Archetype study
Which strategy archetype actually holds up?
84,940 strategies with random settings across 5 fixed risk levels, each run against EURUSD once per year from 2020 to 2026 — 594,580 backtests, no tuning, real spreads.
Random settings are the point. A hand-tuned preset tells you about the person who tuned it; drawing every period, multiplier and threshold at random tells you about the archetype — how it behaves when nobody has curated it for you. Most random strategies lose money after costs, and they should. What matters here is the relative picture: which archetypes survive being left alone, which are consistent across years, and which quietly destroy accounts.
One setting is not random. Risk per trade is fixed to 2.00% / 2.75% / 3.50% / 4.25% / 5.00%, and every strategy is run at all of them with everything else held identical. So alongside the archetype comparison, this study answers a second question with an actual controlled experiment: what does turning risk up actually do?
Everything here runs on H1. An earlier version of this study spread strategies across bar sizes and found timeframe to be its single largest effect — the median run went from −100% on M5 to −0% on D1, monotonically, because a faster bar means more trades and every trade pays the spread. That made the archetype ranking partly a timeframe ranking. Holding one bar size fixed removes the confound entirely, so every number below is comparable to every other number below.
The six answers
“Best” depends on what you are asking. These six questions have different winners, and the disagreement is the most useful thing in the study.
Median of the 2490 random-setting strategies that traded on H1, over 6 full years.
Share profitable in at least 4 of 6 full years — the highest of any archetype.
Top 5% outcome. Upside says what tuning could reach, not what to expect.
Share that lost 95% or more of the account in a single year.
Median of the 2500 that traded. Random settings punish this archetype hardest.
Share whose random settings never produced a single trade in any year. Not a result — the absence of one.
Typical outcome, by archetype (H1)
Median 6-year return of the H1 random-setting strategies that took at least one trade, chained from their yearly results. Zero is the centre line.
Pooled across all 5 risk levels, which is fair here only because every archetype carries the same number at each — that is the point of fixing risk rather than drawing it. Read the counts as 5 variants of each base strategy rather than as that many independent draws.
What risk per trade does (H1)
Risk is the one setting here that is not random. Every base strategy was run at all 5 levels with everything else — structure, indicators, thresholds, timeframe — held byte-identical, so the difference between these rows is caused by the risk and nothing else.
| Risk per trade | Strategies | Median return | Mean return | Best single result | Profitable | Median drawdown | Blew up |
|---|---|---|---|---|---|---|---|
| 2.00% | 16434 | −74.8% | −65.0% | 1199.6% | 4.8% | −34.3% | 8.2% |
| 2.75% | 16434 | −86.3% | −72.4% | 1473.2% | 4.2% | −44.5% | 11.4% |
| 3.50% | 16434 | −93.2% | −77.6% | 654.7% | 3.7% | −53.1% | 15.4% |
| 4.25% | 16434 | −96.7% | −81.4% | 672.7% | 3.3% | −60.6% | 20.0% |
| 5.00% | 16434 | −98.4% | −84.2% | 578.2% | 3.0% | −67.0% | 24.7% |
The same strategy at 2.00% and 5.00%. Of the 16,434 strategies that traded at both ends, 94% came out worse at the higher risk — the median one by 17.7% of its account. The typical result fell from −74.8% to −98.4%.
And yet the ceiling rose: the best single result went from 1199.6% to 578.2%. That is the whole trade. More risk does not improve the typical outcome — it widens the distribution, and a wider distribution has a higher top and a much heavier bottom. A big number from a high-risk setting is evidence of the width, not of the setting being good.
One caveat on the grid. 69,810 of these strategies set a risk percentage directly and 11,195 carry it inside their own sizing rule (a martingale ladder, a Kelly fraction), where the base risk is set and the rule left intact. The remaining 3,935 trade a fixed lot and have no percentage at all; their position size is scaled across the same span instead, which is proportional exposure rather than literally percent of equity risked.
Consistency, by archetype (H1)
Share of each archetype's H1 strategies that finished profitable in at least 4 of its 6 full years. A high median return earned in one lucky year does not show up here.
Year by year, all archetypes together
The median strategy-year across the whole study. Some years are hostile to almost everything, which is worth knowing before reading any single strategy's result.
