Forecasting Leverage-Induced First-Passage Failure with Regime-Persistence Operators: Monte Carlo Evidence and a July 2026 Stress Case

Abstract

Oscar Linares, David T. Daly, Jr., Ričards Bulavs, and Emmanuel Merchain-Cruz

We examine whether regime-persistence measures improve forecasts of leverage-induced first-passage failure. A July 2026 case combines a June 30 Form 13F reconstruction with a sealed Clock-of-Regimes (COR) observation. The disclosed portfolio was highly concentrated: its two largest positions represented 55.6% of reported value and a nine-stock core represented 88.3%. On July 21, COR assigned probability one, to reported precision, to Stress and produced daily stress volatility of 3.54%, tail exposure of 0.0486, and dominant persistence rho(N) = 95.2 In a matched 200,000-path experiment, COR conditioning raises estimated eightfailure probability from 0.22% to 1.77% at 2x leverage, 4.25% to 13.00% at 2.5x, 16.09% to 30.64% at 3x, and 50.60% to 63.95% at 4x. A separate forecasting experiment uses 250 synthetic replications, each containing 1,500 training and 1,000 held-out observations. Relative to conventional predictors augmented with HMM probabilities, COR reduces mean Brier loss by 0.0007 and logarithmic loss by 0.0026, while increasing AUC by 0.0103 and AUPRC by 0.0087. Paired 95% Monte Carlo intervals exclude zero for all four differences. The simulation identifies a transparent data-generating environment in which persistence and tail coordinates add forecast information beyond regime occupancy. COR forecasts survival risk, not return direction.

PDF