EARLY WARNING OF LIQUIDITY STRESS IN U.S. BANKS WITH SEASONALLY ALIGNED FRAILTY HAZARDS AND DELAYED-FEEDBACK CONFORMAL ALARMS
Keywords:
Early warning systems, Liquidity risk, Bank runs, Discrete-time hazard models, Gradient boosting, Conformal prediction, Call ReportsAbstract
Bank liquidity crises unfold within days, but the balance-sheet vulnerabilities behind them build up over several quarters. This paper develops SAFE (Seasonally Aligned Frailty Early-warning), a machine-learning early-warning system for liquidity stress in U.S. banks. Using the complete Call Reports for 2011–2023 (298,996 bank-quarters), we define a liquidity stress event as a quarterly deposit outflow of at least 7.5% of assets that the bank absorbs through wholesale borrowing or the liquidation of assets, or as a bank failure. SAFE forecasts the probability of such an event over horizons of one to four quarters with a censoring-aware discrete-time hazard model in which each horizon-specific hazard is a gradient-boosted ensemble. Because each hazard refers to a single calendar quarter, its inputs are aligned with the quarter being forecast through empirical-Bayes estimates of the bank’s propensity to recurrent and seasonal funding runs. A delayed-feedback adaptive conformal procedure converts the forecasts into supervisory alarms and guarantees that the long-run share of missed events tracks a prescribed target under arbitrary distribution shift. In an out-of-time test on the 2021Q4–2023Q3 monetary-tightening episode, SAFE raises the area under the precision–recall curve relative to LightGBM from 0.205 to 0.232 at the one-quarter horizon and from 0.382 to 0.400 at the four-quarter horizon, outperforms direct classifiers given identical features at all multi-quarter horizons and produces a monotone risk term structure for every bank. SAFE is the most accurate model in each year of a 2016–2023 rolling backtest, and its alarms maintain a recall of 0.80 through the 2022–2023 deposit outflows, whereas a static threshold falls to 0.69.References
[1] Drechsler I, Savov A, Schnabl P, et al. Banking on uninsured deposits. NBER Working Paper 31138, National Bureau of Economic Research, 2023.
[2] Jiang E X, Matvos G, Piskorski T, et al. Monetary tightening and U.S. bank fragility in 2023: Mark-to-market losses and uninsured depositor runs? NBER Working Paper 31048, National Bureau of Economic Research, 2023.
[3] Board of Governors of the Federal Reserve System. Review of the Federal Reserve’s supervision and regulation of Silicon Valley Bank. Board of Governors of the Federal Reserve System, Washington, DC, 2023.
[4] Cole R A, White L J. Déjà vu all over again: The causes of U.S. commercial bank failures this time around. Journal of Financial Services Research, 2012, 42: 5-29.
[5] Betz F, Oprică S, Peltonen T A, et al. Predicting distress in European banks. Journal of Banking & Finance, 2014, 45: 225-241.
[6] Carmona P, Climent F, Momparler A. Predicting failure in the U.S. banking sector: An extreme gradient boosting approach. International Review of Economics & Finance, 2019, 61: 304-323.
[7] Petropoulos A, Siakoulis V, Stavroulakis E, et al. Predicting bank insolvencies using machine learning techniques. International Journal of Forecasting, 2020, 36: 1092-1113.
[8] Gama J, Žliobaitė I, Bifet A, et al. A survey on concept drift adaptation. ACM Computing Surveys, 2014, 46: 44:1-44:37.
[9] Berger A N, Bouwman C H S. How does capital affect bank performance during financial crises? Journal of Financial Economics, 2013, 109: 146-176.
[10] Constantin A, Peltonen T A, Sarlin P. Network linkages to predict bank distress. Journal of Financial Stability, 2018, 35: 226-241.
[11] Climent F, Momparler A, Carmona P. Anticipating bank distress in the Eurozone: An extreme gradient boosting approach. Journal of Business Research, 2019, 101: 885-896.
[12] Gogas P, Papadimitriou T, Agrapetidou A. Forecasting bank failures and stress testing: A machine learning approach. International Journal of Forecasting, 2018, 34: 440-455.