The six families (H1)
| Family | Strategies | Median return | Profitable in ≥4/6 | Median drawdown | Blew up |
|---|---|---|---|---|---|
| Regime / quant | 14620 | −88.4% | 2.9% | −46.9% | 11.8% |
| Mean-reversion | 16985 | −89.5% | 2.9% | −49.2% | 12.5% |
| Breakout / volatility | 12215 | −92.0% | 1.7% | −50.3% | 12.0% |
| Structure / flow | 14565 | −92.7% | 1.9% | −51.5% | 13.7% |
| Trend-following | 7360 | −93.5% | 2.5% | −52.6% | 14.0% |
| Sizing / stateful | 16425 | −97.7% | 1.9% | −64.0% | 28.8% |
And what do the good ones look like?
Everything above measures a strategy nobody chose. That is the right way to compare archetypes, and the wrong way to describe what you end up with — you generate a batch and keep the best of it. Drag the control to rank each archetype on its own best strategies instead of its typical one.
The best 5% of each archetype, ranked among themselves. This is closer to what generating a batch and keeping the winners looks like — selected, in-sample results, not a forecast for a fresh strategy.
| Archetype | Family | In slice | Median | Best | Profitable | Consistent | Blew up | Trades / yr |
|---|---|---|---|---|---|---|---|---|
| Anti-Martingale | Sizing / stateful | 124 | 46.8% | 247.0% | 100.0% | 29.0% | 0.0% | 83 |
| Regime Switch | Regime / quant | 125 | 45.3% | 464.0% | 100.0% | 51.2% | 0.0% | 88.5 |
| Adaptive Kelly | Sizing / stateful | 124 | 22.6% | 201.5% | 100.0% | 27.4% | 0.0% | 94.5 |
| Momentum | Trend-following | 123 | 22.4% | 155.4% | 100.0% | 53.7% | 0.0% | 75 |
| Window | Sizing / stateful | 124 | 21.0% | 202.9% | 86.3% | 32.3% | 0.0% | 81.5 |
| Range Fade | Mean-reversion | 120 | 20.5% | 204.0% | 100.0% | 53.3% | 0.0% | 22 |
| Stationary Reversion | Mean-reversion | 121 | 18.9% | 126.6% | 100.0% | 56.2% | 0.0% | 9 |
| Pattern | Structure / flow | 117 | 16.8% | 97.0% | 100.0% | 33.3% | 0.0% | 28.5 |
| Martingale | Sizing / stateful | 125 | 16.8% | 1473.2% | 68.8% | 28.0% | 0.8% | 80 |
| Regime | Regime / quant | 120 | 16.4% | 404.6% | 100.0% | 53.3% | 0.0% | 27.5 |
| Trend Confirmation | Trend-following | 123 | 14.4% | 180.1% | 94.3% | 44.7% | 0.0% | 31 |
| Orderflow | Structure / flow | 122 | 11.4% | 334.3% | 68.0% | 36.1% | 0.0% | 58.25 |
| Reversion | Mean-reversion | 123 | 9.7% | 126.6% | 72.4% | 24.4% | 0.0% | 32.5 |
| Oscillator | Mean-reversion | 119 | 9.7% | 189.1% | 100.0% | 46.2% | 0.0% | 27.5 |
| Volatility Squeeze | Breakout / volatility | 122 | 9.4% | 73.2% | 91.8% | 37.7% | 0.0% | 16.5 |
| Breakout | Breakout / volatility | 123 | 8.4% | 149.2% | 73.2% | 30.1% | 0.0% | 13 |
| Mean Reversion | Mean-reversion | 121 | 6.5% | 126.6% | 85.1% | 24.0% | 0.0% | 40 |
| Microstructure | Structure / flow | 123 | 6.4% | 269.6% | 63.4% | 38.2% | 0.0% | 10 |
| Trend Gate | Regime / quant | 122 | 6.0% | 114.2% | 78.7% | 47.5% | 0.0% | 19.5 |
| Multi-Step | Sizing / stateful | 79 | 5.4% | 71.5% | 100.0% | 45.6% | 0.0% | 25 |
| Statistical | Regime / quant | 120 | 5.4% | 68.1% | 66.7% | 27.5% | 0.0% | 7.25 |
| Stress Reversion | Mean-reversion | 124 | 5.3% | 126.6% | 54.8% | 29.0% | 0.0% | 24.5 |
| Divergence | Mean-reversion | 124 | 3.6% | 154.5% | 58.9% | 37.1% | 0.0% | 85.5 |
| Skew Breakout | Breakout / volatility | 123 | -0.0% | 35.0% | 49.6% | 20.3% | 0.0% | 18 |
| Session | Structure / flow | 121 | -2.3% | 68.1% | 34.7% | 22.3% | 0.0% | 21 |
| Cycle | Regime / quant | 123 | -2.4% | 216.4% | 40.7% | 29.3% | 0.0% | 68 |
| Trend | Trend-following | 124 | -2.5% | 105.7% | 41.9% | 26.6% | 0.0% | 73 |
| Volatility | Breakout / volatility | 120 | -2.8% | 68.1% | 37.5% | 29.2% | 0.0% | 57.75 |
| Flow | Structure / flow | 124 | -3.2% | 37.1% | 37.9% | 25.8% | 0.0% | 43.5 |
| Regime Router | Regime / quant | 124 | -3.7% | 145.8% | 32.3% | 18.5% | 0.0% | 60.25 |
| Signal | Structure / flow | 123 | -5.4% | 68.1% | 33.3% | 12.2% | 0.0% | 80 |
| Protective Stop | Sizing / stateful | 123 | -8.5% | 68.1% | 26.0% | 11.4% | 0.0% | 95 |
| Volatility Expansion | Breakout / volatility | 124 | -20.8% | 68.1% | 13.7% | 8.1% | 0.0% | 82.75 |
| Grid | Sizing / stateful | 125 | -79.1% | -52.2% | 0.0% | 0.0% | 0.0% | 364 |
Ranked on the selected slice, so the ordering always describes the same question the numbers answer. “Blew up” is the share that lost 95% or more in a single year. All rows are H1, EUR/USD, seven years.
