Skip to content
#small language model Open access

The "True Range" in ATR Is, Most of the Time, Just the Range — At the Median, ATR ÷ the Mean of the Range Alone Is Exactly 1.0000 and the Gap Binds on Only 0.0370 to 0.0897 of Bars, Yet at the Weekly Open It Binds 11.3968 Times as Often [F047]

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

Abstract

ATR は「平均的な値幅」と説明される。だが中身の真の値幅は TR = max(高値 − 安値, |高値 − 前の終値|, |安値 − 前の終値|) であって、前の終値を含んでいる。つまり足の中の動きだけでなく、足と足のあいだに起きたことも測っている。この二つは別の量である。片方は市場が開いているあいだの動きで、もう片方は市場が止まっていたあいだに起きたことだからである。では、前の終値を入れることは実際にどれだけ効いているのか。本稿の分離子は、前の終値を含めたかである。TR は max なので、飛びが効いていない足では TR と値幅がぴったり等しくなり、効いたかどうかが数えられる。 四銘柄(USDJPY・EURUSD・GBPUSD・XAUUSD)の2019年から2024年までの四つの足種(5分・15分・1時間・4時間)で、TR と値幅を両方作った。一時間足は 36,534・36,554・36,535・34,892 本で、一本に入っているティックは中央で 3,671・3,369・3,740・6,743 である。期間は ATR の慣用どおり 14 本、週明けは直前の足との時刻の空きが 24 時間以上の足とし、層別は五分位、先の長さは 24 本とした。基準線は足の中まで歩かせた合成乱数である。 飛びが効くのは足のごく一部だった。五分足で 0.0897、十五分足で 0.0575、一時間足で 0.0446、四時間足で 0.0370 である。しかも効いたときでも TR ÷ 値幅 は 1.0140 から 1.0761 しかない。だから 14 本で平均すると消えてしまい、ATR ÷ 値幅だけの平均 の中央は五分足で 1.0017、一時間足と四時間足ではちょうど 1.0000 になる。つまり ATR は、ふだんただの値幅である。名前に「真の」と付いている部分は、ほとんどの足で何も足していない。予言③(ATR ÷ 値幅 は 1 より大きく足が短いほど大きい)は外れた。 ところが週明けの足だけは別物になる。飛びが効く割合が 0.4841・0.5000・0.4586・0.3119 で、それ以外の足の 11.3968 倍である。EURUSD では二本に一本にあたる。時刻では UTC 21 時の 0.1113 が最大で、UTC 9 時の 0.0236 の 4.7187 倍である。UTC 21 時と 22 時(0.1113 と 0.1007)は日の区切りにあたる。建玉の大きさで言えば、ATR ÷ 値幅だけの平均 の中央は 1.0000 で何も起きないのに、最大は 2.0116 になる。金の ATR が値幅だけの平均の二倍になる 14 本の窓が実在する。中央値では正しく、端では正しくない量である。一方、先の値動きの大きさを分ける力は変わらない。第5層 ÷ 第1層 は TR で 1.5752、値幅だけで 1.5703 で、比は 1.0031 である。 予言④(合成乱数では飛びが効かない)は外れ、しかも逆だった。合成乱数のほうが実測より 5.6573 倍も飛びが効く。理由づけそのものが誤っていた。足の高安はその足の中で歩いた点からしか取らないので、前の終値はそこに入っておらず、一本まるごと同じ向きに動けば飛びが効く。そこで足の中を歩く回数を変えて測り直すと、5 回で 0.4919、20 回で 0.2523、100 回で 0.1117、400 回で 0.0561 となり、実測の 0.0446 はその先にちょうど乗った。つまり飛びが効くのは「市場が閉じていたから」ではなく、その足の中の道が前の終値まで戻らなかったからである。実際の足には中央 3,706 ティック入っているので、ほとんどの足で戻る。金の割合が低いのも、一本に 6,743 ティック入っていて他の三銘柄の 3,369 から 3,740 より多いからと読める。結論として、ATR を「平均的な値幅」と呼ぶのはふだんは正確であり、正確でなくなるのは市場が本当に止まっていた 314 本の足と、分布の端だけである。 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。真の値幅の定義は公表されているものである。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。特定の配信元も論じていない(飛びの大きさは配信元で変わる)。売買を論じておらず、成績も損益も測っていない。予言③と④が外れており、④は理由づけそのものが誤っていた。期間を一つ(14 本)に固定している。薄まったあとしか測っておらず、飛びが効く足だけを取り出して先の値動きを測ってはいない。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。 ----- ATR is described as "the average range". But the true range inside it is TR = max(high − low, |high − previous close|, |low − previous close|), which includes the previous close: it measures not only the movement inside the bar but also what happened between one bar and the next. These are two different quantities — one is movement while the market is open, the other is what happened while the market was standing still. So how much does including the previous close actually do? The separator in this paper is whether the previous close was included. Because TR is a max, on bars where the gap does not bind TR and the range are exactly equal, which makes "did it bind" countable. Using four instruments (USDJPY, EURUSD, GBPUSD, XAUUSD) from 2019 to 2024 across four bar sizes (5 min, 15 min, 1 hour, 4 hour), both TR and the range were built. The hourly series have 36,534, 36,554, 36,535 and 34,892 bars, with a median of 