OPEN-SOURCE SCRIPT
更新済 Auto Trend Channel & Breakout Expectancy in R
OVERVIEW
An auto-drawn trend channel is easy. Knowing whether breaking it has ever paid on YOUR symbol is not — and it is the only thing that matters.
This tool draws the channel objectively and then does the part nobody else does: it forward-resolves every historical break through a real target and stop, and reports, in R, what breaks like the current one have actually returned. Not a prediction. A track record.
Expectancy exp +1.24R · need 15% @ RR 6.0
HV win/loss 38/59% n400 ✓
MV win/loss 32/65% n150 ✓
LV win/loss 17/83% n55
Null / net net 1.29R t=3.1 ✓
It is a research and framing tool. NOT a strategy, NOT a signal service, NOT a validated edge.
HOW THE CHANNEL IS BUILT — objectively, no hand-placed anchors
The support and resistance lines come from a CONVEX HULL over confirmed pivots, not from two points a human chose. The opposite boundary is drawn parallel, and the channel WIDTH becomes the measured-move target. Because the construction is mechanical, the same rule produces the same channel on every symbol and every timeframe — there is nothing to curve-fit.
HOW A BREAK IS GRADED — and why volume is split three ways
A break is a confirmed CLOSE beyond the boundary. Every break is then sorted by its break-bar volume percentile into HV / MV / LV (high / medium / low), and each tier is calibrated SEPARATELY.
That split is the point. "Breakouts need volume" is repeated everywhere and tested almost nowhere. Here you can read straight off the panel whether heavy breaks on your instrument actually pay more than thin ones — and often they do not, which is worth knowing before you wait for a volume confirmation that costs you the entry.
THE FORWARD TEST — expectancy in R, on identical geometry
Every break is resolved through a triple barrier: a target at the channel-width move, a stop at a fixed ATR distance, and a horizon. Win, loss, or chop. The expectancy is the exact realized R per trade — a win contributes +RR, a loss contributes −1, chop contributes 0 — so the win rate and the R:R are on the SAME footing, computed from the same resolved trades. There is no mixing of a win rate from one calculation with an R:R from another.
The R:R shown is the target-to-stop RATIO, and it is deliberately not the headline. A 6:1 target sounds wonderful and is reached maybe a third of the time; quoting it alone flatters a low-hit-rate system. So the panel leads with the realized EXPECTANCY, then states the breakeven win rate the geometry requires, then shows the win rate actually achieved. Expectancy first, because expectancy is the thing that is actually true.
Nothing is rated until a tier has at least 30 resolved trades. Below that, no checkmark, no verdict, no meta call — a Wilson floor on a dozen trials is too wide to certify anything, and pretending otherwise is how bad tiers get a green tick.
THE RANDOM-DIRECTION NULL — the honesty check
Here is the trap this defends against. A 6:1 target with a nearby stop is an asymmetric payoff, and on a drifting instrument an asymmetric payoff makes money in a RANDOM direction — the geometry earns, not the signal. So for every real break the script ALSO runs a coin-flip-direction trade with the identical target and stop, and reports its expectancy alongside.
If the null earns as much as the signal, the "edge" is just the geometry capturing drift, and you should believe the null. And it is now TESTED, not just displayed: a Welch t between the signal's realized-R distribution and the null's turns "net 1.29R" into "net 1.29R, t = 3.1, real" — or into "ns", not significant, which is the more common and more honest outcome.
THE OPTIONAL META-LABEL — a learned second opinion
An online logistic model trains on this script's own resolved outcomes and outputs a calibrated probability of follow-through for the current break. No hand-tuned weights; it self-corrects as more breaks resolve, and it reports its own reliability (the realized win rate inside each probability bin) so you can see whether its confidence is earned. It is off by default and it never overrides the forward test — it is a second opinion, not the verdict.
THE ANTI-BIAS GUARDS
ENTRY IS THE CLOSE (or, in retest mode, the line touch) — a break is a signal, not a fill, and entering at a better price than the trade actually offered would manufacture an edge.
Both barriers on one bar: the STOP is assumed first — conservative, and the only assumption that cannot flatter the result. Unresolved trades at the horizon are scored as chop (zero), not as wins.
THE NULL is direction-randomised on the same geometry, which is what isolates signal from payoff structure. Every rating is gated by a minimum sample and a Wilson lower bound, and the null gap is significance-tested. Nothing here is asserted that has not cleared a test.
NON-REPAINT
Confirmed pivots, confirmed-close breaks, and forward-resolved calibration that never looks ahead. The channel can extend and re-fit as new pivots confirm, but a resolved trade is never re-scored, and no barrier is evaluated on a bar that has not closed.
DATA AND SCOPE
Any symbol, any timeframe. ATR-normalised throughout. Volume is used for the tier split; without it the tiers collapse to one and everything else still works.
EXPORTS (Data Window — consume from other scripts via input.source())
Channel high, channel low, break direction, volume tier, entry, stop, target, per-tier expectancy, null expectancy, meta probability — see the Exports group.