Every archetype, every number
Sorted by typical outcome. Every figure is the H1 slice, so a row is internally consistent and comparable to every other row. Everything the charts above draw is here as a number.
| Archetype | Family | Traded | Never traded | Median return | Profitable in ≥4/6 | Top 5% | Median drawdown | Blew up | Trades / yr |
|---|---|---|---|---|---|---|---|---|---|
| Regime Switch | Regime / quant | 2490 | 0% | −53.2% | 5.8% | 15.1% | −29.2% | 8.3% | 132 |
| Multi-Step | Sizing / stateful | 1570 | 37% | −62.2% | 2.9% | 0.9% | −31.9% | 5.2% | 71 |
| Stationary Reversion | Mean-reversion | 2415 | 3% | −80.7% | 4.7% | 6.5% | −42.7% | 9.7% | 128 |
| Adaptive Kelly | Sizing / stateful | 2475 | 1% | −83.4% | 2.1% | 4.9% | −42.9% | 11.8% | 153 |
| Range Fade | Mean-reversion | 2395 | 4% | −86.3% | 3.6% | 5.0% | −46.8% | 12.0% | 150 |
| Volatility Squeeze | Breakout / volatility | 2440 | 2% | −87.0% | 3.5% | −0.5% | −45.5% | 9.7% | 139 |
| Oscillator | Mean-reversion | 2380 | 5% | −87.0% | 3.1% | −0.2% | −46.3% | 11.3% | 137 |
| Statistical | Regime / quant | 2385 | 5% | −88.3% | 2.1% | −2.9% | −46.6% | 11.3% | 153 |
| Session | Structure / flow | 2420 | 3% | −89.2% | 1.6% | −11.2% | −47.6% | 10.1% | 164 |
| Microstructure | Structure / flow | 2445 | 2% | −89.4% | 2.7% | −4.5% | −48.0% | 11.2% | 154 |
| Regime | Regime / quant | 2390 | 4% | −89.6% | 3.3% | 3.2% | −47.6% | 12.5% | 159 |
| Stress Reversion | Mean-reversion | 2465 | 1% | −89.7% | 2.7% | −10.3% | −50.0% | 11.9% | 171 |
| Mean Reversion | Mean-reversion | 2415 | 3% | −89.8% | 1.5% | −2.6% | −49.1% | 13.3% | 186 |
| Pattern | Structure / flow | 2340 | 6% | −90.2% | 2.5% | 3.8% | −49.2% | 11.4% | 173 |
| Trend Gate | Regime / quant | 2435 | 3% | −90.7% | 3.0% | −3.5% | −48.0% | 11.5% | 160.5 |
| Skew Breakout | Breakout / volatility | 2460 | 2% | −91.2% | 1.3% | −14.2% | −48.6% | 11.1% | 163 |
| Reversion | Mean-reversion | 2450 | 2% | −92.2% | 1.8% | −8.6% | −52.8% | 14.7% | 186 |
| Volatility Expansion | Breakout / volatility | 2465 | 1% | −92.4% | 0.5% | −38.3% | −52.0% | 10.6% | 161 |
| Breakout | Breakout / volatility | 2450 | 2% | −92.4% | 1.8% | −5.6% | −50.1% | 11.9% | 167 |
| Trend Confirmation | Trend-following | 2445 | 2% | −92.5% | 2.6% | −0.4% | −51.5% | 13.5% | 176 |
| Regime Router | Regime / quant | 2465 | 1% | −92.9% | 1.5% | −19.5% | −52.0% | 12.8% | 185 |
| Momentum | Trend-following | 2450 | 2% | −93.5% | 3.5% | 1.3% | −52.8% | 14.2% | 186 |
| Flow | Structure / flow | 2465 | 1% | −94.0% | 1.4% | −18.5% | −53.0% | 14.6% | 222 |
| Divergence | Mean-reversion | 2465 | 1% | −94.0% | 2.7% | −13.7% | −54.7% | 14.4% | 177 |
| Protective Stop | Sizing / stateful | 2460 | 2% | −94.3% | 0.8% | −26.4% | −50.8% | 12.3% | 258 |
| Trend | Trend-following | 2465 | 1% | −94.3% | 1.4% | −23.6% | −53.6% | 14.4% | 182 |
| Cycle | Regime / quant | 2455 | 2% | −94.5% | 1.7% | −22.0% | −53.6% | 14.5% | 214 |
| Signal | Structure / flow | 2455 | 2% | −94.6% | 0.8% | −25.8% | −54.6% | 14.9% | 187 |
| Volatility | Breakout / volatility | 2400 | 2% | −95.1% | 1.7% | −20.8% | −55.5% | 17.0% | 200 |
| Orderflow | Structure / flow | 2440 | 2% | −96.2% | 2.6% | −14.9% | −57.0% | 20.2% | 205 |
| Window | Sizing / stateful | 2465 | 1% | −97.1% | 2.4% | −2.9% | −62.0% | 22.6% | 166 |
| Anti-Martingale | Sizing / stateful | 2465 | 1% | −97.7% | 2.8% | 7.6% | −67.0% | 24.7% | 116 |
