On 9 March 2023, Silicon Valley Bank lost $42 billion in deposits in a single day, the largest bank run in American history. The collapse, which followed a rapid rise in interest rates and a concentrated depositor base, was not a novel phenomenon. It was a modern replay of the fragility that Douglas Diamond and Philip Dybvig captured in their 1983 paper, a model that explains why banks exist, why they are inherently vulnerable to panics, and why government intervention can make them stable. The Diamond-Dybvig model gives analytical form to a centuries-old insight: banks transform short-term, liquid deposits into long-term, illiquid loans, and this maturity transformation creates a beneficial social function but also opens the door to self-fulfilling runs. The model has shaped deposit insurance design, central bank lender-of-last-resort policies, and financial regulation for four decades.
The model’s core contribution is to show that bank runs need not be driven by fundamental insolvency. A run can occur simply because depositors fear that other depositors will run, and each individual’s rational response is to join the queue. This multiple-equilibrium structure sets Diamond-Dybvig apart from purely fundamentals-based views of banking crises. It explains why solvent banks can fail in a panic, and it gives a clear welfare justification for deposit insurance and emergency liquidity provision. After the crises of 2008 and 2023, the model has returned to the centre of policy debate about how to make the financial system resilient without creating excessive moral hazard.
The Logic of Bank Runs
Banks perform two essential functions. They pool deposits from many households, smoothing the idiosyncratic liquidity needs that arise from uncertainty about when consumption will be needed. And they invest those deposits in longer-maturity assets, such as business loans and mortgages, that yield a higher return than if each household held only short-term, liquid securities. This transformation of maturity creates value: it channels savings into productive investment and allows depositors to share the risk of needing funds early. The cost is that the bank’s balance sheet becomes fragile. If all depositors demand their money back at once, the bank cannot honour the demands because its assets cannot be liquidated fast enough without fire-sale losses.
The Diamond-Dybvig model formalises this tradeoff. In the model, agents have an endowment at date 0 and face uncertainty about when they will want to consume: some are “impatient” and must consume at date 1, while others are “patient” and can wait until date 2. Without a bank, each agent must self-insure by holding a mix of liquid and illiquid assets. The bank improves on this arrangement by offering deposit contracts that give a higher return to those who wait, while still providing a reasonable return to those who withdraw early. The deposit contract is a form of insurance, and it Pareto-dominates the autarkic outcome. But the contract has a dark side: it creates a coordination problem. If enough depositors withdraw early, the bank must liquidate its long-term assets prematurely, incurring losses that make it impossible to repay all depositors in full. A depositor who believes others will run has an incentive to run as well, because the bank’s promised payments are feasible only if the run does not occur. This strategic complementarity gives rise to two equilibria: a good equilibrium in which only the genuinely impatient withdraw early, and a bad equilibrium, the bank run, in which everyone withdraws and the bank collapses.
The model’s genius is to show that the bad equilibrium is not a failure of rationality. It is consistent with full information and rational expectations. The run is a sunspot phenomenon, triggered by any event that coordinates depositors’ beliefs. That event could be a rumour, a newspaper headline, or the failure of a neighbouring bank. The corollary is that deposit insurance, which guarantees that depositors will be repaid regardless, can eliminate the bad equilibrium by removing the incentive to run. With insurance, a depositor who believes others are running has no reason to join them, so the run equilibrium disappears. This logic is why deposit insurance is not a bailout but a pre-emptive stabilisation device.
This framework connects directly to the wider issue of information asymmetry in financial markets. When depositors cannot distinguish a temporarily illiquid bank from a genuinely insolvent one, even an unfounded fear can become a self-fulfilling prophecy. The model’s structure also underlies the rationale for the lender of last resort: a central bank that stands ready to lend against good collateral can halt a run without the need for disruptive asset liquidations. Additionally, the insurance mechanism at the heart of the model echoes themes from the broader analysis of risk and insurance in economics, reinforcing the idea that pooling risks, when properly structured, yields large welfare gains.