3,671, 3,369, 3,740 and 6,743 ticks inside one bar. The period is ATR's conventional 14 bars, the weekly open is a bar whose gap in time from the previous one is 24 hours or more, layering is by quintile and the forward horizon is 24 bars. The baseline is a synthetic random walk taken down to within the bar. The gap binds on only a small share of bars: 0.0897 on 5-minute, 0.0575 on 15-minute, 0.0446 on hourly and 0.0370 on 4-hour bars. And even when it binds, TR ÷ range is only 1.0140 to 1.0761. So averaging over 14 bars wipes it out: the median of ATR ÷ the mean of the range alone is 1.0017 on 5-minute bars and exactly 1.0000 on hourly and 4-hour bars. ATR is, most of the time, just the range, and the part of the name that says "true" adds nothing on almost every bar. Prediction ③ (that ATR ÷ range exceeds 1, more so on shorter bars) failed. At the weekly open, however, it becomes a different thing. The gap binds on 0.4841, 0.5000, 0.4586 and 0.3119 of those bars — 11.3968 times the rate elsewhere, and for EURUSD one bar in two. By hour, 21:00 UTC peaks at 0.1113, 4.7187 times the 0.0236 of 9:00, and 21:00 and 22:00 (0.1113 and 0.1007) are the daily boundary. In position terms, the median of ATR ÷ the mean of the range alone is 1.0000 — nothing happens — while the maximum reaches 2.0116: there really are 14-bar windows where gold's ATR is twice the mean of its range alone. It is a quantity that is right at the middle and wrong at the edge. The power to separate the forward magnitude, on the other hand, does not change: the fifth ÷ first quintile is 1.5752 by TR and 1.5703 by range alone, a ratio of 1.0031. Prediction ④ (that the gap never binds on a synthetic random walk) failed, and in the opposite direction: the random walk binds 5.6573 times more often than the measurement. The reasoning itself was wrong — a bar's high and low are taken only from the points walked inside that bar, so the previous close is not among them and the gap binds whenever the bar moves one way throughout. Varying the number of steps walked inside the bar gives 0.4919 at 5 steps, 0.2523 at 20, 0.1117 at 100 and 0.0561 at 400, and the measured 0.0446 sits right at the end of that column. So the gap binds not because "the market was closed" but because the path inside that bar did not come back to the previous close; a real bar holds a median of 3,706 ticks, so on almost every bar it does. Gold's lower share reads the same way, with 6,743 ticks per bar against 3,369 to 3,740 for the others. The conclusion: calling ATR "the average range" is accurate most of the time, and stops being accurate only on the 314 bars where the market really had stopped, and in the tail. Scope of this paper: no new mathematical theorem and no new law is claimed; the definition of the true range is published. No real course, author, or vendor is the subject here, and none is named; nor is any particular data feed (the size of a gap depends on the feed). No trading is discussed, and neither performance nor profit and loss is measured. Predictions ③ and ④ failed, and ④ was wrong in its reasoning. One period is fixed (14 bars). Only what is left after the washing out is measured; bars where the gap binds were not taken out on their own and followed forward. 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.