**I Backtested 5,679 Bitcoin Candlestick Trading Strategies. Here is the Data. — Transcript & Summary | SozAI**
Source: https://sozai.app/transcript/bitcoin-candlestick-trading-strategies-backtest-data/

Backtest of 43 candlestick patterns across five timeframes on Bitcoin finds zero survive all filters; only daily shows marginal, inconsistent positives.

## Key Takeaways

- No candlestick pattern on Bitcoin survived full walk-forward validation and fee-adjusted filters at any tested timeframe.
- Short timeframes are unprofitable largely due to fees and slippage relative to typical Bitcoin bar movements.
- Daily timeframe shows the only marginal positive signals, but they are inconsistent and likely reflect regime timing.
- Apparent out-of-sample success without in-sample support should be treated as luck, not robust edge.
- Market regime, asset class, and timeframe are more important than pattern shape.

## What the video covers

- Tests 43 candlestick patterns on Bitcoin across five timeframes using five years of Binance data and walk-forward validation over five sequential windows.
- Out of 5,679 backtest results, only eight passed walk-forward validation and none survived the full filter stack.
- Candlestick patterns are argued to be a cleaner test than indicator strategies because they have few tunable parameters, reducing overfitting risk.
- Short timeframes (5-minute, 15-minute, 1-hour) show 0% positive results, with fees and slippage compounding into large negative Sharpe values.
- Daily timeframe is the only one with meaningful positive results, with 59% of daily results positive out of sample.
- Several daily patterns (stick sandwich, hanging man, tweezer top) show positive out-of-sample Sharpe despite negative in-sample Sharpe, which the author attributes to favorable regime timing rather than edge.
- Single, two, and three candle pattern categories show no reliable average positive performance across timeframes.
- The author concludes that the market regime, asset class, and timeframe matter more than chart shapes for predicting outcomes.
- The video is a continuation of a four-part series of backtests on retail trading strategies, referencing prior results on 31 assets, Bitcoin indicators, and 1,000 stocks.
- The video ends with a membership promotion and a mild off-topic remark about politeness in comments.

## Chapters

1. 00:00 Introduction: Zero Candlestick Patterns Survive
2. 00:25 Recap of Previous Studies
3. 01:28 Why Candlestick Patterns Are a Cleaner Test
4. 02:27 Methodology and Dataset
5. 02:59 Heat Map and Fee Analysis
6. 03:35 Results by Category and Timeframe
7. 04:32 Daily Timeframe Results and Lucky Windows
8. 06:29 Conclusion and Closing Remarks

Answers

## Questions about this video

Did any candlestick pattern survive all filters on Bitcoin?

No. Out of 5,679 backtest results across 43 patterns and five timeframes, eight passed walk-forward validation but zero survived all filters. The presenter concludes candlestick patterns do not work on Bitcoin at any tested timeframe after fees.

Why are the five-minute results so negative?

The presenter attributes the strongly negative Sharpe values to fees compounding across many small trades. Bitcoin typically moves 0.1 to 0.3% on a five-minute bar, while a round trip costs about 0.1% in commission plus slippage, so patterns would need to be right well over half the time just to break even.

Why does the daily timeframe show positive results if the overall conclusion is negative?

Daily results show 59% positive out-of-sample outcomes and seven of the eight robust patterns, but the presenter argues these are not consistent enough to survive full validation. Several show negative in-sample Sharpe with positive out-of-sample Sharpe, which he attributes to favorable out-of-sample windows rather than genuine edge.

## Full Transcript — Download SRT & Markdown

00:00

Speaker A

Zero. Yes, zero. After testing every candlestick pattern ever documented across five time frames on Bitcoin, zero survived every filter.

00:11

Speaker A

That is the result. Now, let me show you why it is actually the most useful thing this series has produced. This is the fourth video where I backtest everything retail traders actually use at scale.

00:25

Speaker A

Quick recap so the numbers make sense. Video one, 31 assets, 62 strategies, about 131,000 backtests. 65 survived.

00:38

Speaker A

Momentum and mean reversion did most of the work. Video two was Bitcoin only. 350,000 combinations across five time frames. Four survived. All of them trend following. Mean reversion on Bitcoin got nothing through. Video three, 1,000 stocks. 2.3 million backtests. 2,226

01:04

Speaker A

survived. Healthcare dominated and volume-based signals averaged basically zero across the entire universe. Here is what makes this video different from the first three though. Every previous study tested indicator-based strategies.

01:19

Speaker A

Moving averages, RSI thresholds, Bollinger Bands, all of these have parameters, window lengths, threshold levels multipliers.

01:28

Speaker A

My optimizer search up to 60 parameter combination per strategy and keeps whatever performed best historically, which creates a real overfitting risk.

01:38

Speaker A

Walk forward testing is designed to catch that, but it is still a legitimate concern. Candlestick patterns barely have this problem. A hammer is a hammer.

01:48

Speaker A

Small body, long lower wick, almost no upper wick. Maybe one or two tunable parameters at most. How strict that wick ratio needs to be, how many bars can count as a downtrend, the search space is tiny, which means this negative

02:06

Speaker A

result is cleaner than any negative result from the first three videos. You cannot blame overfitting here. The pattern either produces edge or it does not. With candlestick patterns, the test is about as pure as a backtest gets. 43 patterns, five time frames,

02:27

Speaker A

five years of Bitcoin data from Binance, walk-forward validation across five sequential windows, same methodology as every previous video. 5,679 backtest results saved, eight passed walk-forward, zero survived all filters.

