Two centuries of systemic bank runs

Bank runs are notoriously difficult to measure systematically. This column constructs two novel, cross-country databases on bank run events in 184 countries over 1800-2023 and bank sector outstanding deposits. It shows that the costs of systemic bank runs are substantial – real GDP is on average 9% below its pre-run trend. These output losses are observed both in cases where runs are triggered by fundamental factors, such as a monetary policy shock, and non-fundamental factors. Policymakers can dampen the macroeconomic costs with liability guarantees and reduce the likelihood of runs becoming systemic with credible deposit insurance and a lender of last resort.

Since the seminal work of Diamond and Dybvig (1983), bank run risk has remained a topic of critical importance for both academics and practitioners over the past four decades. The 2023 depositor run and failure of Silicon Valley Bank has further fuelled the interest among researchers and policymakers in understanding the economic and regulatory implications of banking panics (Admati et al. 2023, Beck et al. 2024). However, bank runs are notoriously difficult to measure systematically. Moreover, there is limited evidence on their incidence and implications across time and space.

In a recent paper (Jamilov et al. 2024), we make progress by constructing and making publicly available two novel, comprehensive cross-country databases. Data and additional materials are available at www.systemicbankruns.com. First, we put forth new narrative evidence on bank runs that includes 308 events in 184 countries over 1800-2023 and from 463 sources. Second, we assemble and harmonise a novel dataset of banking sectors’ outstanding deposits, allowing us to measure episodes of significant deposit withdrawals. The synergy of our qualitative, narrative-based identification of bank runs with a quantitative, statistical indicator of depositor withdrawals provides an unprecedented advancement in the measurement of bank run risk.

Table 1 provides a detailed comparison of our chronology with four canonical databases of banking crises: Baron et al. (2021), Laeven and Valencia (2018), Jorda et al. (2017), and Reinhart and Rogoff (2009).

Table 1 Comparison of the our bank run chronology with other datasets

Note: This table compares the coverage of the bank run chronology introduced in our paper with existing lists of banking crises. ‘Narrative only’ refers to chronologies that do not incorporate statistical indicators to identify crisis episodes, such as deposit growth rates (this paper) and bank equity returns (Baron et al. 2021).
Source: Jamilov et al. (2024).

How frequent are bank runs? The long view

Have bank runs become more or less common over time? The probability of bank runs is a key object for economic models and policy analysis (Goldstein and Pauzner 2005). Figure 1 visualises the frequency of bank run episodes over the past 200 years. We also plot the frequency of banking crises using our own unified measure of crises that combines the four aforementioned crises lists. The average probability of bank runs across all countries and years is 1.9%. The frequency was trending upwards until WWII, declining during the post-war period, and rising again in the 1980s. These numbers provide evidence that bank runs have been occurring regularly during modern history and all around the globe.

Figure 1 Frequency of bank runs and banking crises

Note: The figures plot the frequencies of bank runs for which we have narrative evidence and a unified banking crises measure from 1800 to 2023. The annual frequency of runs and crises is visualised as a moving average, using a ten-year time window. The frequency in each individual year is calculated as the total number of runs/crises divided by the number of countries for which the authors have data coverage in that specific year.
Source: Jamilov et al. (2024).

Systemic bank runs are associated with large output losses

Should policymakers care about the aftermath of bank runs? While some (singular) bank run incidents could be idiosyncratic, others may lead to contagion effects and systemic liquidity disruptions (Allen and Gale 2000, Diamond and Rajan 2005, Uhlig 2010). We define a narrative bank run as a systemic event if we have objective, statistical evidence that it is accompanied by a contraction in the banking sector’s total deposit base. Thus, systemic runs are measured as the intersection of narrative runs and deposit withdrawal events.

In Figure 2, we show that the macroeconomic costs from systemic bank runs are substantial: five years after a systemic bank run, real deposits contract significantly and real GDP is on average 9% lower relative to its pre-run trend. The economic channel behind this pattern is the severe, roughly 30% contraction in outstanding credit, which is consistent with a critical role for the banking system in providing credit to the real economy (Bernanke and Gertler 1995). Crucially, bank runs are only damaging to the macroeconomy when they are systemic, i.e. accompanied by a decline in aggregate deposits. This finding suggests an important role for the liability side of bank balance sheets in economic downturns, confirming the notion that runs become particularly relevant from a macroeconomic perspective when they are systemic in nature.

