<- all tokdocs
2026.10.25@thequantbuilder

The Noise Band is Swiss Finance Institute paper 24-97, not a survivor @thequantbuilder found by testing 50 papers, and every chart in the video is captioned SCHEMATIC or ILLUSTRATIVE BACKTEST

Finance
Watch on TikTok

View on TikTok ->

The file is a 73.42 second MP4 (container mov,mp4,m4a,3gp,3g2,mj2, format id bytevc1_1080p_213163-1) at 1080x1920, HEVC Main profile with the hvc1 tag in yuv420p, 50 fps, 3,671 video frames at 140,590 bps, with HE-AACv2 audio at 44,100 Hz stereo and 64,163 bps across 1,585 audio frames, 213,166 bps overall and 1,956,332 bytes on disk; ffprobe and metadata.json agree except that the metadata rounds duration to 73 and prints 1:13. It was posted at epoch 1792949467, which is 2026-10-25T17:31:07Z, by uploader thequantbuilder (uploader id 7673208874096722967) under the channel name quantlabs, credited to "Original Sound" by "TikTok Advertiser", with a 327 character description that ends in the hashtags #trading #daytrading #backtesting #tradingstrategy #algotrading. At capture time (metadata epoch 1792991011, 2026-10-26T05:03:31Z) it showed 74,800 views, 5,109 likes, 111 comments, 483 reposts and 3,917 saves. I read 19 frames from the supplied 2 second extraction (001, 002, 003, 005, 008, 011, 014, 017, 020, 022, 024, 025, 026, 028, 029, 031, 033, 036, 037), re-extracted the first nine seconds at 2 fps because the scrolling strategy list skipped entries between samples, read six of those (d_003, d_004, d_005, d_007, d_009, d_011), and pulled four single frames at exact seconds 2, 3, 4 and 5 to pin the list timing; the transcript is 186 words in 12 segments. There is no presenter, no voice-over face and no screen recording anywhere: the whole video is motion graphics on near-black. It opens on a vertical list under a giant numeral that counts "50" then "49", scrolling past "01 OPENING RANGE BREAKOUT", "02 OVERNIGHT GAP FADE", "03 TIME-SERIES MOMENTUM", "04 RSI(2) REVERSION", "05 TURN OF THE MONTH", "06 VWAP REVERSION", "07 PAIRS · COINTEGRATION", "08 52-WEEK HIGH", "09 VOLATILITY BREAKOUT", "10 DUAL MOMENTUM", "11 BOLLINGER FADE", "12 LOW-VOL ANOMALY", "13 FIRST-HOUR REVERSAL", "14 PRE-FOMC DRIFT", "15 OVERNIGHT RETURNS", "16 LAST-HALF-HOUR DRIFT", "17 MOVING-AVERAGE CROSS", "18 DONCHIAN CHANNEL", "19 KELTNER BREAKOUT", "20 VIX TERM STRUCTURE", "21 SELL IN MAY", "22 EARNINGS DRIFT", "23 SHORT-TERM REVERSAL", "24 CROSS-SECTIONAL MOMENTUM", "25 FX CARRY", "26 VOLUME SPIKE FADE", "27 THE NOISE BAND", "28 GAP AND GO", "29 INSIDE-BAR BREAKOUT", "30 ORDER-FLOW IMBALANCE", "31 200-DAY TREND", and on the first frame only, "48 CLOSE-TO-OPEN REVERSAL", "49 SECTOR ROTATION" and "50 Z-SCORE REVERSION"; at 4 s every visible row has a red strikethrough except 27, and at about 5 s row 27 is ringed in gold under the caption "This one didn't." A cream card then reads "QuantLab / Strategy 27 of 50 / The Noise Band / 01 Source code 02 Python version 03 Full analysis / Free in our channel." The middle third is an animated candlestick chart with axis labels "9:30 NEW YORK OPEN", "11:00", "12:00" through "15:00", a "PURE NOISE" title card, a widening gold cone labelled on entry with "BUY", "SL · TRAILING", "TP" and "ENTRY", and a three-line callout "FUNDS / OPTIONS DEALERS / LEVERAGED ETFs". The validation sequence runs four cards, and all four carry a small grey word in the top left: "SCHEMATIC / TESTED AGAINST RANDOM ENTRIES", "SCHEMATIC / SIX BLIND WALK-FORWARD TESTS" with TRAIN and TEST blocks, "SCHEMATIC / TICK-BY-TICK NASDAQ + S&P" with two squiggles labelled NASDAQ and S&P 500, and then "ILLUSTRATIVE BACKTEST / $10,000 → $150,000" over an equity curve with an x-axis running 2018 to 2026, followed by "ILLUSTRATIVE BACKTEST / $10,000 → $150,000 / ~50% LOSING TRADES / 30% DRAWDOWN". It ends on "SOURCE CODE + PYTHON VERSION + FULL ANALYSIS / FREE / LINK IN BIO". No MetaTrader window, no strategy tester report, no terminal, no code, no repository URL and no channel URL appears in any frame I read. The spoken argument is that the narrator tested 50 strategies from research papers, 49 failed, the one survivor draws a band around post-open chop and buys the breakout into the close, and he then tried to kill it with random-entry tests, six blind walk-forward windows and tick-level MetaTrader runs on Nasdaq and S&P since 2018, turning 10,000 into 150,000. The last transcript segment ends at 73.06 s against 73.42 s of video and 73.48 s of audio.mp3, so nothing is timestamped past the end of the audio and no part of this transcript is a Whisper hallucination.

