MES 5-minute — Sweet-Spot Study

Operator's discretionary stack (fractal + MA ribbon + VWMA55 + 九轉序列 TD + VRVP), decoded & backtested · MES · round-trip cost 1 pt · 27931 bars · 2026-01-09 05:10 → 2026-06-08 22:25 (~0.41y) · generated 2026-06-08

Read this first

The fractal Pine script is entry-only — every return here depends on exits we added. "Return" = a single-instrument timing strategy on MES, one position at a time, full equity per trade, net of 1-pt round-trip cost, flattened at session end — not leveraged-futures P&L. The window is ~5 months (one UP regime): high overfitting risk. This headline is the LONG side (the operator's actual signal); the Long vs short vs combined section adds the derived short mirror and the regime caveat. The walk-forward row, not the leaderboard top, is the honest read.

Bottom line

Sweet spot: fractal+TD7+RTH · ATR1.5×RR3 — 26 trades, +2.4% return, Ret/DD 3.28, 50% win, +160 pts ≈ $+799/contract.

Honest caveat: 5 months is one broad regime. Even a config that survives this walk-forward is one bear market away from a different verdict. Treat the sweet spot as a hypothesis to forward-test, not a deployable edge.

What the chart screenshots show

Reading ~7 of the operator's 26 MES 5-minute screenshots, the indicator stack is consistent: green/red fractal triangles at swing lows/highs, the multi-line MA ribbon (white SMA5 hugging price, then the slower lines fanning out), a heavier white VWMA55 line, the 九轉序列 TD count numbers (small 7s and 9s printed at exhaustion points), and the VRVP volume profile histogram down the left edge. The recurring lesson:

These observations drove the confluence rules tested below: fractal + price>VWMA55 + ribbon-rising/stacked + recent-TD-buy is the "good entry" the screenshots depict; bare fractal is the "every triangle" baseline it must beat.

Backtest results — ranked by Ret/DD (risk-adjusted)

27931 bars · 374 config combos · 193 took ≥30 trades (EXCLUDED by the selectivity cap — too busy to be a rare high-conviction setup). Buy & hold here = +6.2% (Ret/DD 0.65, Sharpe 0.96, MaxDD +9.7%).

⚠ Selectivity cap — read before trusting any number

By the operator's deliberate rule, only configs that take fewer than 30 trades in this ~5-month window are eligible to be crowned. The goal is to isolate the rare, high-conviction A+ setups, not the busy high-frequency configs. This is a choice to trade statistical robustness for selectivity: under 30 trades means wide confidence intervals and higher fluke risk. The featured config below takes 26 trades — its win rate, profit factor and Ret/DD are a small-sample read, NOT a statistically proven edge. Treat every crowned config as a hypothesis to forward-test.

+2.4%
Best-config return
3.28
Ret / MaxDD
2.89
Sharpe
+0.7%
Max drawdown
26
Trades
50%
Win rate
2.10
Profit factor
1%
Time in market

Risk-adjusted best: fractal+TD7+RTH entry · ATR1.5×RR3 exit (26 trades, +160 pts ≈ $+799/contract).

best confluence config    buy & hold MES  ·  2026-01-09 05:10 → 2026-06-08 22:25

Top configs (n<30 (selective), ranked by Ret/DD)

