Combining Multiple Trading Strategies: The Ensemble Approach to Bot Trading

A trader builds a great trend-following bot. It returns +60% in year one. They feel brilliant. In year two, the market regimes shift to chop and the same bot returns -8%. They feel less brilliant. In year three, trends return and the bot prints +45%. Over three years: +97% — but with a brutal mid-period drawdown that would have caused most traders to stop the bot before recovery.

The fix isn't a better trend-following strategy. It's running multiple strategies that work in different regimes, sized so their combined equity curve has materially lower drawdown than any individual one. This is ensemble trading — and the math is unambiguous: a portfolio of three uncorrelated strategies produces roughly 1.7x the Sharpe ratio of any single strategy in the set, with no decrease in expected return.

What is an ensemble trading strategy?

An ensemble is a portfolio of two or more independent trading strategies running simultaneously, with capital allocated across them according to a weighting rule. Each strategy operates as if it were a standalone bot. The portfolio's P&L is the sum of all the bots' P&Ls — but the combined equity curve is smoother because the strategies don't all win or lose together.

The key word is *uncorrelated*. Two trend-following strategies on EURUSD and GBPUSD aren't an ensemble — they'll win and lose together because the pairs correlate strongly. Two strategies are properly uncorrelated when their monthly returns have a correlation coefficient below ~0.3.

Why ensembles outperform single strategies

The math comes from portfolio theory. For two strategies with returns R₁ and R₂, standard deviations σ₁ and σ₂, and correlation ρ, the portfolio standard deviation is:

σ_portfolio = √(w₁²σ₁² + w₂²σ₂² + 2w₁w₂ρσ₁σ₂)

When ρ < 1 (the strategies aren't perfectly correlated), σ_portfolio is less than the weighted average of σ₁ and σ₂. Lower portfolio standard deviation with the same expected return means a higher Sharpe ratio.

For three uncorrelated strategies (ρ ≈ 0) with similar Sharpe ratios, the combined Sharpe is roughly √3 ≈ 1.73x the individual Sharpe. A portfolio of three Sharpe-1.0 strategies has a Sharpe of about 1.73. Same returns, less volatility.

For the deeper context on why this works, our piece on why you need uncorrelated strategies covers the correlation diagnostics. This article focuses on building the ensemble itself.

How to pick strategies for an ensemble

Three rules cover most of the discipline:

1. Different strategy *types*, not just different parameters

A 50/200 EMA crossover and a 20/100 EMA crossover are not different strategies. They're parameter variations of the same trend-following idea — they'll win and lose at roughly the same times. Real ensembles combine different *types* of edge:

Pick one strategy from each family. Three different families is the minimum for meaningful diversification.

2. Different instruments where possible

Trading the same instrument with multiple strategies creates correlation through the instrument itself — even uncorrelated strategy logic produces correlated trades when both strategies are reacting to the same price action.

Better: trend on gold (XAUUSD), mean-reversion on EURUSD, breakout on BTCUSD. Different instruments respond to different macro factors, which reduces correlation at the portfolio level.

3. Different timeframes

A 1H strategy and a daily strategy trade at different frequencies. Their drawdowns occur at different times. Mix timeframes deliberately — at least one strategy on a sub-1H timeframe, one on 1H-4H, one on daily.

A working three-strategy ensemble for retail

A concrete example based on what holds up in backtests on real instruments:

Strategy A: Gold trend-following on 1H

EMA crossover (20/50) with ATR stops, trades only during London-NY session, no news days. Backtested 2020–2025 on XAUUSD: ~+40% annual, max drawdown ~18%, Sharpe 1.1.

Strategy B: EURUSD mean-reversion on 4H

RSI-based reversion with Bollinger band confirmation, trades during low-volatility regimes (ATR below 90-day median), excludes news weeks. Backtested 2020–2025: ~+25% annual, max drawdown ~12%, Sharpe 0.9.

Strategy C: BTCUSD breakout on 1D

Donchian channel breakout with trailing stop, position-trades multi-day moves. Backtested 2020–2025: ~+90% annual (crypto volatility helps), max drawdown ~35%, Sharpe 0.95.

