Skip to content
#small language model Open access

Aligning Timeframes Only Reduces the Count by as Much as You Aligned: The Count Falls to 0.3683 While the Directional Information Does Not Rise, and the Apparent Significance Is the Overlap — Removing It Drops the Median Daily t from 2.4643 to 0.5923 [F010]

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

Abstract

上位の足の向きにそろえてから入る、三本の移動平均が順に並ぶのを待つ、という作法がある。条件を足すほど精度が上がると言われる。だが条件を足せば件数は必ず減る。これは市場について何も言っておらず、掛け算の性質である。そして件数が減れば勝率は動きやすくなる。問うべきは、増えるのは情報なのか、減るのは件数だけなのか、である。 四銘柄(USDJPY・EURUSD・GBPUSD・XAUUSD)の2019年から2024年までのティックから五分足・十五分足・一時間足・四時間足を作り、一時間足を基準に測った。上位足は「完成した足」だけを参照している(進行中の四時間足を使うと最大四時間ぶんの先読みになる)。そろえる本数 k を 1 から 4 へ動かすと、件数は四銘柄の比の中央で 0.3683 倍に落ちる。四つが独立なら 0.1250 になるはずなので、実測はその約三倍残っており、四つの足は独立ではなく同じ価格列を四つの粗さで見ている。同じとき、方向の情報(信号の符号と二十四時間先の値動きの積の平均)の中央は +0.2548・+0.4347・+0.8606・−0.2449 pips と動き、単調ではない。三本の移動平均を順に並べる形でも、件数は 0.3413 倍に落ち、方向の情報の増分の中央は −1.3877 pips で四銘柄すべて負である。 本稿の芯は重複にある。k = 4 では日次クラスタの t の絶対値の中央が 2.4643 になり、有意なセルが 4 のうち 4 になる。偶然ならおよそ 0.2 なので、ここだけを見れば「そろえると効く」と読める。だが二十四本先まで見ている以上、隣り合う観測は二十三本ぶん重なっている。日次クラスタは同じ日の中の重なりしか畳まず、日をまたぐ重なりは残る。前の観測が終わるまで次を立てない形にすると、件数は 0.1012 倍になり、日次 t の中央は 0.5923 に、有意なセルは 4 のうち 0 になる。パーフェクトオーダーでも 5.2377 から 0.5286 に落ちる。さらに、そろえた形では符号が 4 のうち 2 の銘柄で反転する。重複を除くことは精度を上げる操作ではなく、別の量に変える操作である。 選択期間では四銘柄すべてで k = 4 が最良だったが、評価期間で最良だったのは 4 のうち 3 の銘柄で k = 3 であり、一致は 4 のうち 1(偶然なら 1.0)だった。足の中まで独立な正規増分で一本の道を作り、それを四つの足にまとめ直すと、k = 4 の件数比は 0.3895 から 0.3918 になる。実測の 0.3630 から 0.3810 より、むしろ高い。そろうことは市場の性質ではなく、同じ価格列を四つの粗さで見ていることの帰結である。なお、この基準線では初稿で誤りを踏んだ。四つの足それぞれに別々の乱数を当てており、その版の件数比は独立の値 0.1250 に対して 0.1243 から 0.1257 とほとんど一致した。基準線が理論値と一致しすぎるときは、基準線の作り方を疑うべきである。 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。移動平均、多時間軸の一致、ランダムウォーク、重複した観測が t を膨らませること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。売買の成績(手数料・約定・資金管理を含む)は測っていない。足の組み合わせを一つに固定しており、別の組み合わせでは数字が変わる。「そろえる」を上下の一致と定義しており、傾きの一致や並びの一致は扱っていない。重複を除いた件数は 584 件から 665 件なので、「有意でない」は「無い」ではなく「この件数では見えない」である。時間帯の層別は重複を残したままであり、そこで有意に見えたセルは市場の性質として読んではいけない。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。 ----- There is a practice of aligning the direction of a higher timeframe before entering, and of waiting for three moving averages to line up, with the claim that adding conditions raises accuracy. But adding a condition necessarily reduces the count, which says nothing about the market and is a property of multiplication; and a smaller count makes the win rate move more easily. The question is whether information rises, or only the count falls. From ticks for four series (USDJPY, EURUSD, GBPUSD, XAUUSD) from 2019 to 2024, five-minute, fifteen-minute, one-hour and four-hour bars were built and measured on a one-hour base. Higher timeframes are read only from completed bars (using a four-hour bar still in progress would look up to four hours ahead). Moving the number aligned k from 1 to 4 cuts the count to a median ratio of 0.3683. Were the four independent it would be 0.1250, so about three times as much survives: the four timeframes are not independent but one price series seen at four resolutions. Over the same move, the median directional information (the signal's sign times the price move over the next twenty-four hours) runs +0.2548, +0.4347, +0.8606, −0.2449 pips, which is not monotone. Requiring three moving averages to line up cuts the count to 0.3413, and the median change in directional information is −1.3877 pips, negative in all four series. The core of the paper is the overlap. At k = 4 the median absolute daily-clustered t reaches 2.4643 and the significant cells become 4 of 4, against about 0.2 by chance, which read alone says "alignment works". But with twenty-four bars of look-forward, neighbouring observations overlap by twenty-three bars, and daily clustering folds only the overlap within a day, leaving the overlap across days. Opening the next observation only after the previous one has finished leaves 0.1012 of the count, drops the median daily t to 0.5923, and takes the significant cells to 0 of 4. For the perfect order the median falls from 5.2377 to 0.5286. Moreover, for the aligned family the sign reverses in 2 of the 4 series. Removing the overlap is not an operation that improves precision; it is one that changes the quantity. In the selection period k = 4 was best in all four series, but in the evaluation period the best was k = 3 in 3 of the 4, a match of 1 of 4 against 1.0 by chance. Building one synthetic path from independent normal increments down to the movement inside each bar and aggregating it into the four timeframes gives a k = 4 count ratio of 0.3895 to 0.3918 — higher than the measured 0.3630 to 0.3810. Alignment is not a property of the market but a consequence of looking at one price series at four resolutions. One error was made and corrected in this baseline: the first version drew a separate random series for each timeframe, and that version gave count ratios of 0.1243 to 0.1257 against the independent value of 0.1250. When a baseline agrees too well with the theoretical value, suspect the baseline. Scope of this paper: no new mathematical theorem and no new law is claimed. Moving averages, multi-timeframe alignment, random walks, and the fact that overlapping observations inflate t are all known. No real course, author, or vendor is the subject here, and none is named. Trading performance, including commissions, fills and money management, is not measured. One combination of timeframes is fixed, and other combinations give other figures. "Aligned" is defined as agreement in sign; agreement in slope or in ordering is not treated. After removing the overlap the counts are 584 to 665, so "not significant" means "not visible at this count", not "not there". The session split still carries the overlap, and cells that look significant there must not be read as a property of the market. 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.