Diamond‑Dybvig Model in Equations
The economy consists of a continuum of ex ante identical agents, each endowed with one unit of a good at date \( t = 0 \). There are three periods, indexed 0, 1, and 2. At date 1, each agent privately learns her type: with probability \( t \) she is impatient and derives utility only from consumption at date 1; with probability \( 1-t \) she is patient and derives utility only from consumption at date 2. Type realisations are independent, so the fraction of impatient agents in the population is exactly \( t \). Utility is given by \( u(c) \), where \( u \) is strictly increasing, strictly concave, and satisfies the Inada conditions.
Two assets are available. A short-term, liquid asset (storage) that returns one unit at date \( t+1 \) for each unit invested at date \( t \). A long-term, illiquid asset that returns \( R > 1 \) at date 2 for each unit invested at date 0, but if liquidated at date 1, yields only \( L < 1 \) per unit. \( R \) represents the productive return on capital; \( L \) captures the fire-sale loss from interrupting the investment. The bank pools deposits and offers a demand-deposit contract that gives the depositor who withdraws at date 1 a fixed amount \( r_1 > 1 \), provided the bank still has resources, and divides the remaining assets among date-2 withdrawers.
The bank’s balance sheet is determined as follows. It invests a fraction \( x \) of total deposits in the long-term asset and \( 1-x \) in storage. To meet date-1 withdrawals, it first uses its storage and then liquidates long-term assets if necessary. Let \( f \) be the fraction of depositors who withdraw at date 1. If \( f \leq (1-x)/r_1 \), the bank can meet all withdrawal requests without touching the long-term asset. If \( f \) exceeds that threshold, the bank must liquidate some of its long-term investments. The condition for the bank to be able to pay \( r_1 \) to all withdrawing depositors is:
The left side is the total amount demanded at date 1; the right side is the maximum resources the bank can raise by liquidating all assets. A run equilibrium exists if, when all patient depositors attempt to withdraw (\( f = 1 \)), the bank cannot honour the full claim, i.e., if \( r_1 > (1-x) + Lx \). In that case, any patient depositor who believes others will run finds it optimal to run as well because waiting would leave her with nothing after the bank’s resources are exhausted. Formally, let \( c_2 \) be the consumption available to a patient depositor who waits when fraction \( f \) of depositors withdraw early. The bank’s allocation rule is:
where \( \hat{f} = (1-x)/r_1 \) is the maximum withdrawal fraction that avoids liquidation. The patient depositor’s decision depends on the expected value of \( c_2 \) relative to \( r_1 \). In the good equilibrium, only impatient depositors withdraw (\( f = t \)), and \( c_1 = r_1 \), \( c_2 = R(1 – r_1 t)/(1-t) \). The bank sets \( r_1 \) to provide the optimal risk-sharing. In the bad equilibrium, all depositors withdraw (\( f = 1 \)), each receiving at most \( (1-x) + Lx \), which is below \( r_1 \) by construction, and the bank fails.
The key variables are summarised below.
| Variable | Definition | Role in the Model |
|---|---|---|
| \( t \) | Fraction of impatient depositors | Determines liquidity demand at date 1 in the good equilibrium |
| \( R \) | Long-term asset return | Productive return on illiquid investment (>1) |
| \( L \) | Liquidation value of long-term asset | Fire-sale return if asset is sold at date 1 (<1) |
| \( r_1 \) | Fixed payment to early withdrawers | Optimal insurance contract payment (>1) |
| \( x \) | Fraction of deposits invested in long-term asset | Bank’s portfolio choice |
| \( f \) | Fraction of depositors withdrawing at date 1 | Strategic variable; \( f = t \) in good equilibrium, \( f = 1 \) in a run |
 | ||
Variable definitions for the Diamond-Dybvig model. The deposit contract \( r_1 \) is chosen by the bank to provide insurance, but it also creates the run equilibrium if \( r_1 \) exceeds the liquidation value of total assets per depositor.
