Skip to content
#small language model Open access

A Stopping Rule Changes the Shape of the Profit and Loss, Not the Direction — Across Twenty Rules the Win Rate Runs from 0.0562 to 0.9443, and a Synthetic Random Walk with Zero Directional Information Moves Just as Much (Correlation +0.9922) [F026]

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

「損小利大にする」「建値に逃がす」「トレールで伸ばす」。どれもいつ切るかの規則であって、どちらへ建てるかの規則ではない。だから方向の情報は原理的に増えない。では停止則を変えると何が動くのか。本稿の分離子はいつ切るかであり、入口・データ・期間・銘柄をすべて固定したまま、停止則だけを二十通りに変える。方向とそれ以外を分けるため、同じ入口から買いと売りを両方走らせ、M =(買 + 売)÷ 2 と D =(買 − 売)÷ 2 に分ける。M は方向の効果を消し、D に方向とドリフトが入る。 四銘柄(USDJPY・EURUSD・GBPUSD・XAUUSD)の2019年から2024年までの1時間足で、全部の足の終値から建て、固定(利確/損切り 4 通り)・トレール(4)・建値移動(3)・時間だけ(3)・利確のみ(3)・損切りのみ(3)の二十通りを走らせた。期限は 24 本、コストは入れていない。曖昧な足(同じ足で両側に触れる)は保守的に倒し、トレールの最大益は前の足までで更新する。形は完全に変わる。勝率は 0.0562 から 0.9443(差 0.8881)、中央値は −40.0000 から +40.0000 pips(差 80.0000)、歪度は −8.4093 から +8.4366(差 16.8459)まで動く。USDJPY の「利確のみ 10」は勝率 0.8325 だが最悪は −702.30 pips で、それは勝率に現れない。同じ USDJPY の「建値 5/SL20」は勝率 0.1076、中央値 0.0000 である。勝率と歪度の相関は −0.9345 で、勝率が高い則ほど小さな勝ちを積んで大きく負ける形になる。勝率は良し悪しではなく形を測っている。 本稿の芯は基準線にある。足の中まで独立な正規増分で作った合成系列、つまり方向の情報が定義からゼロの系列で同じ二十通りを走らせると、勝率は実測と相関 +0.9922、差の中央 −0.0056(幅 −0.0550 から +0.0570)である。どの停止則がどんな勝率を出すかは、当たっていない系列でもほぼ同じである。形は停止則の性質であって、市場の性質ではない。 測る前に書いた予言②の分類は外れた。「対称な停止則では M が恒等的に 0」と書いたが、宣言時に対称とした 36 通りの M は最大の絶対値 3.597932 pips ある。測ってから三つに分け直した。① 柵の無い則(時間だけ)では M はちょうど 0.000000 で、買いと売りが経路ごとに打ち消す恒等式である(12 セル)。② 固定の利確 = 損切りでは、M は −(曖昧率 × 柵) にぴったり一致し、8 セルすべてで差は 0.0000 pips である。負の M は市場ではなく、曖昧な足を保守的に倒して買いにも売りにも損を付けたことそのものである。③ 残り 60 セルでは打ち消さない(M 中央 +0.3573 pips・幅 −4.4154 から +2.1625)。トレールと建値移動を対称と宣言したのは誤りで、買いは最大益から下へ、売りは最小益から上へ追うので経路ごとに打ち消さない。なお測ってから恒等式をひとつ見つけた。「利確のみ」の M と「損切りのみ」の M は符号が逆で絶対値が等しく、ずれは 0.000000 pips である。売りの「利確のみ」は買いの「損切りのみ」の符号を変えたものだからで、これは検定ではない。 一方、買いの期待値そのものは −4.3840 から +27.4471 pips まで動くが、その大半はドリフトである。ドリフトの無い合成乱数でも買いの期待値は中央 +1.5707 pips、幅 −14.9559 から +20.4750 pips まで散る。一本の経路には、その経路の実現したドリフトが乗るからである。だから買いの期待値だけを見て市場を語ってはいけない。結論は、停止則が変えるのは損益の形であって方向ではない、ということである。 この論文の位置づけ:新しい数学定理も新しい法則も主張しない。任意停止定理、トリプルバリア法、トレーリングストップ、ランダムウォークはすべて既知である。特定の教材やその著者、販売元は主題ではなく、一切名指していない。コストを入れていないので、本稿の数字は売買の成績ではない。曖昧な足を保守的に倒しており、その規約の効果は M の負の分としてちょうど説明できる。トレールの最大益を前の足までで更新しており、足の中で更新する規約なら数字が変わるが測っていない。重複を除いていないので t 値を報告しておらず、有意性は論じていない。停止則を二十通りに固定しており、分割決済・ピラミッディング・ボラティリティ連動の幅は扱っていない。測定は日中・少数銘柄・四銘柄六年に限られ、月次の時系列モメンタムのようにより長い保有で文書化されている結果を否定するものではない。 作成にあたって:本稿の着想と内容は、著者自身の考察に基づくものです。文章の構成整理や英訳、集計に用いたスクリプトの作成には AI(大規模言語モデル)の助力を得ました。最終的な内容の解釈や誤りがあれば、それらはすべて著者の責に帰します。お気づきの点があれば、ご教示いただければ幸いです。 ----- "Cut losses short and let profits run", "move the stop to breakeven", "trail it out". Every one of them is a rule about when to exit, not about which way to enter, so none can add directional information in principle. What then does changing the stopping rule move? The separator here is when to exit: the entries, the data, the period and the instruments are all held fixed while the stopping rule is varied over twenty possibilities. To separate direction from everything else, both the buy and the sell are run from the same entry and split into M = (buy + sell) / 2 and D = (buy − sell) / 2; M removes the directional effect while D carries the direction and the drift. Using one-hour bars for four series (USDJPY, EURUSD, GBPUSD, XAUUSD) from 2019 to 2024, entries were taken at the close of every bar and run under twenty rules: four fixed take-profit/stop-loss pairs, four trailing stops, three breakeven moves, three time-only exits, three take-profit-only and three stop-loss-only. The deadline is 24 bars and no costs are included. Ambiguous bars — those touching both sides — are resolved adverse-first, and the trailing maximum updates only up to the previous bar. The shape changes completely: the win rate runs from 0.0562 to 0.9443 (a spread of 0.8881), the median from −40.0000 to +40.0000 pips (a spread of 80.0000) and the skewness from −8.4093 to +8.4366 (a spread of 16.8459). On USDJPY, "take-profit only 10" has a win rate of 0.8325 with a worst of −702.30 pips, which the win rate does not show; the same instrument's "breakeven 5/SL20" has a win rate of 0.1076 and a median of 0.0000. The correlation between win rate and skewness is −0.9345: the higher the win rate, the more it becomes many small wins and a few large losses. The win rate measures the shape, not whether anything is good. The core of the paper is the baseline. Running the same twenty rules on a synthetic series built from independent normal increments down to the movement inside each bar — zero directional information by construction — gives win rates correlating with the measurement at +0.9922, with a median difference of −0.0056 (range −0.0550 to +0.0570). Which stopping rule gives which win rate is nearly the same on a series that is right about nothing. The shape is a property of the stopping rule, not of the market. The classification in prediction ②, written before measuring, failed. It said "for a symmetric rule M is identically 0", but across the 36 rules declared symmetric the largest absolute M is 3.597932 pips, so they were reclassified into three groups after measuring. ① For rules with no barriers (time only), M is exactly 0.000000: the buy and the sell cancel path by path, an identity (12 cells). ② For fixed take-profit = stop-loss, M matches −(ambiguous share × barrier) exactly, with all 8 cells differing by 0.0000 pips; the negative M is not the market but precisely the act of resolving ambiguous bars adverse-first, which puts a loss on both sides. ③ The remaining 60 cells do not cancel (median M +0.3573 pips, range −4.4154 to +2.1625): declaring the trail and the breakeven move symmetric was wrong, since the buy trails down from its maximum and the sell trails up from its minimum. One identity was also found after measuring: the M of "take-profit only" and of "stop-loss only" are equal in magnitude and opposite in sign, differing by 0.000000 pips, because the sell's take-profit-only is the buy's stop-loss-only with the sign flipped. That is not a test. The expected value of the buy alone, meanwhile, runs from −4.3840 to +27.4471 pips, but most of that is drift. Even the driftless synthetic series gives a median of +1.5707 pips and a range of −14.9559 to +20.4750, because a single path carries the drift that path happened to realize. The expected value of the buy alone must therefore not be used to talk about the market. The conclusion is that a stopping rule changes the shape of the profit and loss, not the direction. Scope of this paper: no new mathematical theorem and no new law is claimed; optional stopping, the triple-barrier method, trailing stops and random walks are all known. No real course, author, or vendor is the subject here, and none is named. No costs are included, so the figures are not trading performance. Ambiguous bars are resolved adverse-first, and the effect of that convention is exactly what the negative part of M is. The trailing maximum updates only up to the previous bar; a convention that updates within the bar gives other figures, which were not measured. The overlap is not removed, so no t value is reported and significance is not discussed. The stopping rules are fixed at twenty, and scaling out, pyramiding and volatility-linked widths are not treated. The measurement is intraday and covers four instruments over six years, and does not contradict results documented at longer holding periods, such as monthly time-series momentum. Acknowledgement: The ideas and content of this paper stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and writing the scripts used for the measurement. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated.

