A portfolio of systematic trading models and the theory behind them — empirical model breakdowns and first-principles concept writeups, side by side.
Compiled 2026-06-04 · classical · machine learning · microstructure · alt-data · options
This is the reading room for the trading-models repository. The work runs on two tracks, shown together below. The model breakdowns are end-to-end empirical papers — one per model, each carrying a hypothesis through a lookahead-safe backtest to a verdict, winners and negative results alike. The concept writeups are the theory those models rest on, derived from first principles before any backtest. Filter the cards below, or open either track's full catalogue for sorting; shared conventions live in methodology.md and the glossary. The build history behind it all — why each major change landed, what broke, and what fixed it — runs in the engineering devlog.
The 50/200-day golden-cross rule on SPY, framed as a finite-impulse-response filter with explicit group delay. Why it earns a positive Sharpe yet cannot avoid a deep drawdown.
SPY daily, 2010–2024 · Sharpe +0.75 · Max DD −34% · working
ML-01 · Supervised ForecastingA second-order boosted-tree regressor on eight technical features. The information coefficient collapses out-of-sample; we reconcile a sub-50% hit rate with a positive Sharpe through the Fundamental Law of Active Management.
SPY daily, OOS 2022–2024 · Sharpe +0.96 · Max DD −12% · working
MIC-01 · Order FlowTrade-signed order-flow imbalance on minute-bar BTCUSDT across the 2024-08-05 crash. The continuation hypothesis is rejected; we derive why minute-frequency Sharpe is not comparable to the daily-bar papers and what the scale-invariant metrics say instead.
BTCUSDT 1-min, Aug 2024 · Hit rate 29.7% · Max DD −36% · negative result
ALT-01 · AttentionA rolling z-score of "bitcoin" search interest, traded contrarian. We show the z-transform fails to strip the price-coincident component at weekly resolution, and that even the momentum inverse is untradeable once drawdown is accounted for.
Weekly BTC, 2021–2024 · Sharpe −0.30 · Max DD −80% · negative result
OPT-01 · Volatility TradingA continuously delta-hedged long SPY call, priced at 18% implied vol. Its P&L is the dollar-gamma-weighted gap between realized and implied variance; it bleeds steadily through 2023–2024 because realized sat below implied — a loss the theory predicts before any backtest.
SPY daily, 2023–2024 · Sharpe −6.94 · Max DD −8% · demonstrator
OPT-02 · Event VolatilityA long ATM straddle into scheduled earnings, completing the options strategy SSRN 5921742 specifies but declines to implement. Buying earnings vol loses to the volatility risk premium and the post-report IV crush; the only edge is a selection filter that trades only when the forecast move beats the implied one. On a synthetic Phase-1 surface even the filtered trades do not beat it.
Synthetic, NVDA 7 trades, 2023–2024 · Profit factor 0.82 · bootstrap p 0.83 · demonstrator
Why trades consume, move, and replenish the liquidity they demand. The three dimensions of liquidity; adverse selection as the source of the spread; Kyle's lambda and depth as its reciprocal; permanent vs transient impact and resiliency; the square-root law; and the feedback that drains liquidity in a crash. The theory behind Paper III.
first-principles primer · companion: Paper III (OFI on BTC) · foundational
CON-02 · Non-stationarityA scraped survey of the winning write-ups from Jane Street's two Kaggle competitions plus the firm's own framing, ranking the concepts authors explicitly credit: adapting to non-stationarity (online learning, regime switching) first by a wide margin, then purged time-series CV, autoencoder denoising, training on the true objective, and ensembling — worth ~10× less than adaptation in the '24 winner's own gain ledger.
scraped survey · sources: '21 + '24 Kaggle write-ups, Jane Street brief · survey
No entries match the selected filters.
codes CLA classical · ML machine learning · MIC microstructure · ALT alt-data · OPT options · CON concept note
SEE ALSO: repository · devlog · methodology.md · glossary.md · data-sources.md · latency-notes.md