The model yields two policy solutions. First, suspension of convertibility: if the bank can limit withdrawals once a threshold is reached, depositors know they cannot all withdraw, and the run equilibrium is eliminated. This was the historical practice of US banks before deposit insurance. Second, deposit insurance: a government guarantee that depositors will be repaid, which restores confidence outright. Both solutions work by breaking the strategic complementarity. Deposit insurance is the more complete solution because it does not force impatient depositors to bear liquidity risk.
Key Assumptions and Model Boundaries
The Diamond-Dybvig model’s powerful insights rest on specific assumptions that limit its scope. The most important assumption is that the bank run is a pure sunspot event unrelated to the bank’s underlying asset quality. In the model, the long-term asset always yields \( R \) in period 2; the only problem is premature liquidation. This abstracts from the reality that many bank runs are triggered by genuine concerns about solvency or asset quality. Empirical work has consistently found that both fundamental-based runs and pure panic runs occur, and distinguishing between them has first-order policy implications. A deposit insurance scheme that rescues insolvent institutions creates moral hazard, while one that supports merely illiquid ones is welfare-enhancing. The 2008 financial crisis demonstrated that when asset quality deteriorates across the system, the distinction between illiquidity and insolvency can blur rapidly.
Second, the model assumes that depositors are atomistic and do not coordinate their actions except through the expected behaviour of others. In practice, large institutional depositors, such as the corporate treasuries that held balances at Silicon Valley Bank, can coordinate withdrawals explicitly, accelerating the dynamics beyond what the simple model predicts. The SVB run was not a random sunspot; it was triggered by a disclosed loss on securities sales and amplified by a highly networked depositor base that communicated in real time through social media. This does not invalidate the model’s logic, but it suggests that the speed and information environment of modern runs may require policy responses that go beyond traditional deposit insurance.
Third, the model abstracts from interbank markets, central bank lending, and the possibility of raising new capital. In reality, a bank facing withdrawals can borrow from the interbank market, from the central bank’s discount window, or issue new equity. If these channels are open and the bank is solvent, the run may be contained before it becomes self-fulfilling. The collapse of Northern Rock in 2007 illustrated the limits of these channels under stress; the Bank of England’s role as lender of last resort was invoked too late to prevent the first UK bank run in over a century.
Fourth, the Diamond-Dybvig contract treats depositors as identical ex ante, differing only in their realised liquidity needs. In reality, depositors differ in wealth, risk aversion, and access to information. This heterogeneity can alter the equilibrium structure and affect the efficiency of deposit insurance. A further limitation is that the model is static: it does not capture how the duration of a run evolves or how the bank’s asset portfolio responds dynamically to the withdrawal shock. Still, the core message remains robust: the bank’s dual role as liquidity provider and maturity transformer creates a coordination vulnerability that government intervention can resolve. The assumptions that matter most for policy are those about information and the nature of the shock. As the Akerlof model of information asymmetry also demonstrates, the presence of asymmetric information can fundamentally alter market outcomes and create a role for institutional responses that restore efficiency.
Evidence from Historical Bank Runs
The Diamond-Dybvig framework has been tested against data from historical and modern banking episodes. The broad pattern is that runs are rare in well-insured systems but reappear when insurance is absent, poorly designed, or overwhelmed. The table below summarises several significant bank run episodes and the mechanisms at work.
