Deep-dive summaries of trading strategies that have been systematically backtested — with look-ahead bias prevention, significance testing, transaction costs, and a buy-and-hold benchmark control. Every page shows the full picture: what the strategy claims, what the data actually says, and whether it’s investable.
One page per strategy: explanation, exact signal & filters, applicable markets, full results tables, cost sensitivity, and an honest verdict. New strategies are added as they complete validation.
Buy US small/mid caps that beat consensus EPS by ≥10% and hold 90 trading days. The concordant reaction-day filter adds nothing; the real edge is surprise magnitude (+0.97pp/60d, p=0.0001) — but beta eats it in a realistic portfolio.
1H BOS/CHoCH structure aligned with 4H bias + EMA50 filter across six markets. Gold: Tokyo-session filter and ATR stop lift net avg R 3–4×, all walk-forward confirmed. Silver/SPX cost-wiped; EUR/USD no edge.
18-wave forensic audit of an ex-hedge-fund 'real market edge' framework. Value-area fades complete more often on non-trending days (8/8 assets); reclaim-and-hold sized to daily VA is fee-viable on 6/6 crypto perps (+0.24R BTC / +0.49R ETH).
Four scalping setups across 183 Bybit perps, net of taker fees + slippage. VWAP-deviation fade (+91,893R) and IVA break-in fade (+93,840R, 96.8% WR) survive costs; CVD-divergence fade and IVA breakout lose.
The research program behind the tradeable scalps: 383-perp event studies. Fade family wins (VWAP-fade +0.649 ATR), J1 sweep-reclaim validated wide, L5 crash-warning AUC 0.85. 80% Edge Rule and ITSM falsified; delta dislocation inverted.
Keyless re-implementation of the Royal Trader FX Asset Screener: forecast-surprise + long-term scoring across 9 heatmaps and 28 pairs, COT, CB probabilities, retail sentiment, seasonality. Spreadsheet parity validated cell-by-cell.
230 liquid perps, 90,283 days: open location relative to prior-day value area does NOT predict direction (theories falsified). Real signals: POC shape (+13/−13 bps) and London-session VA (+22 bps P-shape, p<0.0001).
12-month, 546-symbol ORB study. Unfiltered base is negative; short-only + vol/CVD confluence stack trades at +0.042 avg R (n=4,893), beats random-entry by +0.148 and buy-hold by +0.015, block-bootstrap confirmed.
Composite −10..+10 fear/greed score from 5 indicators vs 200-EMA; buy max fear, exit greed. SPY +2.86%, QQQ +7.21% with controlled drawdown but low 10–15% win rate — modest vs buy & hold over the short test window.
Nasdaq ORB (30-min opening range, 1R target, delta filter) tested on 50 Bybit perps × 6 months: aggregate net −$4,234, only 9/50 profitable, walk-forward positive on 15/50. A clean negative transfer result.
Buy new 20-day highs/lows ranked by ATR-normalized strength, 10-day trailing stop, 5y × 4 asset classes net of 10bps. Crypto PF 3.84 net (100% MC profit), stocks PF 1.83; robust across all 30 parameter combos.
Quiet Asia range breaks at 01:00 NY with full-range stop. 2026 window: PF 1.79, +38.5R — but 6-year PAXG out-of-sample: PF 0.90, −220R, 5/7 years lose. The flattering window was a regime artifact.
Replication of Dimaquant's 'Do breakouts work in crypto?' on Bybit top-40: upside 20-day breakouts +76 bps next-day (payoff 1.42), downside −11.7 bps. Signal evidence only — no costs or significance tests.
Rank top-50 Bybit perps by 30-day vol-normalized return; long top-5 (100d MA filter) / short bottom-5 (VNR ≤ −0.5), biweekly, 15% stops. Optimised: +$2,514, Sharpe 1.70 ann, 8.27:1 win/loss ratio.
56-year replication of the Royal Trader SPX down-day strategy: buy at close on ≤−2% open-to-close days, hold 3–24 days. Every config wins >56% with PF >1.29; best config holds OOS (PF 3.05). Lags buy & hold massively.
Short new Bybit perps after pump exhaustion (ATR contraction + below VWAP/EMA50 + >30% launch gain). +17.4% expectancy/trade, PF 2.73, 73.3% WR, 98.9% MC P(exp>0) — crypto-native only, ~2.5 trades/month.
A strategy earns its page here only after passing a fixed evidence pipeline. No cherry-picked windows, no survivorship-blind universes, no compounded-dollars fairy tales.
Point-in-time data only. Entry at the open of the session after the signal is known; consensus estimates verified pre-announcement; BMO/AMC timing honored.
Every long-only result is paired with an equal-weight buy-and-hold of the same universe. The differential vs baseline, not the compounded equity, is the signal.
Welch t-tests, binomial win-rate tests, and p-values on the per-event differential. If p ≥ 0.05 the page says so.
Commission + slippage + market impact modeled at multiple round-trip cost levels (0–50bps). The cost at which the edge dies is reported.
Sequential/fixed-fraction (upper bound) and concurrent portfolio with daily mark-to-market (real economics). Max-exposure capped.
Universes built from point-in-time snapshots (e.g. Jan-2016 R2000) so delisted names are counted, not airbrushed.
Every page ends with an explicit “trade it / don’t / only as a tilt” verdict. Negative results are published as loudly as positive ones.
Source repo, data build scripts, and config are linked from every page. Everything regenerable from free keyless sources.
All backtest code, data build scripts, and raw result CSVs live in the jahrfm/strategy-summaries repo. Each strategy page links to its own evidence-producing research repo.
View on GitHub →