmission mechanism through their effect on exports and imports. The zero bound on the interest rate was taken into account in the stochastic simulations. Later, simpler, three-equation, textbook models were used by other researchers to show that the Taylor rule was optimal using formal dynamic optimization methods (reverse engineering). These simpler models helped students understand how policy rules work.
I referred to the 1987-92 period in my original paper, but the rule was not an estimated rule over that period or any other period. After deriving the rule I noted that it was close to actual policy during in 1987-92 and used that result to work through several case studies of how such a rule would have worked in practice. A pure estimation strategy would not have focused on that period, and even if it did, it would have included more variables on the right hand side (such as money growth or the exchange rate). An example of an estimated rule in existence then was that of Ray Fair of Yale.
When teaching Economics 1, I stress the difference between positive (descriptive) economics and normative (prescriptive) economics. This discussion of policy rules illustrates the importance of understanding the difference in practice. You can’t justify QE2 by saying that the interest rate is negative with the prescriptive policy rule I proposed, because the implied rate is not negative, it’s close to 1 percent.