| Episode | Year | Country | Trigger | Safety Net in Place? | Nature of Run |
|---|---|---|---|---|---|
| US National Banking Era (multiple) | 1873–1907 | United States | Seasonal liquidity tightness, panic | None; clearinghouse certificates used | Panic/sunspot |
| Great Depression | 1930–1933 | United States | Agricultural distress, contagion | No federal deposit insurance | Mixed: fundamentals + panic |
| Continental Illinois | 1984 | United States | Loan losses, large depositor flight | FDIC coverage capped; implicit “too big to fail” | Fundamentals-driven |
| Northern Rock | 2007 | United Kingdom | Subprime losses, wholesale funding freeze | Limited insurance, BoE LLR delayed | Panic / retail run |
| Silicon Valley Bank | 2023 | United States | Interest rate losses, social media coordination | FDIC insurance, but large uninsured deposits | Panic, amplified by digital channels |
| |||||
Sources: Federal Reserve SVB review, FDIC Historical Statistics on Banking, Bank of England. The nature of runs reflects the interplay of fundamentals and panic, as captured by the Diamond-Dybvig multiple-equilibrium logic.
The chart below shows the number of bank failures in the United States per year from 1921 to 2023, capturing two distinct regimes: the pre-FDIC era (1921–1933) and the post-FDIC era (1934–2023). The dramatic decline after deposit insurance was introduced is consistent with the Diamond-Dybvig prediction that insurance eliminates the run equilibrium.
US bank failures per year, 1921–2023. Data from the Federal Deposit Insurance Corporation (FDIC) Historical Statistics on Banking. The vertical dashed line marks the introduction of federal deposit insurance in 1934.
The data are striking. In the four years 1930–1933, over 9,000 banks failed. After the introduction of federal deposit insurance through the FDIC in 1934, failures fell to fewer than 100 annually, and in many years, fewer than 10. The spike in the late 1980s and early 1990s reflects the savings and loan crisis, where the depositor-run mechanism was largely absent because deposits were insured; failures resulted from fundamental insolvency. The spike in 2009–2010, though large by modern standards, was still orders of magnitude smaller than the pre-insurance era.
Cross-country evidence further supports the model’s implications. Demirgüç-Kunt and Detragiache (2002) found that explicit deposit insurance reduces the likelihood of systemic banking crises, but only in countries with strong institutional environments. Where regulation is weak, deposit insurance can increase instability by encouraging excessive risk-taking, a finding consistent with the moral hazard extension of the Diamond-Dybvig framework. More recent work by the Bank for International Settlements and the International Monetary Fund continues to employ the model’s concepts when analysing the stability implications of digital deposit runs, stablecoin structures, and the vulnerabilities of large uninsured depositor bases. The original Diamond and Dybvig (1983) paper, still cited thousands of times, remains the canonical reference for understanding why bank runs are a structural feature of fractional-reserve banking rather than a historical relic.
How Diamond‑Dybvig Shapes Financial Regulation
The Diamond-Dybvig model has had a direct, measurable impact on financial regulation and central banking. Its framework underpins the design of deposit insurance systems in virtually every advanced economy, the articulation of lender-of-last-resort functions, and the post-crisis reforms that strengthen banks’ liquidity buffers. Few academic papers can claim such a sustained influence on real-world institutions.
The most direct application is the nearly universal adoption of explicit deposit insurance. The United States established the FDIC in 1933, before the theoretical model existed, but the model provided the ex post rationale that transformed deposit insurance from a temporary emergency measure into a permanent institution. Today, over 140 countries have explicit deposit insurance schemes, typically covering deposits up to a statutory limit. The European Union’s Deposit Guarantee Scheme Directive requires all member states to provide coverage of €100,000 per depositor per bank, a direct application of the Diamond-Dybvig logic that confidence in the banking system requires a credible promise that small depositors will not lose their savings. The model also clarifies the tradeoff between the benefits of insurance and the moral hazard it creates, which is why schemes are capped and backed by risk-based premiums and prudential regulation.
