PhD Validation · About · Model Developer
Independent PhD Backtest Verification
An independent PhD reviewer validated all four RegimeSignal™ market-prediction signals, confirming that the published historical results could be reproduced exactly from the locked files and that the testing methodology was sound. The review identified no look-ahead bias or data leakage, and concluded that the signals were "Validated with Qualifications," with all limitations fully disclosed.
Overall Conclusion
VALIDATED WITH QUALIFICATIONS
All four published confusion matrices reproduced exactly from the locked files. The methodology was judged sound for the Bear Regime Signal and both Market Break tiers. The qualifications — disclosed below in full — pertain principally to the Regime Recovery Signal and to two macro-guardrail data gaps, and were themselves pre-disclosed by the author as the reason for commissioning the review.
The Engagement
RegimeSignal™ commissioned a fully independent verification of its four validated classifier signals from a doctoral quantitative-finance researcher with no role in the models' construction and no financial relationship with Cronus Market Intelligence beyond the review fee. The reviewer was given a locked evaluator package — code, locked prediction files, data, and a SHA256 manifest — and worked from it directly over a defined review window, reinstalling the environment from scratch and re-running every reproduction script.
The Reviewer
Ali Trabelsi Karoui, PhD
PhD in Finance · University of Sfax (2023)
R&D Team Lead, Financial Innosights · Data Scientist, FundEvolve · former Postdoctoral Researcher, ICADE Business School, Universidad Pontificia Comillas (Madrid)
Peer-reviewed publications in ranked finance journals, with research spanning regime modeling, market-stress indices, exchange-rate econometrics, and applied machine learning in finance (Best Paper Award, International Finance Conference, 2024). His hands-on methodological expertise maps directly onto what these signals use: walk-forward (expanding-window) backtesting and out-of-sample evaluation, L1/L2-regularized logistic regression, gradient-boosted decision trees (XGBoost), confusion-matrix and AUC/precision–recall analysis under class imbalance, bootstrap robustness testing, and Markov regime-switching models.
Scope & Protocol
Four classifier signals were in scope: the Bear Regime Signal (BRS), Market Break Signal Tier 1 (MBS T1), Market Break Signal Tier 2 (MBS T2), and the Regime Recovery Signal (RRS). The Bull and Bear Velocity gauges, the HybridBrain™ ERI overlay, and the AI assistant were explicitly out of scope and were not reviewed.
The structured protocol covered five areas: (a) package-integrity verification against the SHA256 manifest, (b) reproduction of every headline confusion matrix, (c) walk-forward protocol integrity (lookahead / leakage), (d) input-data integrity via FRED spot-checks against primary sources, and (e) a methodology review against known statistical failure modes.
Findings by Area
All four confusion matrices reproduced exactly from the locked files; the three XGBoost signals matched to floating-point tolerance, and every walk-forward fire/no-fire decision was identical. The intervals reflect genuine sampling uncertainty given the modest number of historical signal events — widest for RRS, which fired only 11 times in its out-of-sample window.
⚠ Qualifications — Disclosed in Full
Regime Recovery Signal (RRS). Its decision threshold and hyperparameters were selected on the same out-of-sample window on which its results are reported. Accordingly, the reviewer characterizes the RRS precision as a historical upper bound under an optimized operating point, not an unbiased forward estimate, and notes its wide confidence interval (52%–95%). This was pre-disclosed by the author and was the stated motivation for commissioning the review.
BRS guardrail data gaps. Two of the eight macro guardrails (a high-yield credit-spread series and a forward P/E series) had data gaps and defaulted to PASSED for much of the out-of-sample history, so the two-layer system effectively operated with roughly six of eight guardrails over a large fraction of the period. The reviewer confirms the conservative direction of both defaults is correctly characterized — they cannot inflate false positives, only reduce coverage.
BRS coverage. Bear-Alert recall is intentionally secondary by design (the high precision comes at the cost of missed bear months at the Alert level specifically); broader coverage is supplied by the velocity gauges and lower-tier alert tracks, which were out of scope here.
This independent review validates the documented walk-forward methodology and the reproducibility of the headline confusion matrices from the locked artifacts as of the review date. It does not constitute an opinion on future model results, does not validate live or production deployment, does not cover components outside the four classifier signals, and does not constitute investment advice. Past market signal record does not guarantee future results.