Methodological problems in studies on the Taylor rule

This paper raises concerns about the methodological approaches commonly adopted in typical Taylor rule studies. We find that many empirical studies on the Taylor rule do not follow the required econometric procedures. These studies ignore the presence of unit roots, cointegration, and serial correlation in their tests and estimation. The Taylor rule equation is typically estimated in levels. We show that the Taylor rule can be an unbalanced regression that involves a mixture of I(0) and I(1) variables. Spurious regressions may occur if the variables are not cointegrated and the Taylor rule equation is estimated using variables in levels. In addition, empirical models of the Taylor rule commonly include lags of the dependent variable, and equation residuals are serially correlated. The presence of lagged dependent variables and serially correlated residuals will cause biased and inconsistent least squares estimators. To illustrate our arguments, we re-examine two recent papers to point out the econometric problems that are general in typical Taylor rule studies. We show that an inadequate analysis of the time series properties of the individual series and diagnostic checks of the estimated equations can often lead to invalid conclusions about the empirical validity of the Taylor rule. We demonstrate how autoregressive distributed lag methods can overcome these issues and how the equation can be estimated efficiently.