VoxEU Column COVID-19 Financial Markets Global crisis

On the legacy of financial crises: Lessons from Mexico’s sudden stop history

Borrowing constraints can amplify business cycle dynamics and create significant challenges for countries facing negative economic shocks. Based on a new estimated model of sudden stop crises, this column argues that financial crises are often followed by a quick but partial rebound. Thereafter, economies can experience a very protracted period of stagnation. A similar trajectory may be likely for the current COVID-19 crisis, as many countries will face financial frictions in responding to the economic downturn.

There is consensus that the ongoing COVID-19 pandemic will have a deep negative impact on economic activity (e.g. Barro et al. 2020, Correia et al. 2020). A critical question on the minds of policymakers and market participants is how long it will take for the world economy to recover from the shock once the epidemic is brought under control (Reinhart and Rogoff 2020). While the markets currently expect a quick rebound in economic activity, research on the economic impact of epidemics points to potentially long-lasting negative effects (Ambrus et al. 2020).

A new model of sudden stop crises

In Benigno et al. (2020), we estimate a new model of business cycles and crises with occasionally binding frictions. In the paper, we focus on one particular type of crisis, the so-called sudden stop in international capital flows. This is captured by a binding constraint on the private sector’s ability to borrow from abroad, and we estimate it using data for Mexico since 1981. The framework, however, has general applicability and is a useful laboratory to evaluate the economic consequences of crises.

Occasionally binding borrowing constraints are mechanisms that amplify regular business cycle dynamics. For example, in the case of the COVID-19 crisis, which did not originate in the financial sector, suddenly binding financial frictions in capital markets powerfully amplified the initial impulse, triggering massive policy interventions to counterbalance the impact of the shock.

Most importantly, our new sudden stop model identifies crisis episodes of varying duration and intensity. This finding, which has previously proven difficult to capture in structural models with occasionally binding constraints, is consistent with empirical evidence of large economic dislocation during financial crises and the long-lasting build-up and sluggish recovery phases surrounding them (see Cerra and Saxena 2008, Reinhart and Rogoff 2009, and Boissay et al. 2016).

The estimation results yield a time-varying probability of facing a binding borrowing constraint. The model, therefore, identifies crises episodes as consecutive periods of constrained borrowing.

Figure 1 plots this probability for Mexico (solid black line), together with the peaks of the crisis episodes previously identified in the quantitative sudden stop literature (red bars), corresponding with the troughs in output and capital flows in the data. To illustrate the model’s performance, we also report a purely empirical notion of crisis (grey bars). This indicator is the crisis tally index of Reinhart and Rogoff (2009).1

Estimation results

The estimated model identifies three sudden stop episodes in the recent economic history of Mexico corresponding to periods in which the crisis probability remains above 90% (start and end-quarters marked by vertical dashed lines). The first is the debt crisis, lasting eight quarters, during 1981:Q3-1983:Q2, with the peak in 1983:Q1. The second is the so-called ‘tequila crisis’, lasting nine quarters from 1994:Q1 to 1996:Q1, with its peak in 1995:Q1-Q2. The last one is the Global Crisis that, according to the model, produced a crisis in Mexico from 2008:Q4 to 2009:Q3, for four quarters, with a peak in 2009:Q1-Q2.   

Figure 1 Mexico's model-identified crisis episodes

Notes: The black line is the estimated model implied probability of being in a crisis. The grey bars correspond to the tally index. The red bars indicate model-identified crisis peaks. The vertical dash lines mark the beginning and the end of the estimated crisis episodes.

Figure 1 shows that our estimated model tracks the crisis tally index remarkably well. The crisis peaks are close to the highest values of the tally index. The actual crisis episodes are much more persistent than the crisis peaks previously identified in the literature, consistent with the idea that our new formulation of the typical sudden stop model better characterises how borrowing limits apply to households and firms in the data. The model does particularly well at tracking the consequences of the Global Crisis for Mexico. Importantly, in the paper, we also show that the estimated model does not mistake ordinary recessions – not associated with spikes in the tally index – for crisis periods.

Model-based sudden stop episodes: Duration and dynamics

To draw lessons from Mexico’s experience for the ongoing COVID-19 crisis, we can look at model-simulated crises episodes.

Looking first at the duration of simulated crisis episodes, Figure 2 reports the frequency of sudden stop episodes lasting longer than four quarters, like the shortest of the three estimated crisis episodes, the Global Crisis. Specifically, it shows a histogram of the model-implied conditional crisis duration.

Figure 2 Model-simulated crisis episodes duration

The estimated model generates substantial heterogeneity in crisis duration. The average conditional crisis duration is five consecutive quarters, with some episodes lasting more than 20 quarters, even though they are rare events making up less than 0.5% of all episodes.

Next, considering the other two episodes, the debt and tequila crises, which were more severe than the Global Crisis, we look at the crisis dynamics for episodes of eight consecutive quarters. Figure 3 plots the simulated model dynamics for selected shocks and endogenous variables, five years before and ten years after the event.