The Noise Band is a published paper the video is retelling, not a survivor it discovered

At 0.00 s the video says "I tested 50 trading strategies from research papers. 49 failed. This one didn't. It's called the noise band, and the full code is free in our channel." The framing is that the narrator ran a screen and the Noise Band fell out of it as the lone winner. The chronology runs the other way. "Noise area" is the term of art from Swiss Finance Institute Research Paper N°24-97, "Beat the Market: An Effective Intraday Momentum Strategy for S&P500 ETF (SPY)", by A. Barbon, A. Aziz and C. Zarattini, dated 27 November 2024. The SFI abstract states that the strategy "uses a well-defined 'Noise Area' to filter out unproductive trades and identify intraday zones of abnormal demand and supply imbalances" and reports that "From 2007 to early 2024, the resulting intraday momentum portfolio achieved a total return of 1,985% (net of costs), an annualized return of 19.6%, and a Sharpe Ratio of 1.33."

Every mechanical detail in the video matches that paper rather than any generic breakout. The frame at about 26 s draws a widening gold cone anchored at the open, which is the paper's volatility-scaled noise band. The frame at about 38 s labels the exit "SL · TRAILING", and the published rule uses a VWAP-based trailing stop. At 43.32 s the narration says "you're flat before the day ends", and the paper flattens at the close with no overnight holding. A published review of the paper describes the construction as a volatility-based noise cone built from historical intraday returns that captures intraday volatility seasonality rather than a naive square-root-of-time scaling, with entries triggered when price breaks beyond the noise boundaries multiplied by a factor, a stop if price moves against the trade after entry, a VWAP trailing stop, and positions flattened at the end of the day.

So the correct description is that the Noise Band is item 27 in a list of strategies taken from research papers, exactly as the video's own card says ("Strategy 27 of 50"), and the paper that defines it already published the headline performance claim two years before this video. The video never names Barbon, Aziz or Zarattini, never cites SFI 24-97, and presents the result as the narrator's own kill-test survivor.

