Commentary on " Identifying the Liquidity Effect at the Daily Frequency:

Dan Thornton has written a very interesting paper on estimating the liquidity effect at the daily frequency. The paper starts with a nice discussion of the historical origins of the term “liquidity effect” (tracing it back to Milton Friedman, but somewhat surprisingly, no farther). The paper then discusses ways to identify the liquidity effect in daily data, and applies this discussion to critique the recent methodology of Hamilton (1997). The paper finds that the Hamilton results are not robust across sample periods and appear to be sensitive to outliers. Finally, the paper proposes an alternative way to estimate the liquidity effect via the relationship between nonborrowed reserves and the target funds rate. Here the paper has mixed success. Some variables seem highly correlated in the appropriate way, but the basic relationship between nonborrowed reserves and the target federal funds rate appears fragile—often it is insignificant, and sometimes it is the wrong sign. In my discussion, I intend to provide a brief overview of the literature on liquidity effects, followed by a very simple exposition of a model along the lines of the one presented in the paper. I use this model to illustrate the basic arguments in the paper, and then by changing the assumption regarding the relationship between the Open Market Desk’s (OMD) operating procedure and the desired level of nonborrowed reserves, I argue that we may have reason to expect biased and econometrically fragile parameter estimates from the nonborrowed reserves equation that is estimated in the paper. The modern macroeconomic literature on liquidity effects can be divided into two separate strands: the theoretical literature that examines whether the current generation of dynamic general equilibrium models can generate a liquidity effect in response to innovations in money growth rates, and the identified vector autoregression (VAR) literature which examines whether there exists a contemporaneous negative relationship between money and nominal interest rates in response to monetary policy innovations. In the theoretical literature there are two types of models that are capable of generating a liquidity effect: limited participation, cash-in-advance models with heterogeneous agents; and sticky-price models that generate a demand for real balances through a transactions demand for money or, equivalently, money in the utility function. In both types of models, the theoretical difficulty with generating a liquidity effect stems from the fact that, with positively autocorrelated money growth, a rise in money growth today leads to anticipated inflation. Such anticipated inflation tends to push the nominal interest rate up rather than down, even if the real interest rate is falling in response to the innovation in money growth. Under certain parameterizations, either model may deliver a liquidity effect. Such parameterizations often involve imposing restrictions on quantity responses, either through inflexible production—as in the recent example of Christiano and Eichenbaum (1995)— or through the use of substantial curvature of the transactions benefits associated with real balances, combined with a limited response of interest-sensitive components of spending such as investment. In these frameworks, the response of money-demand to interest rates is a key model parameter that helps determine the liquidity effect. The second strand of this literature using identified VARs to estimate the liquidity effect in time series data also has mixed success, with results depending on specification, sample period, and methods of identification. The problem here is that the assumptions required to identify monetary policy innovations and, hence, the liquidity effect in this literature are not necessarily valid; for example, the specifications considered by Bernanke and Blinder (1992) and Christiano and Eichenbaum (1992) both deliver liquidity effects by imposing the assumption that monetary policy does not have contemporaneous effects on output or prices. Avoiding these types of restrictions motivates Hamilton’s use of daily data and his search for a valid instrument. Let me discuss this methodology by considering a simple version of Thornton’s model for the daily federal funds market.