Identification and the liquidity effect : A

ing from discount window shocks I first consider the case in which shocks emanating from the discount window, υ, are small enough to ignore. To remove the effects of ε from NBR, CEE make the following identification assumption: that aggregate output, y, and the aggregate price level, p, contemporaneously reflect the effects of ε, and not the effects of μ. Their a priori reasoning behind this crucial recursiveness assumption is that— particularly at the monthly level of time aggregation—it is reasonable to think that monetary policy actions have essentially no contemporaneous impact on aggregate output and the aggregate price level. Below, I review the other efforts made by CEE to check the plausibility of this identifying assumption. The CEE identifying assumption rationalizes the following two-step procedure for isolating the monetary policy shock, μ. First, do an ordinary least squares regression of nonborrowed reserves on y and p and treat the residual as something that contains only μ and not ε. In the second stage, regress the interest rate on the residual. In the example, the residual from the first-stage regression would be (1 – a 1 γ) μ if the data set were large. The coefficients in the regression of the interest rate and of TR on the residuals from the first-stage regression are a 1 /(1 – a 1 γ) and 1/(1 – a 1 γ), respectively. Consistent with the sign assumption on (1 – a 1 γ) made above, the latter regression coefficient turns out in practice to be positive, so that the sign of the first regression coefficient coincides with that of a 1 . Thus, under the CEE identification assumption, the sign of the regression of the interest rate on the residuals from the firststage regression constitutes a valid estimate of the sign of the liquidity effect and avoids the pitfalls discussed above. Taking discount window shocks into account The preceding analysis assumes that exogenous discount window shocks, υ, are negligible. If they were important, then CEE’s inference