Backtesting evidence

SSO · spec.md

Derived research evidence · docs/backtesting/evidence/sso/spec.md

SHA-256: 5109fde755eda30547990a143c040ac5d5a75fa857efe0757df06feef1a62a9b

Original source text. A superseded report retains its original assertions; it does not replace a corrected finding. Reproduction code is displayed without execution.

# SSO Opus 5.5 Swing 15m — Backtest Spec (FROZEN)

Source: public TrendSpider share page for "SSO Opus 5.5 Swing 15m", read 2026-11-04.
This spec records the page's rule plus frozen interpretations. Nothing beyond the
page and the interpretations below was invented; every interpretation that could
change results is flagged.

## Page rule (as stated)

- Ticker SSO only, 15-minute bars, regular session only, long only, no short rules.
- Entry, all evaluated at the signal-bar close:
  - SMA(50) > SMA(200)
  - ADX(10) > 20
  - AND (Setup A OR Setup B)
  - Setup A (breakout + volume): close > RANGE(14) high evaluated 1 candle ago;
    bar volume >= 200% of the 20-bar SMA of volume ("by at least 100%").
  - Setup B (VWAP cross + RSI + volume): session day-to-date anchored VWAP crossed
    upward by close (prior bar close <= VWAP and signal bar close > VWAP);
    RSI(14) > 50; bar volume >= 150% of the 20-bar SMA of volume
    ("by at least 50%").
- Exits, all evaluated after candle closes:
  - Stop loss if down more than 3.5%.
  - Take profit if up more than 15%.
  - Exit after 25 candles if PnL < 0%.
  - Exit after 60 candles if PnL < 1%.
  - Trailing stop 7%.
- Page settings: trade by open; market data regular session.
- Page claims (NOT targets — different window): 153% strategy vs 111%
  buy-and-hold, 40% win rate, 4.89 reward/risk, 0.96% avg position return,
  -12.50% drawdown, equity curve May '22–Nov '26. Position sizing and trade
  count are NOT stated on the page.

## Frozen interpretations (assumptions; flagged where results-sensitive)

1. RANGE(14) high 1 candle ago = highest high over the 14 bars ending at the
   prior candle (signal bar excluded). [Could change results if the page meant
   the 14 bars ending at the signal bar.]
2. Anchored VWAP = session VWAP from the 09:30 ET open, reset daily, computed
   from typical price (H+L+C)/3 x volume, cumulative within the session.
   The cross uses the prior bar's close vs the prior bar's VWAP and the signal
   bar's close vs the signal bar's VWAP.
3. ADX(10) = standard Wilder ADX, period 10 (Wilder-smoothed TR, +DM, -DM;
   DX = 100*|+DI - -DI| / (+DI + -DI); ADX = Wilder-smoothed DX over 10).
4. RSI(14) = standard Wilder RSI, period 14.
5. Volume SMA(20) = rolling mean of bar volume INCLUDING the signal bar.
   [Could change results slightly if the page meant the prior 20 bars only.]
6. SMA(50)/SMA(200) = simple moving averages of 15-minute closes, including
   the signal bar.
7. Entry fill = next regular-session bar open after the signal bar. All exit
   fills = next regular-session bar open after the bar whose close first
   satisfies the exit condition. "Next bar" is the next bar in the filtered
   regular-session series, so a fill may land on the next session's 09:30 open.
8. Stop / take-profit PnL is measured from the entry fill price to the
   evaluating bar's close: stop fires when close/entry - 1 < -0.035; take
   profit fires when close/entry - 1 > 0.15.
9. Trailing stop: fires when the evaluating bar's close is below 93% of the
   highest close observed since entry (highest close tracked over bars from
   the entry-fill bar onward, including the evaluating bar's own close, which
   can never trigger against itself).
10. Time exits: bars held counts bars closed from the entry-fill bar onward,
    inclusive. Exit after 25 candles fires at the first close with bars held
    >= 25 and PnL < 0%; exit after 60 candles fires at the first close with
    bars held >= 60 and PnL < 1%. A trade with PnL >= 1% at 60 candles is NOT
    time-exited; it runs until a stop / take-profit / trailing exit fires.
11. Exit-check order on a single close when several fire: stop, take profit,
    trailing, then time exits (25-candle, then 60-candle).
12. One position at a time; no new entry while in a position. A bar's close is
    evaluated for entry only when flat at that close. Full capital per trade
    (sizing unstated; returns are per-trade percentages, so this does not
    change % returns).
13. Costs: gross returns and net returns (gross minus 10 bps round-trip per
    trade) are both reported. No leverage beyond SSO itself (SSO is already
    2x S&P 500 — no leverage is added).
14. A trade whose exit fill bar does not exist (exit condition on the final
    bar of the dataset) is INCOMPLETE and excluded from completed-trade
    statistics; it is listed separately. A signal on the final bar (no fill
    bar) produces no trade.
15. Trade-to-split assignment: by the signal bar's session date.

## Data

- Yahoo Finance 15-minute bars for SSO, prepost=False, explicit windows under
  60 days (a 60-day request is rejected; range=max on intraday silently
  returns monthly and is never used).
- Window fetched: 2026-09-06 to 2026-11-04 (59-day request), yielding bars
  2026-09-08 through 2026-11-03. Regular-session bars only (09:30–15:45 ET
  starts, 26 bars per session).
- Complete sessions: sessions with all 26 regular bars present. Result: 41
  complete sessions, 1,066 bars, zero gaps, zero missing bars.
- Data anomaly recorded, NOT cleaned (cleaning would deviate from the rule):
  a small number of bars carry volumes far above the median (~67k), e.g.
  2026-11-03 14:45 ET with 13,053,077 shares. These bars affect the
  volume-SMA ratios and session VWAP exactly as Yahoo delivered them.

## Split and verdict bar (frozen)

- In-sample = first 60% of complete sessions = first 24 sessions (by date);
  validation = remaining 17 sessions. (24 = floor(0.6 * 41).)
- Verdict bar: validation needs n >= 30 completed trades AND mean net return
  per trade significantly positive (t-stat reported) AND total net return
  ahead of SSO buy-and-hold over the same validation window on a comparable
  basis. If validation n < 30, verdict is INDETERMINATE regardless of mean.
  If validation mean is negative, it is NOT a near-miss.
- Buy-and-hold basis: last validation-window bar close / first
  validation-window bar open - 1. Strategy equity compounds completed-trade
  net returns from 1.0; idle periods earn 0. Comparability caveat: the
  strategy is not always invested, buy-and-hold is.
- t-stat = mean / (std / sqrt(n)), std with ddof=1.

## Honesty constraint

The page's May '22–Nov '26 claim CANNOT be reproduced from Yahoo 15m data
(60-day cap). This test covers only the recent Yahoo window. Its verdict
covers the recent-window replication only and neither confirms nor refutes
TrendSpider's full-period 153% claim.
connecting…