Revisiting empirical studies on the liquidity effect: An identication-robust approach

The liquidity effect, the short run negative response of interest rates to an increase in the money supply, has been the subject of a large number of studies, most of which based on the estimation of structural vector autoregressive models using standard instrumental variable methods (see e.g. Gali, 1992, Quarterly Journal of Economics). Using data from both the United States and Australia, we show that these SVAR models are weakly identified, and therefore the standard IV estimates of the structural coefficients and impulse response functions are biased and inconsistent. We use statistical procedures robust to weak instruments, along with the projection method of Dufour and Taamouti (2005, Econometrica), to construct confidence sets with correct coverage rate for the structural parameters and impact response functions of Gali's four variable IS-LM SVAR model. We find that these confidence sets are in general unbounded or large, and further, contain zero, thus suggesting that the evidence of the liquidity effect found in previous studies is empirically fragile. Our findings align with Pagan and Robertson (1998, Review of Economics and Statistics) who first pointed out possible identification issues in SVAR models.