EntryExitTradesWinPF ReturnRet/DDSharpeMaxDDExpo
fractal+TD7+RTHATR1.5×RR32650%2.10+2.4%3.282.89+0.7%1%
fractal+TD7+RTHATR2×RR32654%2.17+3.1%3.262.88+1.0%2%
fractal+>vwma55+stacked+TD7+RTHvwmaCross743%3.04+1.2%2.431.96+0.5%0%
fractal+>vwma55+stacked+TD7+RTHATR1×RR1771%2.51+0.3%2.371.70+0.1%0%
fractal+>vwma55+stacked+TD7+RTHtime48743%3.65+1.5%2.262.38+0.7%1%
fractal+>vwma55+stacked+TD7+RTHATR2×RR3743%2.68+0.9%2.252.00+0.4%1%
fractal+>vwma55+stacked+TD7+RTHATR1.5×RR3743%2.61+0.7%2.221.88+0.3%0%
fractal+TD7+RTHATR1×RR32646%1.84+1.4%2.182.14+0.7%1%
fractal+TD7+RTHtime242568%1.82+2.2%2.041.96+1.1%2%
fractal+>vwma55+stacked+RTHATR2×RR32241%1.73+1.6%1.851.78+0.9%2%
fractal+>vwma55+stacked+TD7+RTHATR1×RR3743%2.83+0.6%1.851.82+0.3%0%
fractal+>vwma55+stacked+TD7+RTHtime24771%3.49+1.3%1.842.17+0.7%1%
fractal+TD7+RTHATR1.5×RR22650%1.61+1.3%1.811.77+0.7%1%
fractal+>vwma55+stacked+RTHtime482241%2.16+2.4%1.802.10+1.3%3%
fractal+stacked+RTHtime242955%1.96+2.2%1.782.07+1.2%2%
fractal+TD7+RTHATR1×RR22650%1.73+1.2%1.741.97+0.7%1%
fractal+>vwma55+stacked+RTHtime242348%1.83+1.7%1.721.83+1.0%2%
fractal+TD7+RTHATR2×RR22654%1.62+1.6%1.711.85+1.0%2%
fractal+>vwma55+ribbon↑+TD7+RTH+HVNvwmaCross250%3.05+0.2%1.691.32+0.1%0%
fractal+TD7+RTHtime482454%1.74+2.4%1.681.78+1.4%3%
fractal+stacked+RTHATR2×RR32748%1.67+1.7%1.671.61+1.0%3%
fractal+stacked+RTHtime482744%1.93+2.4%1.591.90+1.5%4%
fractal+TD7+RTHATR1.5×RR1.52650%1.43+0.9%1.541.34+0.6%1%
fractal+>vwma55+stacked+TD7+RTHATR1.5×RR2743%2.00+0.5%1.381.38+0.3%0%
fractal+>vwma55+stacked+TD7+RTHATR2×RR2743%2.00+0.5%1.341.46+0.4%0%
fractal (bare fractal, best exit)time4820055%1.39+9.3%3.012.06+3.1%32%
Buy & hold MES+6.2%0.650.96+9.7%100%

Highlighted = the featured sweet spot, chosen as the best Ret/DD among configs that take fewer than 30 trades (the operator's selectivity cap to surface rare high-conviction setups). Each row shows its trade count so the thinness is visible: a high Ret/DD on a handful of trades is a small-sample read, not a proven edge. The "bare fractal" row is the baseline: confluence filters must beat IT (and buy & hold) to be worth the added complexity.

Walk-forward — is the edge real, or in-sample fit?

Pick the risk-adjusted-best config (by Ret/DD, under the same <30-trade selectivity cap) on the first 60% of bars (in-sample). Then run that exact config, untouched, on the last 40% (out-of-sample). If OOS collapses, the leaderboard was fit. Note: a sub-30-trade config split 60/40 leaves very few OOS trades, so the OOS row here is itself a tiny sample — read it as directional, not conclusive.

WindowConfigTradesWin PFReturnRet/DD SharpeMaxDD
In-sample (60%)fractal+>vwma55+stacked+TD7+RTH · time484 50%13.58 +1.7% 4.443.70 +0.4%
→ same config, OOS (40%)fractal+>vwma55+stacked+TD7+RTH · time483 33%0.66 -0.2% -0.23-0.91 +0.7%
best config fit ON the OOS windowfractal+>vwma55+ribbon↑+TD7+RTH+HVN · ATR2×RR2 1100% +0.2% 17.63 6.44+0.0%

If the OOS "same config" row is far worse than in-sample — and worse than the config you'd have picked WITH hindsight on the OOS window — the edge is mostly fit.

Long vs short vs combined

Regime caveat — read before the numbers

The operator trades SHORT as well as long, so we mirrored the green-triangle BUY signal into a faithful derived SELL signal (fractalDownTrendnot in the operator's Pine, which only plots the long side) and re-ran the whole sweep three ways: long-only, short-only, and combined. But the 5-month sample is a predominantly UP regime. That is the single worst environment in which to judge a short strategy — shorts spend the whole window fighting an upward drift. So whatever the short numbers say here, they are not a fair test of the short signal. We report them honestly and refuse to either bury a bad short result or cherry-pick a lucky short winner.