CONCEPT CREDIT
The convex-hull approach to objective trendlines, the triple-barrier forward-labelling method and the meta-labelling idea are from Marcos Lopez de Prado ("Advances in Financial Machine Learning"). The trading-range break as a formal, testable rule was first studied at scale by Brock, Lakonishok and LeBaron (Journal of Finance, 1992); its vulnerability to data-snooping was shown by Sullivan, Timmermann and White (1999) — which is exactly why this tool measures the rule on your instrument instead of asserting it. Wilson score interval — E. B. Wilson (1927). Welch's t-test — B. L. Welch. ATR — J. Welles Wilder.
The convex-hull channel implementation, the volume-tier calibration, the random-direction null with its significance test, and the online meta-label are the author's own. Clean-room implementation; no third-party Pine code is reused. Not affiliated with, nor endorsed by, any of the above.
HONESTY AND LIMITATIONS
Calibration is IN-SAMPLE, fixed-barrier, and assumes idealised fills — no commission, no slippage, no partial fills. It describes the past; it does not predict the future. Real costs will reduce every number, and they reduce the low-hit-rate high-RR tiers most.
Overlapping windows mean the samples are not fully independent, so treat the t-stats as directional evidence, not exact p-values. A tier that is proven in-sample is not guaranteed out-of-sample.
The null being beaten is the single most important line on the panel, and on many instruments it will read "ns". That is a real result. A tool that cannot report "no edge here" is selling you something.
Nothing in this script predicts price.
DISCLAIMER
Research and educational tool only. NOT financial advice, NOT a recommendation, and NO guarantee of results. Entry, stop and target output is arithmetic, not advice. Trading carries risk of loss. Test out-of-sample and make your own decisions. The author accepts no liability for any use.
An auto-drawn trend channel is easy. Knowing whether breaking it has ever paid on YOUR symbol is not — and it is the only thing that matters.
This tool draws the channel objectively and then does the part nobody else does: it forward-resolves every historical break through a real target and stop, and reports, in R, what breaks like the current one have actually returned. Not a prediction. A track record.
Expectancy exp +1.24R · need 15% @ RR 6.0
HV win/loss 38/59% n400 ✓
MV win/loss 32/65% n150 ✓
LV win/loss 17/83% n55
Null / net net 1.29R t=3.1 ✓
It is a research and framing tool. NOT a strategy, NOT a signal service, NOT a validated edge.
HOW THE CHANNEL IS BUILT — objectively, no hand-placed anchors
The support and resistance lines come from a CONVEX HULL over confirmed pivots, not from two points a human chose. The opposite boundary is drawn parallel, and the channel WIDTH becomes the measured-move target. Because the construction is mechanical, the same rule produces the same channel on every symbol and every timeframe — there is nothing to curve-fit.
HOW A BREAK IS GRADED — and why volume is split three ways
A break is a confirmed CLOSE beyond the boundary. Every break is then sorted by its break-bar volume percentile into HV / MV / LV (high / medium / low), and each tier is calibrated SEPARATELY.
That split is the point. "Breakouts need volume" is repeated everywhere and tested almost nowhere. Here you can read straight off the panel whether heavy breaks on your instrument actually pay more than thin ones — and often they do not, which is worth knowing before you wait for a volume confirmation that costs you the entry.
THE FORWARD TEST — expectancy in R, on identical geometry
Every break is resolved through a triple barrier: a target at the channel-width move, a stop at a fixed ATR distance, and a horizon. Win, loss, or chop. The expectancy is the exact realized R per trade — a win contributes +RR, a loss contributes −1, chop contributes 0 — so the win rate and the R:R are on the SAME footing, computed from the same resolved trades. There is no mixing of a win rate from one calculation with an R:R from another.
The R:R shown is the target-to-stop RATIO, and it is deliberately not the headline. A 6:1 target sounds wonderful and is reached maybe a third of the time; quoting it alone flatters a low-hit-rate system. So the panel leads with the realized EXPECTANCY, then states the breakeven win rate the geometry requires, then shows the win rate actually achieved. Expectancy first, because expectancy is the thing that is actually true.
Nothing is rated until a tier has at least 30 resolved trades. Below that, no checkmark, no verdict, no meta call — a Wilson floor on a dozen trials is too wide to certify anything, and pretending otherwise is how bad tiers get a green tick.
THE RANDOM-DIRECTION NULL — the honesty check
Here is the trap this defends against. A 6:1 target with a nearby stop is an asymmetric payoff, and on a drifting instrument an asymmetric payoff makes money in a RANDOM direction — the geometry earns, not the signal. So for every real break the script ALSO runs a coin-flip-direction trade with the identical target and stop, and reports its expectancy alongside.
If the null earns as much as the signal, the "edge" is just the geometry capturing drift, and you should believe the null. And it is now TESTED, not just displayed: a Welch t between the signal's realized-R distribution and the null's turns "net 1.29R" into "net 1.29R, t = 3.1, real" — or into "ns", not significant, which is the more common and more honest outcome.
THE OPTIONAL META-LABEL — a learned second opinion
An online logistic model trains on this script's own resolved outcomes and outputs a calibrated probability of follow-through for the current break. No hand-tuned weights; it self-corrects as more breaks resolve, and it reports its own reliability (the realized win rate inside each probability bin) so you can see whether its confidence is earned. It is off by default and it never overrides the forward test — it is a second opinion, not the verdict.
