Running 10 strategies at once: what diversification is actually worth
Jim Simons put it in one sentence: what makes alpha is being orthogonal — uncorrelated with everything else. Most retail traders spend their time trying to make one strategy better. The math says a second, unrelated strategy that is merely okay is worth more than a large improvement to the one you already have.
This article puts a number on that with our own backtest engine. Every figure below comes from the ten strategy templates on this site, run over the same 5 Aug 2024 – 2 Aug 2026 window, 1% risk per trade, real spreads.
The ten strategies on their own
Each one is a plain retail setup — Ichimoku, Supertrend, MACD, Bollinger, Donchian, KDJ, RSI — on a different market. Nothing exotic. Run standalone on its own $10,000 account:
| Market | TF · strategy | Trades | Return | Max DD | Sharpe |
|---|---|---|---|---|---|
| USDJPY | H1 · Ichimoku cloud breakout | 240 | +62.05% | 11.10% | 1.69 |
| GER40 | H1 · Supertrend trend-following | 226 | +43.18% | 8.39% | 1.31 |
| BTCUSD | H4 · MACD + ADX filter | 65 | +39.60% | 6.68% | 1.85 |
| BTCUSD | H1 · Triple screen pullback | 278 | +29.37% | 13.74% | 0.74 |
| GBPUSD | H4 · Bollinger mean reversion | 147 | +27.49% | 9.15% | 1.23 |
| XAUUSD | H4 · KDJ oversold pullback | 51 | +25.53% | 5.22% | 1.92 |
| ETHUSD | H1 · MACD + EMA200 filter | 158 | +22.66% | 10.55% | 0.73 |
| XAUUSD | H1 · Donchian channel breakout | 222 | +20.67% | 11.35% | 0.97 |
| BTCUSD | H4 · Ichimoku cloud breakout | 81 | +14.34% | 7.78% | 0.92 |
| GBPUSD | M15 · RSI oversold bounce | 76 | +2.26% | 5.64% | 0.32 |
| Average of the 10 | — | +28.72% | 8.96% | 1.17 | |
One decent performer (USDJPY, +62.05%), one near-flat dud (GBPUSD M15 RSI, +2.26%), and eight in between. If you had to pick one and only one, you would pick USDJPY and live with an 11.10% drawdown. Hold that thought.
The number that matters: 0.014
Correlation of daily returns, averaged across all 45 pairs:
- Mean 0.014, median 0.002
- Most correlated pair: BTCUSD H4 Ichimoku × ETHUSD H1 MACD, +0.226 — two crypto strategies, which is exactly what you would expect
- Least correlated pair: GBPUSD H4 Bollinger × GBPUSD M15 RSI, −0.154 — same market, opposite logic, so one tends to make money when the other does not
This is not a clever result. It falls out of the setup: different markets, different timeframes, different logic. Trend-following on the DAX and mean-reversion on cable have no reason to lose money on the same day. That "no reason" is the free lunch.
What it buys you
Two ways to spend the same discovery. First: keep your total risk the same and split it ten ways — each strategy risks 0.1% per trade instead of one strategy risking 1%.
- Return: +29.46% vs +28.72% for the average single strategy — essentially unchanged
- Max drawdown: 1.76% vs 8.96% — one fifth
- Sharpe: 3.12 vs 1.17
Second: keep the drawdown the same and size up. If a 8.96% drawdown was acceptable with one strategy, it is acceptable with ten. That allows 0.52% risk per strategy:
- Return: +272.59%, max drawdown 8.96% — the same pain as the average single strategy
- Against the best single strategy: +272.59% vs +62.05%, and at a smaller drawdown (8.96% vs 11.10%)
- Sized to match the best single strategy's own 11.10% drawdown (0.66% each): +410.13%
- Portfolio · 0.52% risk each (8.96% DD) → $37,263
- Best single strategy, USDJPY (11.10% DD) → $16,205
- One risk budget split 10 ways · 0.1% each (1.76% DD) → $12,946
All three start at $10,000. The orange line's drawdown (8.96%) is smaller than the blue line's (11.10%) — it is not winning by taking more risk. The grey line shows the same total risk budget split ten ways: the return matches the average single strategy, but the curve is nearly straight.
Where the improvement comes from
Volatility is where you can see the mechanism directly. Annualised, for the ten strategies at 1% risk each:
- If they were perfectly correlated (all moving together): 112.51%
- What actually happened: 41.84%
- Theoretical floor if they were perfectly independent: 38.62%
Diversification removed 62.8% of the volatility, and the result sits within 8% of the mathematical floor. There is almost nothing left to extract — these strategies are about as orthogonal as a set of ten can be.
