How the Dallas Fed's Discount Window Model Flags Bank Risk

The standard market read treats discount window borrowing as a binary event: a bank takes a loan from the Federal Reserve and signals distress. The specific mechanism is more instructive. Ben Munyan and Ozge Ozden, economists at the Federal Reserve Bank of Dallas, published a working paper that dismantles the binary framing. They trained a regression model on discount window data from Q3 2010 through Q1 2024. The model was trained only through 2022. It predicted a substantial rise in borrowing going into 2023. That prediction preceded the failures of Silicon Valley Bank and Signature Bank.

The naive interpretation is that discount window borrowing equals trouble. The better market read is that borrowing equals a specific balance-sheet structure. The Dallas Fed paper provides the quantitative framework for that read. For anyone tracking regional lenders, it shifts the question from "Is a bank in trouble?" to "Is a bank structurally close to needing the emergency facility?" That is a finer, more useful distinction, and it has a measurable checklist.

The Borrower Profile: A Five-Point Balance-Sheet Checklist

The model compares average balance-sheet composition over the 14-year sample. Borrower banks are larger on average. They hold fewer core deposits and less capital. The predictive differences show up in five measurable traits.

Reserve Ratios Lead the Signal

Reserves are the most liquid asset – cash or deposits at the Fed used for daily payment flows. Borrower banks hold lower reserves as a fraction of assets than non-borrowers. The gap widened after 2022. The model identifies the reserve-to-asset ratio as the strongest predictor of future borrowing, holding all other characteristics constant. A bank with a falling reserve ratio is structurally closer to needing the window.

Illiquid Assets Force the Hand

Borrower banks carry more commercial and industrial (C&I) loans and illiquid securities. C&I loans are secured by business assets and projected cash flows, not by stable collateral. That matters for discount window access. C&I loans can be pledged to the Fed, unlike at the Federal Home Loan Banks. Banks with high C&I concentration find the discount window a natural fit for contingent liquidity needs.

Non-core funding – Fed funds borrowing, repo, and FHLB advances – also predicts borrowing. The study finds these funding sources act as complements to the discount window, not substitutes. A bank already tapping short-term wholesale funding is mechanically more likely to use the window when stress rises.

Practical rule: A bank with falling reserve ratios, rising C&I loans, and expanding non-core funding is a candidate for discount window use before any visible stress.

Capital Quality Deteriorates First

Borrower banks have lower risk-based capital ratios. The Dallas Fed adjusts tangible common equity (TCE) for unrealized losses on hold-to-maturity (HTM) securities. As the Fed raised rates in 2022, HTM losses surged. The adjusted TCE ratio dropped sharply at banks that later borrowed. That ratio is a second key predictor: as it declines, the probability of borrowing rises.

Comparing Borrower and Non-Borrower Banks

Table: Average balance-sheet composition, Q3 2010–Q1 2024. Source: Federal Reserve Bank of Dallas working paper.

The Geography Variable: Why the Eleventh District Matters

Macro Amplifiers and the 2023 Test Case

The model's training set ended in 2022. The key variables – low reserves, illiquid assets, falling adjusted TCE – were all moving against borrower banks as the Fed tightened. The macro inputs sharpened the forecast. Borrowing generally declines when GDP growth is positive. Fed balance sheet expansion has the opposite effect, correlating with a rise in discount window borrowing the following quarter.

The model predicted a substantial rise in borrowing going into 2023. That prediction validated every significant variable in the regression. The paper's practical value is this: a model trained on 12 years of relatively calm data, using only structural balance-sheet inputs, caught the turning point. It did not need a bank run to happen first. It needed the balance-sheet data to arrive.

Risk to watch: If the model's leading indicators, especially reserve ratios and adjusted TCE, start deteriorating across a cluster of banks in one district, it is worth stress-testing regional bank exposures before public stress appears.

Why Stigma Is Weakening

The paper documents a steady rise in banks with discount window access since 2010. Test loans (under $10,000 to check operational readiness) have become standard practice. The share of active borrowing banks rose sharply after 2023. The practical effect is that a single discount window borrowing event carries less weight than it did in 2008. The market must distinguish between a routine test, a structural liquidity policy, and genuine distress. The Dallas Fed checklist draws that line.

Confirming the Setup and Breaking the Thesis

Sentinel Triggers from the Model

The framework produces a quarterly screening checklist. For any bank with discount window access, watch:

Reserve-to-asset ratio falling below the peer median.

Adjusted TCE ratio declining due to HTM losses.

C&I loans as a share of assets rising above 20%.

Non-core funding growing faster than core deposits.

Asset growth outpacing capital growth.

First quarter seasonal dip – the model finds borrowing is systematically lower in Q1.

A cluster of banks showing this profile across a single district heightens the signal.

Breaking the Thesis

The thesis weakens when the specific inputs reverse. A broad recovery in reserve ratios across the banking sector reduces the model's predictive power. A Federal Reserve that shrinks its balance sheet consistently removes the macro amplifier. A return of discount window stigma, where banks stop taking test loans, would mean the data set becomes less representative of structural demand. The 2023 event was a stress test the model passed. The real test is whether the same indicators work in a different stress scenario – for example, a credit cycle downturn or a commercial real estate shock.

The Next Decision Point

For anyone tracking stock market analysis on regional banks or broader financial exposure through KRE and XLF, the Dallas Fed framework provides a specific timing tool. The 2023 event validated the model. The next quarterly Call Report cycle is the first fresh data point. If a cluster of banks shows falling adjusted TCE and shrinking reserve ratios, the model would flag that cluster before a public stress event.

The paper's contribution is specificity. It tells the reader exactly what to track and what the weighting should be. Reserve ratios matter most. Adjusted TCE comes second. Regional location shifts the baseline. Macro conditions amplify the signal. The combination of a slowing economy and a cluster of banks falling into the borrower profile is the pattern to flag. The market response would likely involve widening credit spreads and a rotation out of those specific names before any official discount window data is published.