Skip to content
#small language model Open access

The Multiplier of a Bollinger Band Changes How Often Price Leaves It, Not Which Way Price Goes: The Count Falls Thirtyfold While the Standard 2.0 Is Best in Only One of Eight Cases [F003]

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)
Consumer Market Behavior and Pricing

Abstract

ボリンジャーバンドには「何σにするか」という選び方がある。倍率を動かすと何が動いて何が動かないのかを、ティックデータで測った。 四銘柄(USDJPY・EURUSD・GBPUSD・XAUUSD)の2019年から2024年までの1時間足で、「上のバンドより上なら買い、下のバンドより下なら売り」を信号にし、一期間ぶん先までの値動きとの符号つきの平均(単位 pips)を測った。逆張りはこの信号の符号を反転しただけなので、片方だけを測っている。 倍率を 1.0 から 3.0 へ上げると、件数は中央 0.0335倍(約三十分の一)に減る。八通りすべてでほぼ同じ割合である。そのとき方向の情報の幅は中央 4.3252 pips あり、同じ期間のスプレッド(0.30 から 3.43)を上回る。だが標準の 2.0 が最良だったのは8通りのうち1通りで(五通りから無作為に選べば1.6通り)、最良の倍率が選択期間と評価期間で一致した銘柄は4つのうち0である。 「2σ の外に出るのは 4.55%」という正規分布の値に対し、実測は 13.4769%(2.9620倍)である。ただし足の中の動きまで独立な正規増分で作った合成乱数でも 11.5897%(2.1388倍)出る。中心も幅も同じ20本から推定するため、価格は自分で作った帯から出やすくなる。裾の重さに帰せられるのは 1.1628倍だけであり、しかも倍率 1.0 と 1.5 では実測の方が外に出にくい(0.9426倍・0.9406倍)。裾が効くのは 2.5 以上で、3.0 では 6.6768倍になる。 標準の設定では、日ごとに平均してからの値は8通りすべてで負であり、全体の平均とは8通りのうち5通りで符号が食い違う。一日あたりの件数は中央 3.0 件で、件数の多い日が全体の平均を作っている。 バンドの幅(上へも下へも同じように広がる量)で五分位に分けると、四銘柄すべてで単調にならない。時間帯を重ならない五区分(東京・ロンドン・重複・ニューヨーク後半・夜間)に分けると、日次クラスタの t の絶対値が 2 以上になったセルは 20 のうち 2 で、偶然なら およそ0.9 セルである。その二つはどちらも負で、方向の情報そのものはスプレッドよりはるかに小さい。 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。移動平均も、移動する標準偏差も、ランダムウォークも既知である。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。測ったのは方向の情報であって、手数料・約定・資金管理を含む売買の成績ではない。利確と損切りで切る出口では測っていない(足の中で上下どちらに先に触れたかは高値と安値から決まらないため)。足は1時間足に固定しており、バンドに触れてから戻るのを待つ読み方は扱っていない。測定は日中・少数銘柄に限られ、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。 ----- A Bollinger band comes with a choice: how many standard deviations. This paper measures, on tick data, what moves and what does not when that multiplier is changed. Using one-hour bars for four series (USDJPY, EURUSD, GBPUSD, XAUUSD) from 2019 to 2024, the signal "buy above the upper band, sell below the lower band" is scored by its signed mean against the price move over the next period, in pips. The mean-reverting reading is the same signal with its sign flipped, so only one of the two is measured. Raising the multiplier from 1.0 to 3.0 cuts the count to a median of 0.0335 of itself, about one thirtieth, with nearly the same proportion in all eight cases. The directional information spans a median range of 4.3252 pips over that move, exceeding the spread over the same period (0.30 to 3.43). Yet the standard 2.0 is best in only 1 of 8 cases (choosing at random among five would give 1.6), and the best multiplier agrees between the selection and evaluation periods in 0 of the 4 series. Against the normal-distribution figure of 4.55% outside two standard deviations, the measured figure is 13.4769%, a factor of 2.9620. But synthetic series built with independent normal increments down to the movement inside each bar already give 11.5897%, a factor of 2.1388: because the centre and the width are estimated from the same 20 bars, price leaves a band it built itself more easily. Only a factor of 1.1628 can be attributed to the tails, and at multipliers 1.0 and 1.5 the measured series in fact leaves the band less often (0.9426 and 0.9406). The tails begin to act at 2.5 and reach 6.6768 at 3.0. At the standard setting the mean of the daily means is negative in all 8 cases and disagrees in sign with the overall mean in 5 of the 8. The median number of observations per day is 3.0, so days with many observations make the overall mean. Splitting by the width of the band — a quantity that grows the same way upward and downward — is not monotone in any of the four series. Splitting into five non-overlapping sessions (Tokyo, London, overlap, late New York, night), the daily-clustered t reaches 2 in absolute value in 2 of the 20 cells against about 0.9 expected by chance; both are negative and the directional information itself is far smaller than the spread. Scope of this paper: no new mathematical theorem and no new law is claimed. Moving averages, rolling standard deviations and random walks are all known. No real course, author, or vendor is the subject here, and none is named. What is measured is directional information, not trading performance including commissions, fills and money management. No exit cutting at a take-profit or stop-loss is used, because which side was touched first within a bar is not determined by the bar's high and low. Bars are fixed at one hour, and the reading that waits for price to return after touching the band is not treated. The measurement is intraday and covers few instruments, and does not contradict results documented at longer holding periods. 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

Microsoft Research Blog Oct 12, 2026

Simulating student learning for better AI tutoring

StudentSim is a research system for evaluating and improving AI tutoring tools before they are tested with learners. It models how individuals may engage with tutors, giving feedback to improve tutor performance across subjects like chess, writing, and mathematics. The post Simulating student learning for better AI tutoring appeared first on Microsoft Research.

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