Research

A century-scale, reproducible measurement of S&P 500 drawdown risk: 73 ATH-anchored underwater episodes of 5%+ across 24,742 trading days (1927–2026), of which 26 reach 10%+ and 12 reach 20%+. Four elements are treated together that rarely are: counting-convention sensitivity (73 vs 339 episodes of 5%+ on the same data), bootstrap uncertainty on recovery times (median 764 days for 20%+ bears, 95% CI [666, 2310]), Kaplan-Meier survival curves of time under water, and robustness across nominal/real × price/total-return conventions. Recession overlap is the watershed: 10%+ episodes overlapping NBER recessions run to a median of -35% over 988 days, versus -14.3% over 216 days without.

2026-07-10

An audited counterfactual of income-scaled DCA — contributions start from a $5,000/month income at a 30% rate and grow 5% per calendar year — applied to 26 assets from 2000 to 2026. The design separates a descriptive own-inception view (Primary B) from fair common windows, with W2018 as the main same-window comparison: NVDA leads W2018 XIRR at 70.98% while TSLA leads Primary B at 43.36%, and SPY anchors the benchmark at 12.25% (317 contributions compounding to $4.63M). Severe drawdowns accompany every extreme outcome, and the paper is explicit that an ex-post universe of winners cannot back a forward-looking selection rule.

2026-07-10

A high-fidelity 1-second backtest lab (Python reference + Rust parity) selects a Donchian rider on 2023–2026 that looks strong — +687 index points, profit factor 2.35, a STRONG_GO on MES futures — then refutes it with a 460,800-combination grid on a 19-year out-of-sample window (2004–2022): the champion ranks in the bottom third, the grid median is negative, and the apparent edge is concentrated in 2021–2022. A textbook post-selection-inference artifact, and the signal-layer companion to the SPX 0DTE Donchian negative-result paper.

2026-06-26

A disciplined, ex-post comparison of realized cross-asset performance (2000–2026, in USD) across 29 assets — equity indices, US stocks, commodities, currencies, cash, and bonds. It is not a strategy or forecast: the contribution is methodological discipline. Equal-window rankings flip the CAGR leader from Tesla to Nvidia; a risk-free-consistent Sharpe lowers headline ratios; intra-window daily drawdown exposes a +11.6% monthly COVID figure as a −38% crash; and significance tests show gold's down-month edge is not significant (p=0.09) while Treasuries' is (p=0.04).

2026-06-25

A fully-reproducible, in-sample audit of a volatility-targeted time-series-momentum overlay that asks whether an individual investor can improve on holding an equity index without giving it up. Implementation friction is quantified layer by layer — strategy (ETF, excess Sharpe 0.57, −0.04 correlation to SPY, 6/6 crisis windows), instrument (CME futures, 0.90–0.97 strict-pair correlation), and contract granularity (327 bp/yr tracking error at a $500K account, dominated by instrument basis rather than rounding). Every number is injected from the repository's outputs; all results are in-sample, with failed criteria, a 100× mis-scaling, and falsified predictions disclosed as results.

2026-06-19

A four-year out-of-sample evaluation of a Donchian 5-minute breakout on SPX 0DTE options. Full-population Databento tick validation across three exit policies gives a real-executable PnL of roughly −$107K to −$117K per year, with bootstrap 95% confidence intervals entirely below zero. The paper documents how Black-Scholes pricing and stratified-sample selection bias produced false-positive headlines in earlier drafts.

A seven-phase Python + Rust research engine validating Donchian breakout structure at the index point level across SPX, ES, and SPY. Covers point-level discovery, risk parameterization, regime filtering, walk-forward analysis, option-aware proxy validation (BSM / delta proxy / VIX-scaled IV), and real SPXW quote-chain checks with 1-second execution replay.

An audit-grade study of SPX opening-range breakout behavior (OR5/10/15/30) on 1-second data, testing whether open-drive signals can be expressed as 0DTE option exposure. Includes an epsilon filter grid, execution proxies with 1–15 second lags, forward-path MFE/MAE, a failed-breakout taxonomy, baseline matching, and a bootstrap Pareto-dominance gate.

A long-only 0DTE strategy targeting 5–20 minute directional moves through a three-layer signal architecture (bias, trigger, exhaustion). Hard risk invariants enforce no naked selling, a single open position, and a daily 2% hard stop. Implemented as a Python/Rust hybrid with iterative work on async execution safety.

A three-layer intraday pipeline: compression-state detection, direction confirmation, then option tradability and exit management. Uses time-of-day percentile normalization, strict closed-versus-forming bar semantics, and cost-conservative exit modeling, with a 300+ test suite enforcing safety invariants at the schema level.

A multi-version study of how breakouts propagate across 1m/5m/15m timeframes, with three strategy variants of increasing complexity, Black-Scholes option pricing, and both 20-year and out-of-sample analyses. The signal is mapped to SPXW 0DTE option exposure, with VIX-regime filtering in the implementation kit.

2026-06-19

A daily US-equity stat-arb pipeline built on IBKR adjusted daily bars: universe construction, Fama-French factor mapping, rolling beta estimation, residual z-score signals, daily target weights, and a full backtest with costs and positions. Walk-forward out-of-sample evaluation is paired with robustness gates on costs, borrow fees, gross exposure, and capacity.

A long-only S&P 500 overlay that de-risks during recessions using a labor-market composite (Initial Jobless Claims, Nonfarm Payrolls, Unemployment Rate). Out-of-sample Sharpe is 0.586 versus 0.43 for buy-and-hold, with max drawdown −27.6% versus −57.4% and Sharpe decay of only −0.006, validated over 21 out-of-sample years (2006–2026) on vintage first-print data.

An event-study framework measuring the SPX impact of macro releases (NFP, CPI, FOMC, ISM, Retail Sales, PCE) using surprise (actual minus consensus) computed from FRED vintage prints and TradingEconomics consensus. Multi-window direction probability and move magnitude are translated into 0DTE long call/put entry and exit parameters.