The Predictive Failure of the Baba, Hendry and Starr Model of the Demand for M1 in the United States

Baba, Hendry and Starr (1992) attempt to restore a stable specification for the demand for M1 balances in the United States. Estimating an error correction model that simultaneously models the long-run and short-run determinants of money demand, they introduce two new explanatory factors of the demand for M1. First, they provide theoretical and empirical evidence that the volatility of long term interest rates affects the demand for M1. Second, they introduce an adjusted set of interest rates that allow for learning behavior in response to the introduction of various new accounts for components of M1 and non-M1 M2s. Most importantly, they show that their specification passes a large battery of diagnostic tests and performs well during hard-to-model episodes such as the 'Missing Money Period.' This paper replicates their results and extends their sample range to evaluate its predictive performance. Their model breaks down completely over the longer sample period. This predictive failure could have been anticipated to someextent in that 'sensitivity' analysis might have been helpful in foreshadowing the model's breakdown. It appears that their specification underestimated the interest rate elasticity of money demand in part because of the learning- adjustment mechanism. We also provide results that call into question their use of volatility ad the model's over- reliance on movements in volatility