All four validation cards carry an on-screen SCHEMATIC or ILLUSTRATIVE BACKTEST label

This is the single most important thing in the video and it is never spoken aloud. At 43.68 s the narration says "So, I tried to kill it. Against random entries, it won." The matching card (frame_024, about 46 s) prints "SCHEMATIC" in small grey capitals above "TESTED AGAINST RANDOM ENTRIES", over a gold line labelled "THE NOISE BAND" rising out of a fan of grey random walks. At 48 s the walk-forward card (frame_025) again prints "SCHEMATIC" above "SIX BLIND WALK-FORWARD TESTS", and at 50 s (frame_026) "SCHEMATIC / STILL POSITIVE OVERALL" sits above an equity line labelled "BLIND ONLY". At 54 s (frame_028) the card reads "SCHEMATIC / TICK-BY-TICK NASDAQ + S&P" over two unlabelled squiggles with no axis, no scale and no units.

Then the money shot changes word but not substance. At 60 s (frame_031) and 64 s (frame_033) the header reads "ILLUSTRATIVE BACKTEST" above "$10,000 → $150,000", "~50% LOSING TRADES" and "30% DRAWDOWN". "Illustrative" and "schematic" are the producer's own admission that these curves are drawn, not plotted from a run. Nothing in the video shows output from a backtest: no MetaTrader strategy tester report, no equity curve with a y-axis, no trade count, no Sharpe ratio, no profit factor, no parameter set, and no date of the run.

The video is therefore making quantitative claims in the voice-over while labelling the supporting visuals as illustrations. A viewer who hears "every single tick in MetaTrader" at 51.5 s and sees a chart two seconds later is being shown a drawing of what such a test might look like.

$10,000 to $150,000 implies roughly 40 percent a year; the paper behind the strategy reports 19.6 percent

At 56.0 s the narration says "Nasdaq and S&P since 2018. 10,000." and at 58.78 s it completes with "150,000." The x-axis on the matching card (frame_029, about 56 s) runs 2018, 2019, 2020, 2021, 2022, 2023, 2024, 2025, 2026. Fifteen times money over the eight years from the start of 2018 to the end of 2025 is a 40.3 percent compound annual rate. Over the full 8.75 years from January 2018 to the video's post date of 25 October 2026 it is 36.3 percent a year. Either figure is roughly double the 19.6 percent annualized return that SFI 24-97 reports for the same strategy family over 2007 to early 2024.

The closest published test on the instruments the video names is tighter still. "Noise-band Intraday Momentum in Nasdaq-100 Futures" by Justin Chuk, dated 1 October 2026, adapts the Zarattini noise-area rule to Nasdaq-100 futures at a fixed size of one micro contract. Its abstract reports that on one-minute data from 2020 to July 2026 the executable rule earns $10.26 per trading day net of costs, with a day-block bootstrap 95 percent interval of $3.27 to $17.25, a daily Sharpe ratio of 1.14, and positive results in each of seven calendar years, two of them only marginally. At roughly 252 trading days a year that is about $2,586 a year on one micro contract. The same abstract states that this rule is "one of two survivors among 144 preregistered tests on the same archive", that all intervals are in-sample, and that no live contract has been traded. I could not fetch the SSRN page for this paper directly, which returned HTTP 403, and the academia.edu mirror also returned 403, so I am reporting the abstract as surfaced in search results rather than from the primary page.

The video gives no account rules at all: no position size, no leverage, no contract, no starting margin, no commission or slippage assumption. A 15x result and a $10.26-per-day result are not contradictory if the first compounds aggressive leverage, but the video never says so, and the on-screen label says the curve is illustrative.

"Half the trades lose" understates the published win rate, and the out-of-sample record runs the other way

At 63.1 s the narration says "Half the trades lose." The on-screen card at 64 s softens it to "~50% LOSING TRADES". A published review of SFI 24-97 puts the strategy's win rate at approximately 43 percent, which means roughly 57 percent of trades lose, and notes that the low win rate combined with a decent Sharpe implies a highly convex payoff. The video's number is directionally right and numerically generous.

