Stephan Luck on What History can Teach Us about Financial Stability

Episode Summary

Executive Summary: Stefan Luck explains how historical microdata and AI/LLMs are reshaping research on bank runs, showing that bank failures are usually driven by weak fundamentals and solvency problems rather than runs alone. He also compares the national banking era to today’s stablecoin debate and argues German hyperinflation reveals an important debt-repricing channel, though more work is needed on general equilibrium effects.

Main Topics: How historical data and LLMs changed the research agenda (Priority: 5/5): Luck describes moving from theory toward historical empirical work by digitizing bank records and later using LLMs to sift millions of newspaper articles for true bank-run events. This allowed construction of a large bank-run dataset that would have been infeasible manually. Bank failures: solvency vs. liquidity (Priority: 5/5): The main empirical finding is that bank failures are strongly linked to deteriorating fundamentals and deep insolvency. Bank runs occur, but they are often not the primary cause of failure; healthy banks can survive runs, while weak banks are vulnerable. Policy implications for lender of last resort and resolution (Priority: 4/5): Luck argues liquidity support can matter, but the bigger policy question is how to identify insolvent banks and recapitalize or resolve them rather than merely lending to them. He frames historical evidence as favoring solvency-focused policy over pure liquidity provision. Silicon Valley Bank through a historical lens (Priority: 4/5): He interprets SVB as resembling a 19th-century-style failure: uninsured deposits, deep insolvency, and a run that came later than one might expect. The example is used to reinforce the importance of fundamentals over panic narratives. Stablecoins and the national banking system (Priority: 4/5): Luck compares the Genius Act/stablecoin framework to the national banking era, when private banknotes were backed by Treasuries. He argues stablecoins may echo that system, but likely have more promise in cross-border or high-inflation settings than in domestic retail payments. German hyperinflation and the debt-inflation channel (Priority: 4/5): Luck and coauthors find that unexpected inflation can boost activity for indebted firms by reducing real debt burdens. This suggests nominal debt rigidity is a meaningful friction, though the magnitude and general equilibrium effects remain open questions.

Key Arguments: Historical bank failures are highly predictable and are mostly associated with poor bank fundamentals, not random panic. Bank runs can happen at healthy banks, but healthy banks tend to survive them; weak banks suffer permanent damage or fail. Low recovery rates in historical bank receiverships make it hard to argue that franchise value loss from a run was the main driver of most failures. LLMs are especially useful for filtering false positives and clustering newspaper reports into true distress events from massive archival corpora. Liquidity support is not useless, but the better policy is to identify insolvent institutions and either recapitalize or resolve them. SVB fits the pattern of a fundamentally weak bank with a delayed run, rather than a pure liquidity crisis. Stablecoins resemble national banknotes more than pre-Civil War free banking, but domestic use may be limited because modern deposits already offer safety, interest, and fast payments. German hyperinflation shows that inflation can ease constraints for indebted firms, increasing investment and hiring when nominal debt contracts are fixed. The debt-inflation channel is real, but its macro importance depends on who is constrained and on general equilibrium effects that are not yet fully understood.

Data Points: Newspaper articles searched: 374 million - Luck says the team searched a massive archive of publicly available historical newspapers to identify bank-run events. Dataset size of bank runs: more than 3,500 - LLMs helped convert newspaper coverage into a structured dataset of historical bank runs and related distress events. Historical document source: annual reports of the Office of the Comptroller of the Currency - Used to digitize balance sheets of national banks from the Civil War era through the Great Depression. Time span of national bank balance-sheet data: Civil War through 1941 - The OCC reports covered national banks across this long historical period. German bond interest rates observed: 4%, 4.5%, and 5% - Luck noted limited variation in pre-WWI German bond coupons in the hyperinflation paper. FDIC loss on SVB: around $20 billion - Used as evidence that Silicon Valley Bank was deeply insolvent. Depression-era note business vs. deposits: deposits eventually became more attractive than notes - In the stablecoin discussion, he argues bank deposits overtook banknotes as payments infrastructure improved. Inflation in early German hyperinflation: price level rose by a factor of about 30 - He cites the period up to summer 1922 to illustrate the scale of the shock.

Pivotal Quotes: "bank failures tend to be extremely predictable" — Stefan Luck: Summarizing the core empirical finding that weak fundamentals, not random runs, drive failures. "the better policy would be to actually find out which institutions are insolvent and then recapitalize them, or if that's not feasible, resolve them" — Stefan Luck: His preferred policy response to banking fragility over pure liquidity provision. "we've just put a little bit too much emphasis on [bank runs] as a sort of a cause of why bad things happen" — Stefan Luck: His broad critique of the standard narrative that bank runs are the main driver of bank failures.

Implications: Listeners should expect more emphasis on capital and solvency in financial-stability policy, with liquidity viewed as secondary. The work also suggests AI can unlock archival evidence and that stablecoins may matter most where banking and monetary institutions are weak.