02:45

Speaker A

Look at the heat map on screen. Every column is a time frame. Every row is a pattern. The color represents out-of-sample Sharpe ratio. You are looking at a wall of red. The five-minute column is not just red, the

02:59

Speaker A

numbers are negative 30, negative 50, negative 90. That is not a strategy failing. That is fees compounding to zero across hundreds of tiny trades. On a five-minute bar, Bitcoin typically moves 0.1 to 0.3%.

03:16

Speaker A

A round trip costs 0.1% in commission plus slippage. The pattern needs to be right significantly more than half the time just to break even on fees. Candlestick patterns are not that precise. Nothing really is. The red gets less extreme as

03:35

Speaker A

you move to longer time frames. 4-hour starts showing values near zero. Daily shows the only positive numbers in the entire table, and even those are not strong enough to survive a strict validation regime. Breaking it down by category and by time frame makes the

03:55

Speaker A

picture even clearer. Single candle patterns, two candle patterns, three candle patterns, none of them average positive once you pull across all five time frames. Two candle patterns are the worst category on average, -20. Those same patterns look completely different

04:12

Speaker A

on daily. The more complex multi-candle patterns and the composite strategies do not even show up in this breakdown.

04:21

Speaker A

Across the entire study, on every time frame, not one of them produced a single usable walk-forward result. By time frame, the cliff is even sharper.

04:32

Speaker A

5-minute, 15-minute, and 1-hour all sit at exactly 0% positive. Not mostly negative zero. 4-hour finally cracks open slightly, 14% positive. Daily is the outlier. 59% of daily results come back positive out of sample. Seven of the eight robust patterns in this entire

04:52

Speaker A

study live on the daily time frame alone. If candlestick patterns have any signal left in them anywhere, it is on daily bars. And even there, it was not enough to survive the full filter stack.

05:04

Speaker A

By the way, if you want to support this work and get access to the full database and pipeline for all four studies, the members section link is below. Back to the results. Stick sandwich, hanging man, and tweezer top all show positive

05:19

Speaker A

out of sample sharp ratios on daily Bitcoin. Each paired with a negative in sample sharp. The out-of-sample windows happen to land on favorable regimes.

05:30

Speaker A

That is not edge. That is timing. A strategy whose own optimized sharp across the full history is negative has no business turning positive just because of one slice of out-of-sample testing. It is a lucky window, not a robust finding. Evening star is a

05:49

Speaker A

slightly different case. Sharp is barely positive at 0.159. Out of sample jumps to 0.92, almost six times higher. Same conclusion though, a small positive edge in sample does not explain a jump that large. The out of sample window was just kind of to

06:09

Speaker A

this pattern than its own history was. The only honest conclusion from the daily results is this. Marginal positive signals exist, but they are not consistent enough to survive a full walk-forward validation. If you lowered every filter to its minimum, no minimum

06:26

Speaker A

trade count, no consistency requirement between in sample and out of sample, you would be able to produce a list of surviving patterns. That list would not mean anything though. The methodology exists to prevent exactly that. So, the honest answer to whether candlestick

06:45

Speaker A

patterns work on Bitcoin is no. [snorts] Not at any time frame tested. Not after accounting for fees. Not after walk-forward validation across five sequential periods of history. The retail trading industry sells the idea that reading charts is a skill. Specific

07:02

Speaker A

patterns predict specific outcomes. The hammer is bullish. The evening star is a reversal signal. 2.3 million backtests on stocks. And now, 43 patterns across five time frames on Bitcoin say something different. The shapes on the chart are not the signal.

07:22

Speaker A

The market regime, the asset class, and the time frame are the signal. Tell me in the comments what you actually want tested next. I read every single one, including the mean ones. And on those, basic politeness goes a long way. Your

07:38

Speaker A

parents probably covered this already. Just a friendly reminder in case it slipped. Otherwise, my next video is going to be a basic tutorial on politeness. Happy to do it as well as there seems to be an inherent demand for

07:55

Speaker A

it. Everything from all four studies is in the member section. The databases, the notebooks, the pipelines. If you have built something similar yourself and want to compare methodology, or if you just want to save the time that went

08:09

Speaker A

into validating all of this, the link is below.

Topics: bitcoin candlestick patterns backtesting walk-forward validation algorithmic trading overfitting technical analysis crypto trading Sharpe ratio trading strategies

Study this video

- [Flashcards from this video](https://sozai.app/tools/flashcard-maker/#from=transcript&id=96573)
- [Quiz on this video](https://sozai.app/tools/quiz-generator/#from=transcript&id=96573)
- [Crossword from this video](https://sozai.app/tools/crossword-maker/#from=transcript&id=96573)
- [Word search from this video](https://sozai.app/tools/word-search-maker/#from=transcript&id=96573)
- [Study guide from this video](https://sozai.app/tools/study-guide-maker/#from=transcript&id=96573)

Free, made by AI from this transcript. No signup.


---
This is the markdown twin of https://sozai.app/transcript/bitcoin-candlestick-trading-strategies-backtest-data/ — the same content, without the markup.
Published by SozAI (https://sozai.app). Reuse and quotation are allowed with attribution and a link back.
Machine-readable index: https://sozai.app/llms.txt · data API: https://sozai.app/api/