Correlation matrix across the three (monthly returns 2020–2025): all pairwise correlations below 0.25. The strategies genuinely move independently.

Combined ensemble with equal weights: ~+50% annual return, max drawdown ~14%, Sharpe ~1.6. The individual strategies were good. The portfolio is better than any of them.

How do you allocate capital across an ensemble?

Three approaches in order of complexity:

Equal weight

Allocate the same dollar amount to each strategy. Simplest, works surprisingly well, and doesn't require ongoing rebalancing.

Risk parity

Allocate so each strategy contributes the same dollar volatility to the portfolio. A higher-volatility strategy gets less capital; a lower-volatility strategy gets more. The math is straightforward but requires periodic rebalancing as strategy volatilities drift.

Mean-variance optimisation

The formal portfolio-theory optimum. Weights solve for the highest expected Sharpe given the correlation matrix and expected returns. Theoretically optimal, practically fragile — the optimisation is sensitive to input estimates, and small changes in expected return estimates produce large allocation changes.

For most retail traders, equal weight is the right starting point. Risk parity is the next level. Mean-variance optimisation is for traders with strong views on expected returns who can stomach the parameter sensitivity.

Should I run multiple bots on the same broker account?

You can. PineForge supports multiple bots on a single MT5 account, isolated via magic numbers so each bot's trades don't interfere with the others'.

The constraints to watch:

For larger ensembles (4+ strategies), splitting across two or three broker accounts can be operationally cleaner. Each account has independent margin, simpler P&L tracking, and reduced single-point-of-failure risk.

How do you rebalance an ensemble portfolio?

Three approaches:

For most retail ensembles, annual rebalancing is the sweet spot. Frequent enough to prevent drift, infrequent enough that transaction costs don't dominate.

What if a strategy in the ensemble starts decaying?

Stop the decaying strategy. Don't keep capital allocated to it hoping for recovery. See our piece on detecting strategy decay for the signals.

When you stop a strategy, redistribute its capital to the remaining strategies (proportionally to their existing weights) or to a new strategy if you have one ready. The ensemble has fewer strategies for a period, which slightly reduces diversification benefit — accept the reduction rather than hold capital in a known-failing strategy.

The discipline: an ensemble is not a static portfolio. Strategies enter and exit over time as decay and discovery proceed. The portfolio composition evolves; the portfolio approach is constant.

Common ensemble mistakes

Four patterns that show up repeatedly:

How does an ensemble change my [risk management](/blog/risk-management-strategies)?

Risk management operates at two levels: per-strategy (each bot's individual stop, position size, daily limits) and per-portfolio (combined exposure caps, account-level daily loss limit, drawdown thresholds that trigger rebalancing or pause).

The portfolio level is where many retail ensembles fail. They set rigorous per-strategy risk rules and no portfolio-level rules. Then a correlated drawdown across multiple strategies produces a portfolio loss larger than any single strategy could.

The fix: set a portfolio-level maximum drawdown threshold. When the combined portfolio hits, say, 12% drawdown from peak, pause all strategies for review. This isn't a stop-out — it's a checkpoint. Resume after verifying that the drawdown is consistent with normal variance.

Conclusion

Single-strategy bots have an inherent ceiling on risk-adjusted returns. The math of correlation makes a portfolio of uncorrelated strategies strictly better than any individual strategy at the same expected return. The improvement isn't marginal — it's roughly 1.7x Sharpe for three uncorrelated strategies.

Build three strategies of different types, on different instruments, on different timeframes. Allocate equal weights. Rebalance annually. Monitor each strategy independently for decay. Stop the ones that decay and redeploy capital to new strategies you've validated.

Backtest each candidate on PineForge before adding it to the portfolio. Test the ensemble itself — combining the trade-level outputs of multiple backtests gives you the actual portfolio characteristics, not just theoretical projections. The traders who consistently print across regimes aren't the ones with the best strategy. They're the ones running the best portfolio.