| Martingale | Sizing / stateful | 2490 | 0% | −100.0% | 2.5% | −15.0% | −86.3% | 53.2% | 100 |
| Grid | Sizing / stateful | 2500 | 0% | −100.0% | 0.0% | −89.5% | −94.6% | 62.4% | 301 |
How this was run
- Random settings, real structures. Each strategy takes an existing structure for its archetype — which indicators, which conditions, which exit modules — and draws every numeric setting at random: lookbacks 2–200, multipliers 0.5–6, reward:risk 0.5–5. Nothing is optimised, and no result was discarded for being bad.
- Risk per trade is the exception — it is controlled. Every other setting is drawn at random; risk is fixed to 2.00% / 2.75% / 3.50% / 4.25% / 5.00%, and 16,988 base strategies are each run at all of them. That is deliberate: when risk was drawn at random it turned out to be the single strongest driver of the result, which also made it a confound — an archetype whose structures happened to draw more risk looked worse for a reason that had nothing to do with the archetype. Fixing it balances every archetype on risk, and makes the risk effect itself measurable within one strategy.
- Six full years, not seven. Runs cover 2020–2026, but the bar data stops at 2026-06-30, so 2026 is a half year. It is shown in the year-by-year section marked ½ and excluded from every chained return and consistency figure — counting a half year as a year would understate both. Consistency therefore means profitable in at least 4 of 6 full years.
- One run per calendar year, each starting from the same account. That is what makes consistency answerable — a single compounded number cannot tell “up every year” apart from “one lucky year”. The multi-year figure chains those yearly results.
- Only strategies that traded are scored. Random settings often produce a strategy that never fires — a 200-bar lookback with a three-sigma trigger can go a whole year without a signal. Those return exactly 0.0%, which is not a result but the absence of one, so they are excluded from every return, consistency and blow-up figure and reported separately as never traded. It matters more than it sounds: scored the other way, the archetype that does nothing most often would top the study at 0.0% while the strategies it actually ran lost almost everything.
- Real costs. Every run uses the recorded per-bar spread from our own tick-derived data, not a frictionless model. Costs alone are enough to turn a marginal strategy into a losing one.
- EURUSD only. One instrument. An archetype that suits gold or an index may rank differently there.
- One timeframe for the rankings. Every strategy in this run was generated on H1. An earlier version spread them across bar sizes and found that the single largest effect in the study was the bar size itself, not the archetype — so it is held fixed here and the comparison is clean by construction.
- Reproducible. Seed 20260808, 2500 strategies per archetype. The same seed regenerates the same strategies.
- What this is not. Not a forecast, and not a claim about any tuned strategy — including ours. It is a map of which archetypes are forgiving and which are unforgiving when left entirely alone.
Curious how a tuned one compares? The parity study runs real strategies through both our engine and the MetaTrader 5 tester on the same 24.6-million real ticks, and composing your own takes a sentence.