THE ANTI-BIAS GUARDS
ENTRY IS THE CLOSE (or, in retest mode, the line touch) — a break is a signal, not a fill, and entering at a better price than the trade actually offered would manufacture an edge.
Both barriers on one bar: the STOP is assumed first — conservative, and the only assumption that cannot flatter the result. Unresolved trades at the horizon are scored as chop (zero), not as wins.
THE NULL is direction-randomised on the same geometry, which is what isolates signal from payoff structure. Every rating is gated by a minimum sample and a Wilson lower bound, and the null gap is significance-tested. Nothing here is asserted that has not cleared a test.
NON-REPAINT
Confirmed pivots, confirmed-close breaks, and forward-resolved calibration that never looks ahead. The channel can extend and re-fit as new pivots confirm, but a resolved trade is never re-scored, and no barrier is evaluated on a bar that has not closed.
DATA AND SCOPE
Any symbol, any timeframe. ATR-normalised throughout. Volume is used for the tier split; without it the tiers collapse to one and everything else still works.
EXPORTS (Data Window — consume from other scripts via input.source())
Channel high, channel low, break direction, volume tier, entry, stop, target, per-tier expectancy, null expectancy, meta probability — see the Exports group.
CONCEPT CREDIT
The convex-hull approach to objective trendlines, the triple-barrier forward-labelling method and the meta-labelling idea are from Marcos Lopez de Prado ("Advances in Financial Machine Learning"). The trading-range break as a formal, testable rule was first studied at scale by Brock, Lakonishok and LeBaron (Journal of Finance, 1992); its vulnerability to data-snooping was shown by Sullivan, Timmermann and White (1999) — which is exactly why this tool measures the rule on your instrument instead of asserting it. Wilson score interval — E. B. Wilson (1927). Welch's t-test — B. L. Welch. ATR — J. Welles Wilder.
The convex-hull channel implementation, the volume-tier calibration, the random-direction null with its significance test, and the online meta-label are the author's own. Clean-room implementation; no third-party Pine code is reused. Not affiliated with, nor endorsed by, any of the above.
HONESTY AND LIMITATIONS
Calibration is IN-SAMPLE, fixed-barrier, and assumes idealised fills — no commission, no slippage, no partial fills. It describes the past; it does not predict the future. Real costs will reduce every number, and they reduce the low-hit-rate high-RR tiers most.
Overlapping windows mean the samples are not fully independent, so treat the t-stats as directional evidence, not exact p-values. A tier that is proven in-sample is not guaranteed out-of-sample.
The null being beaten is the single most important line on the panel, and on many instruments it will read "ns". That is a real result. A tool that cannot report "no edge here" is selling you something.
Nothing in this script predicts price.
DISCLAIMER
Research and educational tool only. NOT financial advice, NOT a recommendation, and NO guarantee of results. Entry, stop and target output is arithmetic, not advice. Trading carries risk of loss. Test out-of-sample and make your own decisions. The author accepts no liability for any use.
リリースノート
v1.1 — Correctness + export cleanup- Correctness: in the order-flow VWAP-anchor proxy, a pivot-high call sat behind a short-circuiting
OR and was skipped on some bars, which could corrupt its rolling buffer. Both pivots now evaluate
on every bar, as Pine requires for series functions. Clears the compiler warning.
- Housekeeping: the Data Window exports are renamed to a consistent EXP_ prefix (EXP_ResLine,
EXP_SupLine, EXP_Break, EXP_VolTier, EXP_ChanWidth, EXP_Entry, EXP_Stop, EXP_Target, EXP_SetupState,
EXP_ContProb, EXP_MetaCall, EXP_MetaSamples). The values are identical — only the titles changed.
Note: if you wired the old "Break Direction" export into another chart's confirmation input, re-select
it as "EXP_Break" in the source dropdown after updating.
Descriptive research tooling, in-sample statistics, not investment advice.
オープンソーススクリプト
TradingViewの精神に則り、このスクリプトの作者はコードをオープンソースとして公開してくれました。トレーダーが内容を確認・検証できるようにという配慮です。作者に拍手を送りましょう！無料で利用できますが、コードの再公開はハウスルールに従う必要があります。
免責事項
これらの情報および投稿は、TradingViewが提供または承認する金融、投資、取引、またはその他の種類の助言もしくは推奨であることを意図したものではなく、またこれらに該当するものでもありません。詳細は利用規約をご覧ください。
オープンソーススクリプト
TradingViewの精神に則り、このスクリプトの作者はコードをオープンソースとして公開してくれました。トレーダーが内容を確認・検証できるようにという配慮です。作者に拍手を送りましょう！無料で利用できますが、コードの再公開はハウスルールに従う必要があります。
免責事項
これらの情報および投稿は、TradingViewが提供または承認する金融、投資、取引、またはその他の種類の助言もしくは推奨であることを意図したものではなく、またこれらに該当するものでもありません。詳細は利用規約をご覧ください。