Figure 3 illustrates the distinctive combinations of shocks that drive the economy before, during, and after the crisis episode. Crisis episodes are preceded by a long-lasting ‘boom’ phase, driven by improving technology and a favourable international environment, with a notable fall in borrowing costs and compressed country spreads. These forces drive the expansion gradually, with increasing output, consumption, and investment, in a manner consistent with empirical characterisations of the boom phases of financial crises (Boissay et al. 2016).

The economy enters the crisis episode at t=0, after a final acceleration. The large crash is precipitated by a combination of adverse supply and demand shocks – a sudden drop in technology, an increase in patience, and a reversal in the cost of borrowing. During the crisis episode, borrowing costs and patience continue to increase, technology stagnates, and the country spread remains wide open. The constraint on borrowing limits consumption smoothing and curtails the output supply further through limiting working capital. This causes output, consumption (not reported), and investment to drop sharply. The output drop from peak to trough is eight percentage points, in line with what observed during the tequila crisis in Mexico. Credit flows (measured by the trade-balance-to-output ratio in Figure 3) suddenly dry up (reverts), remaining at this contracted level throughout the crisis phase. The credit contraction is about six percentage points as a share of output from trough to peak.

Figure 3 Simulated sudden stop dynamics

Notes: The figure plots model-simulated dynamics during crisis episodes of eight quarters, five years (20 quarters) before the crisis, and ten years after the crisis (40 quarters). The economy is in the constrained regime from period t=0 to period t=7 (vertical dashed lines). The plotted dynamics are medians across all crisis episodes identified, in log-levels, setting t-20=0.

The economy rebounds quickly from these episodes, but only partially, making up only half of the ground lost during the crisis episode, or about four percentage points in these simulations. After the initial bounce, a combination of persistently adverse circumstances produces a protracted output decline, as we can see in the Mexican data after the debt crisis (not reported), and also in line with empirical evidence on the long-term consequences of financial crises in other emerging and advanced economies (Cerra and Saxena 2008, Reinhart and Rogoff 2009). The international cost of borrowing remains elevated for an extended period, even though spreads revert after the crisis. The productivity decline is sizable and very long-lasting, with technology reaching a level that is below the one at the beginning of the boom. During the post-crisis period, investment and to a lesser extent consumption also stagnate below their pre-crisis levels (Benigno and Fornaro 2017). As a result, credit flows (the level of the trade balance in Figure 3) remain above (below) their pre-crisis level long after the crisis has ended, although the economy is no longer financially constrained.


Our new estimated model of business cycles and sudden stop crises, which fits Mexico’s experience of severe crises episodes well, suggests that these episodes can have a very long-lasting negative economic impact. When crises end, a rebound in economic activity can occur quickly but might be only partial. Our model indicates that protracted periods of stagnation, such as those experienced after the Global Crisis in the US, are relatively common occurrences, and that economic recovery from crises amplified by financial constraints can be slow and anaemic.

Authors’ Note: The views expressed are solely those of the authors and do not necessarily reflect the views of the Federal Reserve Banks of New York, San Francisco, or St. Louis, or the Federal Reserve System.


Ambrus, A, E Field and R Gonzalez (2020), "Loss in the Time of Cholera: Long-Run Impact of a Disease Epidemic on the Urban Landscape", American Economic Review 110(2): 475-525.

Barro, R, J Ursua and J Weng (2020), "Coronavirus Meets the Great Influenza Pandemic",, 20 March.

Benigno, G, A Foerster, C Otrok and A Rebucci (2020), “Estimating Macroeconomic Models of Financial Crises: An Endogenous Regime Switching Approach”, CEPR DP No. 14545.

Benigno, G and L Fornaro (2017), "Stagnation Traps", Review of Economic Studies 85(3): 1425-1470.

Boissay, F, F Collard and F Smets (2016), "Booms and Banking Crises", Journal of Political Economy 124(2): 489-538.

Cerra, V and S C Saxena (2008), "Growth Dynamics: The Myth of Economic Recovery", American Economic Review 98(1): 439-57.

Correia, S, S Luck and E Verner (2020), Pandemics Depress the Economy, Public Health Interventions Do Not: Evidence from the 1918 Flu, manuscript, MIT.

Reinhart, C M and K Rogoff (2020), "The Coronavirus Debt Threat", The Wall Street Journal, 26 March.

Reinhart, C M and K S Rogoff (2009), This Time Is Different: Eight Centuries of Financial Folly, Princeton University Press.


1 The crisis tally index of Reinhart and Rogoff (2009) ranges from zero to six, depending on whether a country-year observation is deemed to be in one or more of the following six varieties of crisis, assigning the value of one if a variety is present: currency, inflation, stock market, sovereign domestic or external debt, and banking crisis. In Figure 1, the index is normalized to range between zero and one.

1,050 Reads