The model also justifies the lender-of-last-resort function of central banks, which was formalised by Walter Bagehot in the 19th century but received a new, rigorous economic rationale from Diamond and Dybvig. In the model’s bad equilibrium, a bank that is fundamentally solvent but temporarily illiquid can collapse. A central bank that stands ready to lend against good collateral can prevent the fire sale of assets and the destruction of value that follows. This logic was applied during the 2008 financial crisis, when the Federal Reserve, the European Central Bank, and the Bank of England extended emergency liquidity to banks facing wholesale funding runs. In 2023, the Fed’s new Bank Term Funding Program was designed explicitly to provide liquidity against the high-quality securities that SVB and other banks held, allowing them to meet depositor demands without realising their mark-to-market losses. This was a Diamond-Dybvig response to a Diamond-Dybvig crisis. The broader functions of central banks are inseparable from this role, as the lender-of-last-resort function is a core stabilising mechanism that modern central banking would not be complete without.
Banking regulation has incorporated the model’s insights through liquidity requirements. The Basel III Liquidity Coverage Ratio (LCR) requires banks to hold enough high-quality liquid assets to survive a 30-day stress scenario of deposit outflows. This is a direct acknowledgment of the model’s central vulnerability: a bank that relies on short-term deposits to fund long-term assets must have a buffer against the risk that depositors’ liquidity demands unexpectedly spike. When SVB failed, its LCR was not binding because the bank was not subject to the full Basel III regime due to a tailored regulatory framework for midsize banks. The subsequent regulatory debate has focused on whether the Diamond-Dybvig fragility that the LCR was designed to address should be applied more broadly, regardless of bank size.
More recently, the rise of digital banking and social media has given the Diamond-Dybvig model a new urgency. The SVB run showed that withdrawals can occur with unprecedented speed: $42 billion in a single day, a pace far exceeding the assumptions embedded in most liquidity regulations. The model’s sunspot mechanism, in which any event that coordinates expectations can trigger a run, fits the digital environment well, where information, including misinformation, spreads in minutes. The Bank for International Settlements and the Financial Stability Board are studying how to adapt liquidity regulation and deposit insurance to this new speed of panic. The core analysis remains rooted in Diamond-Dybvig, but the parameters have changed.
The model also informs the analysis of stablecoins and money market funds, which share the same maturity transformation structure as banks but operate outside the traditional safety net. A stablecoin that promises perfect convertibility but holds reserve assets that are less than perfectly liquid can be driven into a run by a loss of confidence, just as a bank can. The failure of TerraUSD in 2022 and the brief de-pegging of several major stablecoins have been analysed through the Diamond-Dybvig lens. The emergence of central bank digital currencies is partly motivated by the desire to offer a perfectly safe digital payment instrument that does not carry the risk of runs inherent in privately issued money-like claims.
The intellectual longevity of the Diamond-Dybvig model rests on the simplicity and generality of its core mechanism. The problem it identifies is not specific to a particular institutional context or historical period. It is a property of any contract that promises liquid claims backed by illiquid assets. As long as financial intermediation involves maturity transformation, the Diamond-Dybvig logic will apply. The policy challenge is to design intervention mechanisms that eliminate the bad equilibrium without destroying the good one, and every generation of policymakers rediscovers that balance.
MASEconomics Explains
Four economic concepts behind the Diamond-Dybvig model
Conclusion
The Diamond-Dybvig model provides the analytical foundation for understanding why banks are simultaneously valuable and fragile. Formalising the maturity transformation at the heart of banking, it shows how deposit contracts that improve welfare also create a coordination problem that can produce self-fulfilling runs. The model’s dual-equilibrium structure explains why solvent banks can fail in a panic and why government intervention, through deposit insurance and lender-of-last-resort facilities, can eliminate the bad equilibrium. Empirical evidence from the pre-FDIC era, the Great Depression, and modern banking crises consistently affirms the model’s predictions. The 2023 banking turmoil, far from refuting Diamond-Dybvig, provided a dramatic reaffirmation, updated for the digital age. The model remains the indispensable starting point for any serious analysis of bank fragility, deposit insurance design, and financial stability regulation.
Did you find this article helpful? Share it with someone who loves economics. And remember, at MASEconomics, we make complex ideas simple.