The larger problem is at 66.0 s, where the narration says "But the edge kept coming back." That claim has a direct published test and it does not support the video. "Out-of-Sample Evaluation of an Intraday Momentum Strategy for the S&P 500 ETF (SPY)" by Sheimy Paz Serpa, dated June 2026 and posted to SSRN on 19 August 2026, replicated the Zarattini rule on IQFeed one-minute data and then ran it forward. Larry Swedroe's 22 October 2026 review of that paper confirms the in-sample replication reproduced the original almost exactly, at a 19.8 percent annualized return against the original 19.6 percent and a Sharpe ratio of 1.34 against 1.33, and states that the out-of-sample period of May 2024 through March 2026 "told a very different story than the in-sample ones." The specific out-of-sample numbers reported in the SSRN abstract are a 9.4 percent total return against 29.5 percent for passive buy-and-hold and a Sharpe ratio of 0.39. The Swedroe post paywalls those figures and the SSRN page returned HTTP 403 to a direct fetch, so I treat the 9.4 / 29.5 / 0.39 trio as unverified at the primary source while the replication figures and the out-of-sample window are confirmed.

This is the one place where the video's chronology is reversed in a way that matters. The strategy was published in late 2024, an independent replication reproduced its in-sample result, and the two years after publication underperformed buy-and-hold. "The edge kept coming back" describes the in-sample picture and not the post-publication one.

"49 of 50 failed" has no published counterpart, and the list never shows 16 of the 50 names

A 49-of-50 failure rate is 98 percent. No replication study I could verify reports anything that severe. The NBER working paper "Replicating Anomalies" by Kewei Hou, Chen Xue and Lu Zhang (Working Paper 23394, issued May 2017) examined 447 anomaly variables and found that "286 anomalies (64%) including 95 out of 102 liquidity variables (93%) are insignificant at the conventional 5% level", and that "imposing the cutoff t-value of three raises the number of insignificance to 380 (85%)." The version published in the Review of Financial Studies in 2020 uses 452 anomalies and reports 65 percent failing the t-value 1.96 hurdle and 82.1 percent failing a 2.78 multiple-testing hurdle. The other standard reference, "Does Academic Research Destroy Stock Return Predictability?" by R. David McLean and Jeffrey Pontiff, Journal of Finance volume 71, issue 1, pages 5 to 32, 2016, examined 97 return predictors identified in 79 papers and won the American Finance Association's 2016 Amundi Smith Breeden Prize; its headline finding is that portfolio returns are 26 percent lower out-of-sample and 58 percent lower post-publication, figures I could not confirm on the Wiley or SSRN pages because both returned HTTP 403.

So the published literature puts anomaly failure somewhere between 64 and 85 percent depending on the hurdle, and post-publication decay at about 58 percent of the original return rather than a binary pass or fail. The video's 98 percent is higher than any of these, and the video supplies no methodology for how a strategy was scored as failed: no sample, no cost assumption, no significance threshold, no hold-out rule.

The list itself is also incomplete. The scroll shows numbers 01 through 31 and, on the opening frame only, 48, 49 and 50. Numbers 32 through 47 never appear on screen at all, which is 16 of the 50 names the video says it tested. Two of the named entries have specific recent evidence against them that the video does not mention. "01 OPENING RANGE BREAKOUT" is the subject of "Opening-Range Breakout Does Not Survive Trading Costs: A Pre-Registered 225-Cell Study on Sixteen Years of Futures Data" by Mulham Fetna, posted to SSRN on 7 September 2026, whose abstract reports that zero of 225 pre-registered cells across nine liquid U.S. futures markets met the positive bar, that the grid earned $1.57 million gross but lost $6.49 million at $25 per round turn, and that the five-minute range favoured by the literature is the worst window. The arXiv preprint "Structural Limits of OHLCV-Based Intraday Momentum Signals in MNQ Futures: A Systematic Falsification Study" by Mathias Mesfin (arXiv 2605.04004, submitted 5 May 2026) tested fourteen signal families on 947 trading days of Micro E-Mini Nasdaq 100 futures from 2021 to 2025 and found none satisfied all five success criteria, mostly because gross returns fell below the friction threshold, while two positive control signals did show an edge. I could not fetch the Fetna SSRN page directly, which returned HTTP 403.

