Persistence, duration and transition dynamics between stability and exchange-rate stress regimes

Abstract

Persistence, duration and transition dynamics between stability and exchange-rate stress regimes in Bolivia's parallel dollar market (2024–2026): evidence from an autoregressive Markov-switching model

Since February 2023 the Bolivian economy has operated under a de facto multiple exchange rate system: an official rate pegged at BOB 6.96 per US dollar since 2011, and a parallel market that, facing foreign currency rationing, migrated towards trading the USDT stablecoin. This paper characterises the persistence, duration and transition dynamics between calm and stress phases in that market using a two-state Markov-switching autoregressive model, MS-AR(10), with regime-dependent variance, estimated by maximum likelihood on 462 daily returns of the USDT/BOB exchange rate between July 2024 and June 2026.

The results identify two regimes that differ fundamentally in volatility rather than in trend: a speculative regime with 68 times higher variance and a mean duration of 4.77 trading days, and a calm regime with a mean duration of 11.25 trading days and a stationary probability of 70.2%. The resulting dynamics are cyclical rather than monotonically deteriorating: episodes of exchange rate pressure are intense but transitory and are followed by normalisation, a pattern consistent with a demand for digital dollars driven by temporary hedging against confidence shocks rather than by expectations of sustained depreciation.

Out-of-sample evaluation on a 70/30 temporal split shows that the model substantially outperforms naive, constant-variance AR(10) and GARCH(1,1) specifications in predictive log-score (−1.37 versus −1.88 for the best competitor), and that its one-day Value at Risk is rejected neither by Kupiec coverage nor by Christoffersen independence tests. The one-day-ahead probability of stress discriminates well between stress and calm days in the test period (ROC-AUC of 0.86) and, with the threshold calibrated on the training set, flags 21 of the 23 stress days, albeit with a precision of 37.5%, which delimits the model’s scope: probabilistic identification, early warning and risk quantification rather than exact prediction of the timing of regime breaks.

Keywords: parallel exchange market, Markov regime switching, stablecoins, Bolivia, Value at Risk, out-of-sample evaluation.

Key results

68×

higher variance in the speculative regime than in the calm regime

4.77 trading days

mean duration of a stress episode

11.25 trading days

mean duration of a calm spell

70.2%

long-run share of time in the calm regime

−1.37 vs −1.88

out-of-sample predictive log-score, against the best competing model

0.86

out-of-sample ROC-AUC for anticipating the stress state one day ahead