Extending the approach of Bernanke and Blinder (1992), Strongin (1992), and Christiano, Eichenbaum, and Evans (1994a, 1994b), we develop and apply a VAR-based methodology for measuring the stance of monetary policy. More specifically, we develop a "semi-structural" VAR approach, which extracts information about monetary policy from data on bank reserves and the federal funds rate but leaves the relationships among the macroeconomic variables in the system unrestricted. The methodology nests earlier VAR-based measures and can be used to compare and evaluate these indicators. It can also be used to construct measures of the stance of policy that optimally incorporate estimates of the Fed's operating procedure for any given period. Among existing approaches, we find that innovations to the federal funds rate (Bernanke-Blinder) are a good measure of policy innovations during the periods 1965-79 and 1988-94; for the period 1979-94 as a whole, innovations to the component of nonborrowed reserves that is orthogonal to total reserves (Strongin) seems to be the best choice. We develop a new measure of policy stance that conforms well to qualitative indicators of policy such as the Boschen-Mills (1991) index. Innovations to our measure lead to reasonable and precisely estimated dynamic responses by variables such as real GDP and the GDP deflator.