Tick-level MetaTrader testing back to 2018 is possible, but it depends on a broker's tick archive

At 51.5 s the narration says "Then, every single tick in MetaTrader." The capability is real. MetaTrader 5's strategy tester has an "Every tick based on real ticks" modelling mode, and the MetaTrader 5 help states that in this mode testing is performed using real ticks of financial instruments accumulated by brokers, that no simulation is performed, and that downloading the tick data from a broker's server on the first run can take a long time. The same documentation adds the caveat that matters here: if real ticks are absent for some time intervals, the tester simulates the price just as in the ordinary "Every tick" mode.

That caveat is the whole question for a claim that runs back to 2018 on Nasdaq and S&P. Real tick history in MetaTrader comes from the broker, not from the exchange, and a retail CFD or futures-tracking symbol frequently has no stored ticks going back eight years, in which case the tester silently falls back to synthetic ticks. A credible version of this claim would name the broker, the symbol, the tick count downloaded and the modelling quality figure the tester prints. The video names none of those, and the card that accompanies the claim is the one labelled "SCHEMATIC".

Key Takeaways

  • Correction: The video says at 0.00 s that it tested 50 strategies and "this one didn't" fail, presenting the Noise Band as its own discovery. The noise area rule is defined and benchmarked in Swiss Finance Institute Research Paper N°24-97 by A. Barbon, A. Aziz and C. Zarattini, dated 27 November 2024, which the video never cites.
  • Correction: The spoken "10,000 turned into 150,000" since 2018 implies 36 to 40 percent a year. SFI 24-97 reports 19.6 percent annualized and a Sharpe of 1.33 for the same strategy family over 2007 to early 2024.
  • Correction: "But the edge kept coming back" at 66.0 s is contradicted by the post-publication record. Sheimy Paz Serpa's out-of-sample evaluation reproduced the in-sample result at 19.8 percent and Sharpe 1.34, then found the May 2024 to March 2026 window told a very different story, per Larry Swedroe's 22 October 2026 review.
  • Correction: Every chart in the validation sequence is self-labelled. Frames at about 46, 48, 50 and 54 s print "SCHEMATIC" and frames at about 56, 60 and 64 s print "ILLUSTRATIVE BACKTEST". No actual backtest output appears in any frame I read.
  • Partial correction: "Half the trades lose" at 63.1 s is generous. A published review of SFI 24-97 puts the win rate at about 43 percent, so closer to 57 percent of trades lose.
  • Clarification: "Every single tick in MetaTrader" is a real MetaTrader 5 mode, but the MetaTrader 5 help states the tester falls back to simulated ticks wherever a broker's real tick history is missing, which is common for 2018 data.
  • Context: The video's 98 percent failure rate is harsher than the replication literature. Hou, Xue and Zhang found 64 percent of 447 anomalies insignificant at the 5 percent level and 85 percent at a t-value of three; McLean and Pontiff found returns 58 percent lower post-publication across 97 predictors in 79 papers.
  • Context: Opening Range Breakout is item 01 on the video's own list, and Mulham Fetna's pre-registered 225-cell futures study reports zero cells surviving $25 per round turn, with the grid losing $6.49 million net against $1.57 million gross.
  • Unverified: The 30 percent maximum drawdown, the specific 10,000 to 150,000 figures, the six walk-forward windows and the random-entry comparison have no source outside the video, and the video labels the charts that display them as illustrative.
  • Unverified: The exact out-of-sample figures of 9.4 percent versus 29.5 percent and Sharpe 0.39 come from the SSRN abstract as surfaced in search; the SSRN page returned HTTP 403 and the Swedroe post paywalls them.
  • Unverified: The "free in our channel" source code, Python version and full analysis could not be checked. The TikTok profile page returned no bio or link to a direct fetch, and no repository or channel URL appears in any frame.
  • Context: The on-screen list shows numbers 01 to 31 plus 48, 49 and 50. Numbers 32 to 47 never appear, so 16 of the 50 claimed strategies are never named.

Resources

Published October 25, 2026. Writeup generated from a favorited TikTok.