Abstract
The literature on fixed exchange-rate regimes has long recognized that pegs are inherently impermanent, and that the time until a peg is realigned or abandoned depends on observable macroeconomic fundamentals — reserve adequacy, real exchange-rate misalignment, fiscal discipline — as well as on structural and political factors that are harder to monitor systematically but that the duration-analysis literature has shown to matter, including terms-of-trade shocks and executive political transitions (Klein and Marion, 1997).
This paper formalizes peg sustainability as a Cox (1972) proportional-hazards survival problem, in the same statistical tradition used to model corporate default under incomplete information (Obstfeld and Rogoff, 1995), and partitions the covariate vector into a monitored block (reserve coverage, real effective exchange-rate misalignment, and the fiscal balance — the variables a central bank or an IMF Article IV mission routinely tracks) and a structural/shock block (terms-of-trade shocks, political transition intensity, and regional monetary-union contagion pressure — factors the duration literature flags as important but that are not part of routine peg-sustainability monitoring in every currency union).
Motivated by the institutional history of the CFA franc zone, whose two monetary unions have maintained a fixed external peg since 1945 with a single 50% realignment in January 1994 (International Monetary Fund, 1994), we study the properties of this hazard-based framework under a fully specified, parametrically known data-generating process rather than fitting it to a real cross-country panel, because no de-identified, spell-level dataset of peg durations with verified realignment timing and both covariate blocks was available to us for this study.
Across 1,500 simulated 30-year peg spells, adding the structural/shock block raises the concordance index from 0.641 to 0.693 and is jointly significant against the monitored-only model (likelihood-ratio χ²₃ = 91.2, p < 10⁻¹⁶); the discrimination gain scales with the assumed strength of the structural signal, degrades gracefully rather than catastrophically as structural-data coverage becomes incomplete, and the maximum-likelihood estimator shows correct confidence-interval coverage down to samples of 100 peg-spells — a sample size on the order of what a real historical cross-country panel of fixed-exchange-rate episodes could plausibly provide. We interpret these as necessary, not sufficient, conditions for using hazard-based peg-sustainability monitoring in practice, and set out explicitly what an empirical study using a real historical panel of exchange-rate regimes would still need to show.