I Tested the Strategy That Turned $5,000 Into $15 Million

ELI5/TLDR

A YouTuber rebuilt Ed Seykota’s old trend-following system in code and backtested it on six markets over five years. The rules are dead simple: trade with the 100-day moving average, enter on 20-day breakouts, add to winners up to three times, and trail your stop with a volatility measure. Across a diversified basket it returned 198% over six years with a 33% win rate — a losing percentage of trades, but the few big winners more than paid for the many small losses. The video is also a sponsored ad for a data API, which is worth keeping in mind.

The Full Story

The legend, and the asterisk on it

The hook is Ed Seykota turning a client’s $5,000 into over $15 million — a 300,000% return. That number does the heavy lifting in the title and never gets examined. Seykota was a real and influential trend follower, and the rest of the video quietly walks away from the $15M and tests something far more modest. Worth noting up front: Seykota wrote no book, so the whole strategy is reverse-engineered from descriptions of his systems. The “five rules” are the presenter’s reconstruction, not gospel.

The five rules

The system is genuinely simple, which is the point.

His first rule was a trend filter. This is primarily an EMA 100. Above that EMA, we’re going long… and below that EMA, we have a short bias.

The rest:

Trend filter — 100-period exponential moving average sets the bias (long above, short below).
Entry — a new 20-day high triggers a long; a new 20-day low triggers a short. This is a classic Donchian breakout.
Pyramiding — add up to three more units as price keeps moving in your favour, each new unit triggered once price has travelled half an ATR (Average True Range, a volatility measure) from the last entry.
Trailing stop — an ATR-based stop that lets winners run and cuts losers fast.
Position sizing — fixed risk percentage per trade, scaled down when volatility (ATR) is high.

None of this is novel; it’s the skeleton every trend follower from Donchian to the Turtles has used.

Why a strategy that loses most of the time makes money

This is the genuinely useful part. Trend following inverts the retail instinct to chase high win rates.

The best trend following systems in history have a win rate below 40%… One huge trade pays for a year of small losses.

Three principles do the work: cut losers fast (the trailing stop), let winners run and add to them (pyramiding — the opposite of averaging down), and diversify across many markets because you can’t know in advance which one will trend. Run trend following on a single market and you’ve gutted it. The emotional cost is real — six losses in a row, then one win that erases them — which is why most people abandon it before it pays.

The build and the backtest

The presenter coded it with Claude Code, giving it reference code and a simple prompt. The only tricky bit for the AI was pyramiding, because the natural (wrong) instinct is to add to losers. He tested six symbols — SPY, QQQ, gold, USO (oil), BTC, ETH — on daily bars over roughly five years, using a single data provider’s API.

The headline result, with no parameter tuning and including commissions and slippage:

Overall, it did a 198% return… around a 16.7% CAGR… only a 12% drawdown… 334 trades… a 33% win rate… and a profit factor of 2.0.

By market: SPY was the weakest (33%), QQQ and gold strong, oil middling (48%), and the two crypto names carried the most because they trended hardest. Verdict: rules written before personal computers existed still seem to work in 2026 — not every market, not every year, but across a diversified basket with strict risk control.

The fine print

The video is paid promotion for Public’s data API, disclosed mid-roll, and ends with a pitch for the presenter’s paid “Crypto Momentum Group.” The strategy demo is partly a vehicle for both.

Key Takeaways

Trend following deliberately accepts a sub-40% win rate; profit comes from winners that are 3–10x the size of losers, not from being right often.
The reconstructed Seykota system: 100 EMA trend filter, 20-day breakout entry, pyramid up to 3 units (each at +0.5 ATR), ATR trailing stop, ATR-scaled position sizing.
Diversification across uncorrelated markets is structural, not optional — you can’t predict which market trends, so you must be in all of them.
Pyramiding means adding to winners, the deliberate inverse of the retail habit of averaging down on losers.
Backtest result: 198% total / ~16.7% CAGR / 12% max drawdown / profit factor 2.0 / 334 trades over ~5 years on six diversified symbols, no optimization.
Crypto (BTC, ETH) drove most of the return because it trended hardest; SPY was the laggard.
The $5k-to-$15M headline is never tested or substantiated — the actual test compounds a notional $100k to ~$300k over six years.

Claude’s Take

The signal-to-noise here is better than the clickbait title implies, but the title earns the skepticism. The “$5,000 into $15 million” is pure thumbnail bait — it’s attributed to Seykota, never verified, and has nothing to do with what’s actually tested. What’s tested is a textbook breakout system producing a 16.7% CAGR. Respectable, real, and roughly in line with published trend-following research — but a different universe from 300,000%.

The educational core is sound and honestly stated: low win rate, fat-tailed winners, cut losers, diversify, pyramid. That’s the actual lesson of trend following and the video gets it right. The catch is the backtest itself. Five years (~2020–2025) is a short, favourable window — a long crypto bull run and a gold rally do a lot of the lifting, and the presenter says as much. No mention of out-of-sample testing, transaction-cost realism on crypto, slippage on breakout fills, or whether the parameters are the ones that happened to survive. “No optimization” is asserted, not shown. And it’s a paid ad for the data provider, with a paid community pitch on the back end, so the incentive is to make the strategy look workable.

Net: a 5. The trend-following primer is worth the eleven minutes and the rules are a fair starting reference. Treat the equity curve as a sponsored demo, not evidence, and ignore the title entirely.