StrategyTradesWinPF ReturnRet/DDSharpeMaxDD ExpoNote
Long-only — fractal+TD7+RTH · ATR1.5×RR32650%2.10+2.4%3.282.89+0.7%1%the prior study's winner
Short-only — fractal↓+250%5.33+0.7%2.652.55+0.3%0%DERIVED mirror · evaluated in an UP regime
Combined L/S — L/S+>vwma55+stacked+TD7+RTH · time241275%4.79+2.2%3.072.80+0.7%1%one position at a time, either direction
Buy & hold MES+6.2%0.650.96+9.7%100%benchmark

On this data the SHORT side subtracts: the best short-only config looks marginally positive in-sample but does not survive validation, and the combined long+short book does NOT beat long-only on a risk-adjusted, out-of-sample basis. Out-of-sample (the honest read), the frozen short config returned -0.2% (Ret/DD -0.59, 2 trades) — it did not hold up. This is exactly what theory predicts: the 5-month sample (2026-01→06) is a predominantly UP regime — a broad grind higher. Shorting a rising tape is structurally a losing proposition; every short is fighting the drift. We are therefore evaluating shorts in the worst possible environment for them, so a poor short result here is uninformative about whether the short mirror is a good signal — it only confirms you should not short an uptrend.

Long/short verdict

The TD-timed LONG remains the best risk-adjusted config. Adding shorts drags on this window — so the recommendation is unchanged: trade the long, hold the short as a hypothesis. A fair short evaluation NEEDS a sustained down/chop regime; until the tape turns, the short mirror is a hypothesis to forward-test, not a live edge. The short signal is mechanically the exact reflection of the long one, so if the long signal is real, the short one is a credible candidate the moment the regime cooperates — but this window cannot prove it.

Walk-forward — short & combined (out-of-sample, last 40%)

Mode · configWindowTradesReturnRet/DDSharpeMaxDD
Combined L/S · L/S+>vwma55+stacked+TD7+RTH · time24IS 60%8+2.3%5.194.04+0.4%
→ same config OOSOOS 40%4-0.1%-0.18-0.79+0.7%
Short-only · fractal↓+IS 60%2+0.7%2.653.29+0.3%
→ same config OOSOOS 40%2-0.2%-0.59-2.39+0.3%

A short/combined config that survives OOS here still only survived an UP regime — the drawdown that would matter (a sustained selloff) is not in this sample.

What the signals + knobs do

Signal / knobWhat it encodes
fractal (long base)Pine v4 fractalUpTrend: bullish bottom-reversal triangle — down-fractal shape + 3-bar bullish thrust + body expansion + new high + prior EMA(5)<EMA(11). The operator's actual BUY signal.
fractal↓ (short base)DERIVED mirror fractalDownTrend — axis-flipped reflection: up-fractal HIGH shape + 3-bar bearish thrust + body expansion + new low + prior EMA(5)>EMA(11). NOT in the operator's Pine (which only plots the long side); built so the study can test the short direction.
>vwma55 / <vwma55long: close ≥ rolling 55-bar VWMA; short: close ≤ VWMA. Price holding above/below the dynamic S/R.
ribbon↑ / ribbon↓long: close > SMA44 & rising; short: close < SMA44 & falling.
stacked / stacked↓long: SMA22 > SMA44 > SMA120 (stacked bull); short: SMA22 < SMA44 < SMA120 (stacked bear).
TD7 / TD7↓long: DeMark BUY-setup ≥7 within 6 bars (downside exhaustion); short: SELL-setup ≥7 within 6 bars (upside exhaustion).
RTHtime-of-day filter: only the US cash session (09:30-16:00 ET, timezone-aware). Direction-agnostic.
exit ATR k×RRstop = entry − k·ATR(14); target = entry + RR·(entry−stop). k∈{1,1.5,2}, RR∈{1,1.5,2,3}.
exit timeflat after N 5m bars (6/12/24/48 ≈ 30min/1h/2h/4h).
exit vwmaCrossflat when close falls back below VWMA55.
Source: scripts/mes_5m_sweetspot_study.py · long signal decoded from the operator's Pine source + chart screenshots; short signal is a DERIVED mirror (not in the operator's Pine) · exploratory study, NOT a productionized emitter. Returns = single-instrument timing on MES, one position at a time (long, short, or combined), full equity per trade, net of 1-pt round-trip cost, positions flattened at session end — NOT leveraged-futures P&L. 5-month sample = one (UP) regime; shorts are evaluated in their worst environment. Treat as directional. Past performance is not predictive.