EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

EARLY WARNING OF LIQUIDITY STRESS IN U.S. BANKS WITH SEASONALLY ALIGNED FRAILTY HAZARDS AND DELAYED-FEEDBACK CONFORMAL ALARMS

JiYuan Wang, Yao Ge, JingYi Liu, BingYing Jiang

Journal of Trends in Finance and Economics · 03.02.2024 · 1 (1) · 85-100

DOI: 10.61784/jtfe4078 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Trends in Finance and Economics, online erschienen am 03.02.2024. Verfasst von JiYuan Wang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 16-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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.

Zitieren

APA

Wang, J., Ge, Y., Liu, J., & Jiang, B. (2024). EARLY WARNING OF LIQUIDITY STRESS IN U.S. BANKS WITH SEASONALLY ALIGNED FRAILTY HAZARDS AND DELAYED-FEEDBACK CONFORMAL ALARMS. Journal of Trends in Finance and Economics, 1(1), 85-100. https://doi.org/10.61784/jtfe4078

Vancouver

Wang J, Ge Y, Liu J, Jiang B. EARLY WARNING OF LIQUIDITY STRESS IN U.S. BANKS WITH SEASONALLY ALIGNED FRAILTY HAZARDS AND DELAYED-FEEDBACK CONFORMAL ALARMS. Journal of Trends in Finance and Economics. 2024;1(1):85-100. doi:10.61784/jtfe4078

BibTeX
@article{Wang2024,
  author = {Wang, JiYuan and Ge, Yao and Liu, JingYi and Jiang, BingYing},
  title = {EARLY WARNING OF LIQUIDITY STRESS IN U.S. BANKS WITH SEASONALLY ALIGNED FRAILTY HAZARDS AND DELAYED-FEEDBACK CONFORMAL ALARMS},
  journal = {Journal of Trends in Finance and Economics},
  year = {2024},
  volume = {1},
  number = {1},
  pages = {85--100},
  doi = {10.61784/jtfe4078}
}
Kostenlos starten →