← hub

Trading strategy research: YouTube, web and our own tests

Prepared 25 Sep 2026. Question: which trading methods actually make money, and when? Scope: scalping, day trading, swing/trend, ICT/SMC, session-based, bots. YouTube limited to uploads from the last 18 months (since 25 Mar 2025). Not financial advice.

How the research was done

• 72 YouTube searches across 36 topics (scalping, day trading, swing, trend following, ICT/SMC, liquidity sweeps, New York / Asian / London sessions, opening-range breakouts, VWAP, mean reversion, order flow, liquidation heatmaps, grid / DCA / market-making / AI bots, funding arbitrage, reality checks). Half used YouTube's "this year" filter.
• 903 unique videos found, 527 uploaded on or after 25 Mar 2025. Each was checked for exact upload date, views, likes and comment counts, then classified by method.
• 58 of the most-viewed videos were read in detail: description plus 3,358 top comments. Six transcripts were read in full, including three independent large-scale backtests.
• Web research: academic papers and industry write-ups on session effects, trend following, retail day-trader results, funding arbitrage and grid bots.
• Our own tests on Binance data (fees included, train/test halves) of every claim that could be turned into rules.

What is being taught (527 recent videos)

Most-viewed recent examples: The BEST 5 Minute Scalping Strategy Ever (1.8M views), The "ONE CANDLE" Scalping Strategy (1.8M), My Simple 9:30AM Candle Scalping Strategy (432K), Copy This 5 Rule SMC Strategy (482K), The 3 Step A+ Supply & Demand Strategy (548K), TTrades: The Only Trading Strategy You Need For 2026 (366K).

The independent tests (the videos that actually tested things properly)

1. Claude Tested Over 9,000 Trading Strategies (Jun 2026). 15 years, 30 assets (stocks, ETFs, gold, oil, BTC, ETH), daily bars, walk-forward validation.

• Of 9,000 strategies, 524 survived all six filters. Only 44% of strategies that looked good in testing stayed good on unseen data: more than half of good-looking backtests were fake.
• Mean reversion was the only family positive on average (RSI snap-backs, Keltner / Bollinger reversion): 64% of the survivors. Trend, volume, volatility and pattern strategies were negative on average in their naked form. Cross-sectional momentum (rank a basket, hold the strongest) scored much better than single-asset momentum.

2. Claude Tested Over 25,000 ICT Strategies (Aug 2026). 16 years of 1-minute NQ / ES futures, 10 million simulated trades, real commissions and slippage, 2 years locked away.

• The famous 70-80% win rate is not there: top versions won 34-48%, the survivors about 52%.
• Only 1.7% of the 25,000 versions survived the full test. The "Silver Bullet" time windows are not special: they sat in the middle of every other hour.
• Only one rule added an edge: requiring the liquidity sweep. Adding the higher-timeframe bias filter that guides teach made results worse. The surviving version is simply a sweep-and-snapback (mean reversion), traded often.
• Caveats from the author: perfect fills assumed (real life misses the winners), the two hold-out years were a strong rising market, and a trader would need to skip 17% of losing trades in advance for the textbook version to work.

3. The Dark Side of the 9:30AM Candle Scalping Strategy (Jan 2026, the strategy's own creator). Buying every breakout: 1,023 trades, 31% won, profit factor 0.6 (loses money). His "refined" version (wait for a retest) claims 68% won and a profit factor of 4.15 over 1,138 trades. That is a hand-marked backtest with hindsight, from a channel that sells courses: treat it as unproven.

4. I Backtested 1,000 Trading Strategies (Feb 2026). Rules of thumb: a 98% win rate with tiny drawdown "is not an edge, it's not durable"; you need enough trades, realistic slippage and fees, and a reason the trade works.

What viewers say (3,358 comments read)

• Most top comments are thank-yous. A large share of the "I turned $5k into $31k" comments are copy-paste bot testimonials (identical stories appear under unrelated videos).
• The genuine first-hand comments lean negative: "11 months trying every YouTube strategy and VIP group, my account is still at zero" (repeated on several videos), "I tried this 4 times this week, all 4 trades were losers", "ORB didn't work for me on NQ due to heavy wicks and low volume", "every time I backtest a strategy and it works, the market changes the rules the next day".
• The few positive first-hand reports are tiny samples: "13 wins against 12 losses on gold over a month" is a coin flip; another tested "on demo" only.
• A commenter on the 25,000-ICT video reported that when he replicated it, Claude found a look-ahead leak in the selection step: even the "rigorous" AI-run tests can flatter themselves.
• Several comments make the same point: 1-minute charts and scalping were where they lost the most; higher timeframes and patience worked better.

Best time to trade (sessions)

• Volatility and volume peak in the US / Europe overlap: 13:00-18:00 UTC (16:00-21:00 Nairobi time) for BTC. The quietest hours are 04:00-05:00 UTC. Academic studies agree: Bitcoin volume and volatility are higher during European and US daytime hours; options activity clusters at 08:00-09:00 and 14:00-15:00 GMT.
• Direction is another matter. On our own 3.4 years of hourly data (BTC plus a basket of 10 coins) the average return in any hour is a few basis points (0.03-0.05%), far below the ~0.14% cost of a trade. The most consistent: the 13:00 UTC hour (US open) is slightly negative for the coin basket (-5 bp, both halves); 21:00-22:00 UTC slightly positive (+3 to +5 bp). None is tradeable on its own.
• The Monday Asia-open effect described in a 2025 study is present but weak in our data (+12 bp per week, not statistically significant).
• The 25,000-strategy ICT test found the "special" ICT hours are no better than any other hour.
• Practical rule: trade when liquidity is high (US / Europe hours) for tight spreads and fills, but the session does not create an edge by itself.

What the evidence says works and what doesn't

Recommendation

1. Core: the daily trend system (Turtle-style breakouts, optionally with the BTC uptrend filter) plus the weekly momentum rotation as a low-drawdown second sleeve. It is slow (a few trades a month), wins about 40-46% of the time, and is the only approach that shows up in independent evidence, in our tests, and in both halves of the data.
2. Tactical: the STANDARD 4h buy (ICT sweep at support, daily trend up) as a small satellite. Only the sweep concept survived the large ICT test, and it needs the trend filter in crypto.
3. Skip: 1m / 5m scalping, opening-range and "first candle" strategies, mean reversion on perps, discretionary news trading, grid bots without a trend filter.
4. How to test the waters: paper trade first, then smallest real size, with a stop on every trade and a kill switch (stop at -20% or 8 losses in a row). Win rates matter less than average R and drawdown.
5. Be realistic about capital: trend following returned about 20% a year in the backtest (before survivorship effects). On $20 that is about $4 a year. Growth speed is set by capital, and no tested strategy reliably "flips" a very small account.

Sources

• The intraday dynamics and intraday price discovery of bitcoin (ScienceDirect)
• Seasonality in Bitcoin Intraday Trend Trading (Concretum Group)
• Systematic Trend-Following with Adaptive Portfolio Construction in Crypto (arXiv, 2026)
• The Profitability of Day Traders (Financial Analysts Journal)
• Funding rate arbitrage strategy guide
• I backtested the most popular Binance grid bots on 4 years of real data
• YouTube: Claude Tested Over 9,000 Trading Strategies, Claude Tested Over 25,000 ICT Strategies, The Dark Side of the 9:30AM Candle, I Backtested 1,000 Trading Strategies
• Our tests: crypto_scanner\tournament.py, scalp_lab.py, session_returns.py (in trading_research), wade_lab.py, sr_backtest.py; results summarised in the analysis standard.