View source

Similar papers

#small language model Dataset Open access Oct 2026

Socratic guiding questions in synthetic arithmetic data: matched LoRA runs (revision v2)

Supporting data, adapters, predictions and code for the article *Low-Cost LoRA Fine-Tuning of Small Language Models for Multi-Step Arithmetic Reasoning* by Jake O'Grady, Asena Isik Gürhan, Chee Fong Ting and Effirul Ramlan (University of Galway). We generated 20,000 GSM8K-derived arithmetic problems with step-by-step s...

O'Grady, Jake, Gürhan, Asena Isik, Chee, Fong Ting et al. · 465 citations
#computer vision Open access Jun 2016

Software Development in Startup Companies: The Greenfield Startup Model

The results are packaged in the Greenfield Startup Model (GSM), which explains the priority of startups to release the product as quickly as possible, and the need to shorten time-to-market, by speeding up the development through low-precision engineering activities.

Carmine Giardino, Nicolò Paternoster, M. Unterkalmsteiner et al. · 178 citations · ⚡14
#computer vision Open access Oct 2016

Software Startups - A Research Agenda

Software startup companies develop innovative, software-intensive products within limited timeframes and with few resources, searching for sustainable and scalable business models.

M. Unterkalmsteiner, P. Abrahamsson, Xiaofeng Wang et al. · 157 citations · ⚡17
#machine learning Review Open access Oct 2016

“Failures” to be celebrated: an analysis of major pivots of software startups

This study conducts a case survey study based on the secondary data of the major pivots happened in 49 software startups, and demonstrates that customer need pivot is the most common among all pivot types.

Sohaib Shahid Bajwa, Xiaofeng Wang, Anh Nguyen-Duc et al. · 127 citations · ⚡15
#computer vision Review Open access May 2015

A survey study on major technical barriers affecting the decision to adopt cloud services

The comparison of adopter and non-adopter sample reveals three potential adoption inhibitor, security, data privacy, and portability, which underlines the importance of the technical and security perspectives for research investigating the adoption of technology.

Nattakarn Phaphoom, Xiaofeng Wang, S. Samuel et al. · 111 citations · ⚡8
#computer vision Open access Feb 2018

Lean Internal Startups for Software Product Innovation in Large Companies: Enablers and Inhibitors

This study investigates how Lean internal startup facilitates software product innovation in large companies and identifies its enablers and inhibitors, and shows the potential of the method-in-action framework to investigate the Lean startup approach in non-startup context.

Henry Edison, Nina M. Smørsgård, Xiaofeng Wang et al. · 78 citations · ⚡6

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.