AN EXPLAINABLE HYBRID MACHINE LEARNING FRAMEWORK FOR CASH-FLOW-CONDITIONED MULTI-HORIZON LIQUIDITY RISK EARLY WARNING: AN SME-ORIENTED BENCHMARK STUDY
Keywords:
SME, Liquidity risk, Financial distress, Cash-flow ratios, LightGBM, Probability calibration, SHAP, Early warningAbstract
Small and medium-sized enterprises (SMEs) are particularly exposed to liquidity shocks, yet public, transaction-level SME cash-flow panels suitable for reproducible forecasting remain scarce. This paper therefore studies a narrower but operationally related task: cash-flow-conditioned, multi-horizon liquidity-risk early warning. A real public benchmark containing 43,405 financial-statement observations, 64 ratios, missing values, and bankruptcy labels one to five years ahead is used as an evidence base. The proposed framework combines three views: an L2-regularized logistic model, a LightGBM model using all ratios, and a dedicated LightGBM expert using 23 cash-flow and liquidity ratios. Non-negative horizon-specific weights are selected on a calibration split in log-odds space, followed by isotonic probability calibration, a recall-constrained warning threshold, and SHAP-based feature and semantic-group explanations. Across five held-out horizons, the calibrated hybrid achieved mean ROC-AUC 0.914, PR-AUC 0.617, and Brier score 0.0260. The uncalibrated ensemble had higher ranking performance (PR-AUC 0.663) but materially worse probability error (Brier 0.0424), exposing a genuine calibration-discrimination trade-off. The cash-flow expert received positive weight at the two- and five-year horizons, while the full-feature tree dominated the remaining horizons; the PR-AUC gain over LightGBM-all was not statistically significant (one-sided Wilcoxon p=0.50). The study supplies executable code, data hashes, result tables, figures, and verified references. Because the benchmark does not identify firms by a legal SME definition and is not a cash-flow time series, the findings support an SME-oriented methodology rather than a claim of direct SME field validation.References
[1] Beaver W H. Financial ratios as predictors of failure. Journal of Accounting Research, 1966, 4: 71-111. DOI: 10.2307/2490171.
[2] Altman E I. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. The Journal of Finance, 1968, 23(4): 589-609. DOI: 10.1111/j.1540-6261.1968.tb00843.x.
[3] Ohlson J A. Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research, 1980, 18(1): 109-131. DOI: 10.2307/2490395.
[4] Zmijewski M E. Methodological issues related to the estimation of financial distress prediction models. Journal of Accounting Research, 1984, 22: 59-82. DOI: 10.2307/2490859.
[5] Shumway T. Forecasting bankruptcy more accurately: A simple hazard model. The Journal of Business, 2001, 74(1): 101-124. DOI: 10.1086/209665.
[6] Campbell J Y, Hilscher J, Szilagyi J. In search of distress risk. The Journal of Finance, 2008, 63(6): 2899-2939. DOI: 10.1111/j.1540-6261.2008.01416.x.
[7] Altman E I, Sabato G. Modelling credit risk for SMEs: Evidence from the U.S. market. Abacus, 2007, 43(3): 332-357. DOI: 10.1111/j.1467-6281.2007.00234.x.
[8] Ciampi F, Gordini N. Small enterprise default prediction modeling through artificial neural networks: An empirical analysis of Italian small enterprises. Journal of Small Business Management, 2013, 51(1): 23-45. DOI: 10.1111/j.1540-627X.2012.00376.x.
[9] Brown I, Mues C. An experimental comparison of classification algorithms for imbalanced credit scoring data sets. Expert Systems with Applications, 2012, 39(3): 3446-3453. DOI: 10.1016/j.eswa.2011.09.033.
[10] Saito T, Rehmsmeier M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLOS ONE, 2015, 10(3): e0118432. DOI: 10.1371/journal.pone.0118432.
[11] Barboza F, Kimura H, Altman E. Machine learning models and bankruptcy prediction. Expert Systems with Applications, 2017, 83: 405-417. DOI: 10.1016/j.eswa.2017.04.006.
[12] Lessmann S, Baesens B, Seow H-V, et al. Benchmarking state-of-the-art classification algorithms for credit scoring: An update of research. European Journal of Operational Research, 2015, 247(1): 124-136. DOI: 10.1016/j.ejor.2015.05.030.
[13] Niculescu-Mizil A, Caruana R. Predicting good probabilities with supervised learning. In: Proceedings of the 22nd International Conference on Machine Learning, 2005: 625-632. DOI: 10.1145/1102351.1102430.
[14] Zadrozny B, Elkan C. Transforming classifier scores into accurate multiclass probability estimates. In: Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2002: 694-699. DOI: 10.1145/775047.775151.
[15] Lundberg S M, Lee S-I. A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems 30, 2017. DOI: 10.5555/3295222.3295230.
[16] Lundberg S M. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2020, 2: 56-67. DOI: 10.1038/s42256-019-0138-9.
[17] Ribeiro M T, Singh S, Guestrin C. Why should I trust you? Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 1135-1144. DOI: 10.1145/2939672.2939778.
[18] Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 2019, 1: 206-215. DOI: 10.1038/s42256-019-0048-x.
[19] Balcaen S, Ooghe H. 35 years of studies on business failure: An overview of the classic statistical methodologies and their related problems. The British Accounting Review, 2006, 38(1): 63-93. DOI: 10.1016/j.bar.2005.09.001.
[20] Sun J, Li H, Huang Q-H, et al. Predicting financial distress and corporate failure: A review from the state-of-the-art definitions, modeling, sampling, and featuring approaches. Knowledge-Based Systems, 2014, 57: 41-56. DOI: 10.1016/j.knosys.2013.12.006.
[21] Friedman J H. Greedy function approximation: A gradient boosting machine. The Annals of Statistics, 2001, 29(5): 1189-1232. DOI: 10.1214/aos/1013203451.
[22] Ke G. LightGBM: A highly efficient gradient boosting decision tree. In: Advances in Neural Information Processing Systems 30, 2017. DOI: 10.5555/3294996.3295074.
[23] Zieba M, Tomczak S K, Tomczak J M. Ensemble boosted trees with synthetic features generation in application to bankruptcy prediction. Expert Systems with Applications, 2016, 58: 93-101. DOI: 10.1016/j.eswa.2016.04.001.
[24] Geng R, Bose I, Chen X. Prediction of financial distress: An empirical study of listed Chinese companies using data mining. European Journal of Operational Research, 2015, 241(1): 236-247. DOI: 10.1016/j.ejor.2014.08.016.
[25] Tomczak S. Polish Companies Bankruptcy. UCI Machine Learning Repository, Dataset, 2016. DOI: 10.24432/C5F600.
[26] Brier G W. Verification of forecasts expressed in terms of probability. Monthly Weather Review, 1950, 78(1): 1-3. DOI: 10.1175/1520-0493(1950)078<0001:VOFEIT>2.0.CO;2.
[27] Kapoor S, Narayanan A. Leakage and the reproducibility crisis in machine-learning-based science. Patterns, 2023, 4(9): 100804. DOI: 10.1016/j.patter.2023.100804.
[28] Barth M E, Cram D P, Nelson K K. Accruals and the prediction of future cash flows. The Accounting Review, 2001, 76(1): 27-58. DOI: 10.2308/accr.2001.76.1.27.
[29] Dechow P M, Kothari S P, Watts R L. The relation between earnings and cash flows. Journal of Accounting and Economics, 1998, 25(2): 133-168. DOI: 10.1016/S0165-4101(98)00020-2.