Work

Money, Interest Rates and Inflation in Mexico, 1985–2025: Extreme Bounds Analysis Within Clusters of Policy-Homogeneous Years

Year: 2026
Type: preprint
Abstract: This paper estimates the Mexican money–inflation relationship three ways on annual data for 1985–2025. Log inflation y is the dependent variable, log broad-money growth x is the focus variable, and six doubtful variables — real output growth, unemployment, peso depreciation, terms-of-trade growth, the public sector borrowing requirement and the log policy interest rate — generate a model space of K = 2⁶ − 1 = 63 specifications. The first exercise estimates that space over the whole sample with money growth entered contemporaneously; the second repeats it with money growth lagged one year; the third partitions the sample into three clusters of policy-homogeneous years and re-estimates within each. The first two exercises agree completely and appear conclusive. Every one of the 63 specifications is positive and significant at the 5% level in both, so the robustness share Θ equals 1.000, and no estimate is wrong-signed. The mean coefficient is 0.618 contemporaneously and 0.552 at one lag, with Leamer extreme bounds of [0.005, 1.261] and [0.043, 1.238]. Both criteria return robust. On the conventions used throughout this research programme the Mexican relationship would be recorded as settled. The clusters show that it is not. Over 1985–1994, when the peso was anchored and Banco de México was not autonomous, the share is 0.143 and the bounds are [−0.729, 1.548]. Over 1995–2009, spanning the float and the transition to targeting, the share is 1.000 with a mean of 0.886. Over 2010–2025, under mature inflation targeting, the share is 0.000, the mean coefficient is −0.094, one of seven estimates is wrong-signed and adjusted R² falls to between 0.09 and 0.26. The pooled result is produced almost entirely by the middle cluster. Two methodological points follow. A pooled extreme bounds exercise reports an average across regimes that may correspond to no regime that existed, and the Mexican pooled coefficient of 0.618 lies between a cluster estimate of 0.886 and one of −0.094. And the model space must shrink with the sample: enumerating 63 specifications on a ten-year cluster would leave two residual degrees of freedom in the largest models, so the cluster exercise uses a three-variable doubtful set and K = 7. Section 5.3 sets out the arithmetic.
Language: English
Open Access status: green
OpenAlex ID: W7236740947
Cites: 17
Cited by:
Related to: 10