Figure 2 Output losses from bank runs and deposit contractions

Note: The figures plot the local projection impulse response function of real GDP growth rates, real deposit growth rates, real credit growth rates, and changes in deposits-to-GDP for (i) systemic bank runs (blue line), (ii) non-systemic bank runs (red line), and (iii) deposit contractions outside of run episodes (green line). The error bands depict 95% confidence bands based on standard errors double-clustered by country and year.
Source: Jamilov et al. (2024).

Runs are harmful even when not triggered by fundamentals or solvency issues

What is the nature of bank runs? A large theoretical literature models runs as (behavioural) sunspot phenomena (Diamond and Dybvig 1983, Gertler and Kiyotaki 2015). The second important view is that runs are driven by poor macroeconomic fundamentals (Bryant 1980, Allen and Gale 1998). The third class of frameworks allows for both fundamental and non-fundamental causes of panics (Goldstein and Pauzner 2005). As such, the question of whether runs are triggered by fundamentals, the sunspot residual, or both, is a central issue in the literature (Goldstein 2013).

We adopt a narrative approach and classify every run in our dataset based on whether it was triggered by fundamental (e.g. a monetary policy shock or exchange rate devaluation) or non-fundamental causes. Our approach identifies some 55 systemic runs as non-fundamental. In Figure 3, we show that the aggregate output losses from these non-fundamental systemic runs are very similar to those from fundamental ones. This suggests that the underlying fundamental factors are not the sole determinants of GDP contractions following systemic bank runs.

Figure 3 The output losses of non-fundamental systemic bank runs

Note: The figures plot the local projection impulse response function of real GDP, real deposits, real credit, and deposits-to-GDP for (i) systemic runs (blue line) and (ii) non-fundamental-based systemic runs (red line). We define systemic runs not based on fundamentals as those for which we find no mentions of macroeconomic causes such as business cycle downturns, monetary policy decisions, capital inflows, or credit booms in narrative accounts of the bank run events in our chronology. The error bands depict 95% confidence bands based on standard errors double-clustered by country and year.
Source: Jamilov et al. (2024).

In addition, we find that systemic runs are more harmful when coinciding with (i) episodes of widespread bank failures or (ii) episodes of banking crises, or (iii) when the banking sector is poorly capitalised. However, systemic bank runs are associated with significant output losses even when they do not coincide with either (i), (ii), or (iii). We therefore provide multi-faceted evidence that weak fundamentals are sufficient but not necessary for systemic bank runs to have negative macroeconomic ramifications.

Depositors run on highly leveraged banks with poor solvency

To understand the triggers and consequences of bank runs on a granular level, we supplement our cross-country chronology with detailed US bank-level data over 1867-1904 from Carlson et al. (2022) and over 1976-2020 from the Call Reports. We show that depositors run on highly leveraged institutions, which is in line with the solvency view of banking crises (Correira et al. 2024). We document that besides banks with high ex-ante leverage, those with low profitability or high reliance on deposit funding experience larger deposit outflows in both samples. By comparing banks with and without deposit outflows during bank runs using an event study approach, we show that affected banks cannot make up for the loss of deposits by raising other types of funding, even in modern times where banks rely more on non-deposit funding sources such as interbank borrowing. As a result, they see a reduction in their leverage ratios and a contraction of their balance sheets and loan books.

Regulators can dampen bank runs with liability guarantees; prevention is harder

Can government policies prevent systemic bank runs and/or mitigate their macroeconomic damage? Using the data from existing databases of government interventions and institutional setups in Demirguc-Kunt et al. (2014), Laeven and Valencia (2018), and Metrick and Schmelzing (2021), we find that liability guarantees are an effective form of ex-post government intervention and can almost halve the long-run output losses from systemic runs. In addition, the existence of credible deposit insurance schemes and of a lender of last resort can reduce the likelihood of a run becoming systemic ex ante, but to a limited extent conditional on the run already occurring.

Conclusion

Our research provides a novel, comprehensive chronology of bank runs and aggregate deposit contractions around the world over two centuries. Bank runs affect the macroeconomy only if they are accompanied by deposit withdrawals, i.e. they are systemic in nature. Contributing to an important debate in the literature, we show that systemic bank runs can be contractionary even if they are not triggered by deteriorating fundamentals, bank capitalisation issues, or widespread banking failures. A granular look at bank-level data reveals that depositors do not run on banks indiscriminately but, instead, target highly leveraged institutions, which in turn leads to a reallocation of deposits and a credit crunch. Finally, policymakers can tame the macroeconomic effects of bank runs by means of liability guarantees and reduce the likelihood of runs turning systemic with credible deposit insurance and a lender of last resort.

Source : VOXeu