ONDS · onds-orb-ema-backtest-report.md
Derived research evidence · docs/backtesting/evidence/onds/onds-orb-ema-backtest-report.md
SHA-256: 6e67ee6c8578804084c1eb4eab2541be6876c3897abb4c811b3d06805c05a496
Original source text. A superseded report retains its original assertions; it does not replace a corrected finding. Reproduction code is displayed without execution.
# ONDS 1h ORB hold till EMA20 break — backtest report Source: TrendSpider shared strategy (published 2026-10-28). Spec (frozen before running): `~/workspace/onds-orb-ema-backtest-spec.md`. Script: `~/workspace/onds-orb-ema-backtest.py`. Data: Yahoo 60-min regular-session bars, 2024-10-29 → 2026-10-28 (3,484 bars used, 14 null bars dropped). Yahoo caps 60-min history at ~2y; that cap IS the sample. ## Rules as tested - Entry: 60-min close above the day's opening-range high (OR = high of the first 60-min bar of the day) AND bar volume >= 1.3x the 20-bar SMA of volume. - Exit: close crosses below EMA(20) on 60-min closes. - Execution at next bar's open. Long only, one position, 100% capital, no leverage, no pyramiding. Costs 10 bps round trip; totals below are net. ## Results (net of costs) - Trades: 56 (1 forced close at sample end). Entry signals: 56. - Win rate: 33.9% (TrendSpider claims 46%). - Mean trade: +5.48%, t = 2.24. Avg win +23.62%, avg loss -3.83%, win/loss 6.17 (claims 4.15). Profit factor 3.17. Median hold 13 bars. - Strategy total: +894.1% vs buy-and-hold +688.4% (claims 2159% vs 723%). - Max drawdown (marked to market): -41.8% (claims -38.30%). ## Pre-registered PASS bar — MET on all three legs (a) Net beats B&H: yes (+894% vs +688%). (b) Mean trade > 0 with t >= 2.0: yes (+5.48%, t = 2.24). (c) n >= 30: yes (56). VERDICT: PASS. ## Caveats (read before building) 1. TrendSpider's 2159% was NOT replicated (ours: 894%). Costs don't explain the gap (gross is only ~6pp higher). Likely causes: a different/longer sample (their "as of" date is blank, so their window is unknown — our B&H 688% vs their 723% suggests a similar era but not the same window), or a different opening-range length (undisclosed on the page; a sub-hourly OR cannot be computed from 60-min bars, so it is untestable with this data). 2. The edge is front-loaded: first 28 trades averaged +10.38%, last 28 averaged +0.59%. The system beat B&H overall, but recent trades are nearly flat — this looks like it rode ONDS's 2025 run more than a steady edge. 3. Single-ticker, single-regime risk: 56 trades on one small-cap. The win profile (lose ~4% often, win ~24% occasionally) is coherent trend-following, but it is one ticker's story, not a diversified system. 4. Volume-SMA sensitivity: using the prior-20-bar average instead of the including-current-bar average gives 57 trades at +4.83% mean — same conclusion. ## Build status PASS means build is authorized, but the form is open: the strategy is intraday (60-min) and single-ticker, so it does NOT fit a daily screener preset (that would be a different, untested strategy), and Strategy Lab's DSL has EMA/volume/cross but no opening-range primitive and no intraday interval in its runner — a faithful in-app build needs intraday bars + an OR primitive. Build form awaiting user pick. ## Generalization test (2026-10-29, same frozen rules, 8 tickers, same window) Basket frozen before running: SPY, QQQ, IWM, NVDA, TSLA, PLTR, SMCI, SOUN. Per ticker (n / mean trade / t / strategy vs B&H): SPY 64 / -0.00% / -0.03 / -0.7% vs +33.1%; QQQ 43 / +0.31% / 1.29 / +13.9% vs +46.3%; IWM 40 / +0.45% / 1.09 / +18.3% vs +27.3%; NVDA 37 / -0.52% / -0.91 / -19.2% vs +53.5%; TSLA 39 / -0.47% / -0.52 / -21.2% vs +53.4%; PLTR 43 / +2.16% / 1.49 / +114.1% vs +338.2%; SMCI 51 / +2.34% / 0.95 / +95.9% vs -25.1%; SOUN 61 / +0.39% / 0.36 / +5.3% vs -26.2%. Pooled ex-ONDS: 378 trades, mean +0.61%, t=1.42 (below the 2.0 bar). CONCLUSION: the edge does NOT generalize. It beats B&H only on SMCI/SOUN, both negative-B&H crash windows, and lags B&H on every benchmark and mega-cap. The ONDS pass stands on its own bar, but the strategy is ONDS-specific — the Strategy Lab build must carry ONDS-only labeling with this result stated.