But more is not automatically better
The obvious next thought is "add everything, correlation will handle it." It will not. Drop each strategy in turn and re-measure the portfolio:
| Remove this strategy | Its own return / Sharpe | Portfolio Sharpe | Portfolio return (same DD) |
|---|---|---|---|
| Keep all 10 | — | 3.12 | +272.59% |
| BTCUSD H1 Triple screen | +29.37% / 0.74 | 3.39 | +249.05% |
| ETHUSD H1 MACD | +22.66% / 0.73 | 3.23 | +263.98% |
| GBPUSD M15 RSI | +2.26% / 0.32 | 3.10 | +280.38% |
| BTCUSD H4 Ichimoku | +14.34% / 0.92 | 3.09 | +248.89% |
| XAUUSD H1 Donchian | +20.67% / 0.97 | 3.04 | +275.30% |
| GBPUSD H4 Bollinger | +27.49% / 1.23 | 2.86 | +208.95% |
| XAUUSD H4 KDJ | +25.53% / 1.92 | 2.86 | +221.90% |
| GER40 H1 Supertrend | +43.18% / 1.31 | 2.85 | +280.93% |
| BTCUSD H4 MACD+ADX | +39.60% / 1.85 | 2.81 | +198.24% |
| USDJPY H1 Ichimoku | +62.05% / 1.69 | 2.70 | +159.69% |
Read the Sharpe column. Removing BTCUSD H1 Triple screen raises portfolio Sharpe from 3.12 to 3.39 — it returned a respectable +29.37% on its own, but it is the fourth crypto strategy in the set, so it adds risk that the other three already carry. Removing the weakest strategy by far, GBPUSD M15 RSI at +2.26%, barely moves anything (3.12 → 3.10): it earns almost nothing, but it earns it at times unrelated to everything else, so it pays for its seat.
The rule is not "keep the profitable ones." It is keep the orthogonal ones. A weak uncorrelated strategy can be worth more than a strong correlated one. That is the whole Simons point, and it is visible in four decimal places of our own data.
When they do fail together
The portfolio's worst stretch was 19 days, 24 Dec 2024 → 12 Jan 2025, −16.53% (at 1% risk each). Here is every strategy over that window:
| Strategy | Over those 19 days |
|---|---|
| GER40 · H1 | -5.56% |
| ETHUSD · H1 | -4.00% |
| BTCUSD · H1 | -3.21% |
| XAUUSD · H1 | -2.65% |
| BTCUSD · H4 | -1.28% |
| GBPUSD · M15 | -0.89% |
| XAUUSD · H4 | -0.80% |
| BTCUSD · H4 | -0.46% |
| GBPUSD · H4 | +0.92% |
| USDJPY · H1 | +0.94% |
Only 2 of 10 made money. The losses cluster exactly where you would guess: DAX, ETH, BTC and gold all fell in the same risk-off move. Low average correlation does not mean low correlation in the moments that hurt — it means the clusters are smaller and rarer.
Worth noting what did not happen: in the portfolio's worst month (March 2025, −6.10%) average pairwise correlation went from 0.014 to just 0.033. The usual "correlations go to 1 in a crisis" did not show up in this sample — but two years is not long enough to claim it never will.
What this backtest does not prove
Being specific about the holes matters more than the headline number:
- These ten were selected on this same history. Their individual edges are partly in-sample. The correlation structure is far more trustworthy than the return figures.
- Two years, 25 months. 22 of them were profitable. That is not a long enough record to rule out a regime that hits everything at once.
- Fat tail. The best five days contributed a large share: excluding them cuts the 1%-risk portfolio from +1029% to +636%. Compounding at ~42% annualised volatility is not a smooth ride, whatever the equity curve looks like from a distance.
- Real-world friction. Ten EAs means ten charts, margin for ten simultaneous positions, swaps, and a VPS that stays up. Spread is in the engine; swap and commission are not.
- The model assumes positions can always be opened. No margin-call logic, no broker-specific limits.
How to reproduce this
Nothing here needs code. Backtest several template strategies on this site, note each one's return and max drawdown, then run them together in MetaTrader 5 on one demo account with the risk per trade divided by the number of strategies. The correlation is what you are buying; you can see it within a month of forward testing.
If you want the exact method: each strategy's equity curve is resampled to calendar days, converted to daily returns, and the portfolio is the sum of those returns scaled by the risk multiplier. Annualisation uses 365 days because the series includes weekends at zero.
FAQ
How many strategies should I run at once?
The volatility benefit is largest going from one to about five, and our ten sit within 8% of the theoretical floor. Past that, adding more only helps if the new strategy is uncorrelated with what you already have — a sixth crypto trend-follower adds nothing.
Should each strategy risk 1% or 0.5%?
Neither is "correct" — it is a leverage choice, not a skill choice. Ten strategies at 1% each produced a 16.53% drawdown in this test; at 0.5% each, 8.55%. Pick the drawdown you can sit through, then divide.
Does a losing strategy still help the portfolio?
A flat one can, if it is uncorrelated — our +2.26% RSI strategy costs the portfolio almost nothing and occasionally profits when everything else is down. A consistently losing strategy does not; diversification reduces volatility, it does not turn a negative expectancy positive.
Can I run 10 Expert Advisors on one MetaTrader 5 account?
Yes — one chart per EA, each with its own magic number so they do not touch each other's positions. Every EA generated here already uses a unique magic number. Check that your margin covers all ten holding positions simultaneously.
Is this the same as buying a portfolio of assets?
No. Asset diversification lowers your exposure to one market while leaving you exposed to the market direction (beta). These strategies go both long and short, so what is being diversified is the edge itself — closer to what Simons meant by alpha.