[13] Le H H, Viviani J-L. Predicting bank failure: An improvement by implementing a machine-learning approach to classical financial ratios. Research in International Business and Finance, 2018, 44: 16-25.
[14] Sarlin P. On policymakers’ loss functions and the evaluation of early warning systems. Economics Letters, 2013, 119: 1-7.
[15] Holopainen M, Sarlin P. Toward robust early-warning models: A horse race, ensembles and model uncertainty. Quantitative Finance, 2017, 17: 1933-1963.
[16] Lang J H, Peltonen T A, Sarlin P. A framework for early-warning modeling with an application to banks. ECB Working Paper No. 2182, European Central Bank, 2018.
[17] Beutel J, List S, Schweinitz G von. Does machine learning help us predict banking crises? Journal of Financial Stability, 2019, 45: 100693.
[18] Bluwstein K, Buckmann M, Joseph A, et al. Credit growth, the yield curve and financial crisis prediction: Evidence from a machine learning approach. Journal of International Economics, 2023, 145: 103773.
[19] Diamond D W, Dybvig P H. Bank runs, deposit insurance, and liquidity. Journal of Political Economy, 1983, 91: 401-419.
[20] Iyer R, Puri M. Understanding bank runs: The importance of depositor-bank relationships and networks. American Economic Review, 2012, 102: 1414-1445.
[21] Iyer R, Puri M, Ryan N. A tale of two runs: Depositor responses to bank solvency risk. Journal of Finance, 2016, 71: 2687-2726.
[22] Egan M, Hortaçsu A, Matvos G. Deposit competition and financial fragility: Evidence from the US banking sector. American Economic Review, 2017, 107: 169-216.
[23] Acharya V V, Mora N. A crisis of banks as liquidity providers. Journal of Finance, 2015, 70: 1-43.
[24] Bai J, Krishnamurthy A, Weymuller C-H. Measuring liquidity mismatch in the banking sector. Journal of Finance, 2018, 73: 51-93.
[25] Hong H, Huang JZ, Wu D. The information content of Basel III liquidity risk measures. Journal of Financial Stability, 2014, 15: 91-111.
[26] Drechsler I, Savov A, Schnabl P. The deposits channel of monetary policy. Quarterly Journal of Economics, 2017, 132: 1819-1876.
[27] Choi D B, Goldsmith-Pinkham P, Yorulmazer T. Contagion effects of the Silicon Valley Bank run. NBER Working Paper 31772, National Bureau of Economic Research, 2023.
[28] Grinsztajn L, Oyallon E, Varoquaux G. Why do tree-based models still outperform deep learning on typical tabular data? In: Advances in neural information processing systems 35, datasets and benchmarks track, 2022.
[29] Chen T, Guestrin C. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016: 785-794.
[30] Ke G, Meng Q, Finley T, et al. LightGBM: A highly efficient gradient boosting decision tree. In: Advances in neural information processing systems 30, 2017: 3146-3154.
[31] Angelopoulos A N, Bates S. Conformal prediction: A gentle introduction. Foundations and Trends in Machine Learning, 2023, 16: 494-591.
[32] Tibshirani R J, Barber R F, Candès E J, et al. Conformal prediction under covariate shift. In: Advances in neural information processing systems 32, 2019.
[33] Barber R F, Candès E J, Ramdas A, et al. Conformal prediction beyond exchangeability. Annals of Statistics, 2023, 51: 816-845.
[34] Gibbs I, Candès E. Adaptive conformal inference under distribution shift. In: Advances in neural information processing systems 34, 2021.
[35] Tong X, Feng Y, Li J J. Neyman-Pearson classification algorithms and NP receiver operating characteristics. Science Advances, 2018, 4: eaao1659.
[36] Lundberg S M, Lee S-I. A unified approach to interpreting model predictions. In: Advances in neural information processing systems 30, 2017: 4765-4774.
[37] Basel Committee on Banking Supervision. Basel III: The liquidity coverage ratio and liquidity risk monitoring tools. Bank for International Settlements, Basel, 2013.