AN EXPLAINABLE MACHINE LEARNING FRAMEWORK FOR PREDICTING DIVIDEND SUSTAINABILITY AND FINANCIAL RISK OF U.S. REITS UNDER INTEREST-RATE SHOCKS
Keywords:
Real estate investment trusts, Dividend sustainability, Interest-rate shocks, Explainable machine learning, Monotonicity constraints, SHAP, Stress testingAbstract
Rising policy rates transmit to real estate investment trusts (REITs) through asset revaluation, debt repricing and refinancing frictions, and they can force cuts in the distributions that define the REIT contract. We propose RS-XMTL, a rate-shock-aware, explainable multi-task learning framework that jointly predicts (i) a dividend cut or suspension and (ii) a covenant-style financial-risk event over a four-quarter horizon. The framework combines an engineered rate-shock exposure block, a monotonicity-constrained gradient-boosting learner, a sign-restricted logistic component that extrapolates when the rate state leaves the training support, out-of-fold cross-task transfer, and a second-stage blender that also calibrates probabilities. Exogenous drivers are observed public series (10-year Treasury yields, CPI, S&P composite, CBOE VIX, 1999-2023), from which nine rate-shock quarters are identified; firm-level dynamics come from an accounting-consistent simulator of 300 REITs with five documented transmission channels, giving 21,442 firm-quarters. In five expanding-window out-of-time folds with a four-quarter embargo (12,544 test observations), RS-XMTL reaches AUC 0.794 and 0.904 on the two tasks, significantly above random forest, XGBoost, LightGBM and a multi-task network, with expected calibration error of 0.063 and 0.055 versus 0.323 and 0.130 for a class-balanced logistic benchmark. We introduce Regime SHAP Divergence, which quantifies how attribution shifts between normal and shock regimes, and show that macro and cross-task channels gain 7.4 and 9.5 percentage points of attribution mass in shock quarters. A counterfactual +300 bp stress test raises the mean predicted probability of a dividend cut from 26.7% to 45.0%, monotonically and with zero sign violations, whereas the unconstrained model implies the economically absurd conclusion that higher rates make cuts less likely.References
[1] Chan K C, Hendershott P H, Sanders A B. Risk and return on real estate: evidence from equity REITs. AREUEA Journal, 1990, 18(4): 431-452.
[2] Mueller G R, Pauley K R. The effect of interest-rate movements on real estate investment trusts. Journal of Real Estate Research, 1995, 10(3): 319-326.
[3] Allen M T, Madura J, Springer T M. REIT characteristics and the sensitivity of REIT returns. Journal of Real Estate Finance and Economics, 2000, 21(2): 141-152.
[4] Bredin D, O’Reilly G, Stevenson S. Monetary shocks and REIT returns. Journal of Real Estate Finance and Economics, 2007, 35(3): 315-331.
[5] Cotter J, Stevenson S. Multivariate modeling of daily REIT volatility. Journal of Real Estate Finance and Economics, 2006, 32(3): 305-325.
[6] Boudry W, Coulson N E, Kallberg J G, et al. On the hybrid nature of REITs. Journal of Real Estate Finance and Economics, 2012, 44(1): 230-249.
[7] Bradley M, Capozza D R, Seguin P J. Dividend policy and cash-flow uncertainty. Real Estate Economics, 1998, 26(4): 555-580.
[8] Hardin III W, Hill M D. REIT dividend determinants: excess dividends and capital markets. Real Estate Economics, 2008, 36(2): 349-369.
[9] Boudry W I. An examination of REIT dividend payout policy. Real Estate Economics, 2011, 39(4): 601-634.
[10] Hayunga D K, Stephens C P. Dividend behaviour of US equity REITs. Journal of Property Research, 2009, 26(2): 105-123.
[11] Ghosh C, Sun L. Agency cost, dividend policy and growth: the special case of REITs. Journal of Real Estate Finance and Economics, 2014, 48(4): 660-708.
[12] Devos E, Spieler A, Tsang D. Elective stock dividends and REITs: evidence from the financial crisis. Real Estate Economics, 2014, 42(1): 33-70.
[13] Altman E I. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance, 1968, 23(4): 589-609.
[14] Ohlson J A. Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research, 1980, 18(1): 109-131.
[15] Breiman L. Random forests. Machine Learning, 2001, 45(1): 5-32.
[16] Friedman J H. Greedy function approximation: a gradient boosting machine. Annals of Statistics, 2001, 29(5): 1189-1232.
[17] 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.
[18] Ke G, Meng Q, Finley T, et al. LightGBM: a highly efficient gradient boosting decision tree. In: Advances in Neural Information Processing Systems, 2017, 30: 3146-3154.
[19] Chawla N V, Bowyer K W, Hall L O, et al. SMOTE: synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 2002, 16: 321-357.
[20] Barboza F, Kimura H, Altman E. Machine learning models and bankruptcy prediction. Expert Systems with Applications, 2017, 83: 405-417.
[21] Fitzpatrick T, Mues C. An empirical comparison of classification algorithms for mortgage default prediction: evidence from a distressed mortgage market. European Journal of Operational Research, 2016, 249(2): 427-439.
[22] Bernanke B S, Kuttner K N. What explains the stock market’s reaction to Federal Reserve policy?. Journal of Finance, 2005, 60(3): 1221-1257.
[23] Kuttner K N. Monetary policy surprises and interest rates: evidence from the Fed funds futures market. Journal of Monetary Economics, 2001, 47(3): 523-544.
[24] Hardin III W, Highfield M J, Hill M D, et al. The determinants of REIT cash holdings. Journal of Real Estate Finance and Economics, 2009, 39(1): 39-57.
[25] Riddiough T J, Wu Z. Financial constraints, liquidity management and investment. Real Estate Economics, 2009, 37(3): 447-481.
[26] Gilstrap C, Petkevich A, Sezer O, et al. REIT debt pricing and ownership structure. Journal of Real Estate Finance and Economics, 2022, 64(4): 546-589.
[27] Downs D H, Zhu B. Property market liquidity and REIT liquidity. Real Estate Economics, 2022, 50(6): 1462-1491.
[28] Gu S, Kelly B, Xiu D. Empirical asset pricing via machine learning. Review of Financial Studies, 2020, 33(5): 2223-2273.
[29] Bianchi D, Büchner M, Tamoni A. Bond risk premia with machine learning. Review of Financial Studies, 2021, 34(2): 1046-1089.
[30] 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.
[31] Lundberg S M, Lee S-I. A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems, 2017, 30: 4765-4774.
[32] Lundberg S M, Erion G, Chen H, et al. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2020, 2(1): 56-67.
[33] Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 2019, 1(5): 206-215.
[34] Slack D, Hilgard S, Jia E, et al. Fooling LIME and SHAP: adversarial attacks on post hoc explanation methods. In: Proceedings of AAAI/ACM Conference AI, Ethics, and Society, 2020: 180-186.
[35] Apley D W, Zhu J. Visualizing the effects of predictor variables in black box supervised learning models. Journal of the Royal Statistical Society: Series B, 2020, 82(4): 1059-1086.
[36] Niculescu-Mizil A, Caruana R. Predicting good probabilities with supervised learning. In: Proceedings of the 22nd International Conference on Machine Learning (ICML), 2005: 625-632.
[37] Fuster A, Goldsmith-Pinkham P, Ramadorai T, et al. Predictably unequal? The effects of machine learning on credit markets. Journal of Finance, 2022, 77(1): 5-47.