At Prismatic Algos, portfolio construction is central to our research: how do we combine algorithms without concentrating the same risks? Here we share our approach to five decisions that deserve attention before systems trade together: understanding the rules, comparing losses, mapping shared exposure, sizing positions and recording the evidence.
Strong standalone backtests can hide a fragile combination. Different algorithms may lose together, depend on the same market conditions or compete for the same account capital. These five checks will help you uncover those weaknesses before you commit capital. They provide a review framework, not a guarantee that a portfolio will avoid losses.
1. Write down what each algorithm does
Before comparing returns, describe each algorithm's entry rule, exit rule, position sizing and operating conditions. An indicator name alone does not explain the system.
For example, two EAs might use different indicators but both buy after an upward move and exit after a reversal. That gives you a hypothesis about shared exposure to investigate. It does not establish their correlation or prove that they will lose together.
2. Compare losses on the same timeline
Use the same dates and a consistent return definition. Inspect marked equity where available, including open positions, rather than relying only on the balance after trades close. Look at when drawdowns begin, how long recovery takes and whether losses overlap.
Illustrative arithmetic, not a backtest: suppose two separate sleeves each begin a period with $5,000. If both lose 4%, each loses $200 and the combined $10,000 falls by $400, or 4%. If one instead gains 4%, the two dollar changes offset for that period. The example assumes fixed starting allocations and returns after costs. It says nothing about future returns or the path of equity within the period.
The useful information is the relationship between the results. Historical differences can disappear when conditions change. Repeat the comparison across different periods instead of treating one full-period correlation as a permanent property.
3. Map shared market and execution exposure
Several symbols may express a similar currency or market exposure. Several systems may also enter during the same session, hold through the same announcements or depend on narrow spreads. List these dependencies before assuming that more EAs mean more diversification.
| Check | Record for each system |
|---|---|
| Market exposure | Symbols, direction and overlapping positions |
| Loss timing | Drawdown dates and recovery periods |
| Execution | Spread, commission, slippage and latency assumptions |
| Operating conditions | Sessions, holding periods and rules for pausing |
4. Evaluate sizing at account level
Equal lot sizes do not imply equal risk across instruments. Contract specifications, price movement, stop distance and concurrent positions all matter. Examine the combined exposure and margin requirements under the proposed settings.
Standalone tests do not automatically reproduce an account running every system together. Check whether the portfolio study models shared capital, simultaneous orders and account limits. If it only aligns separate return series, label it a reconstruction and describe the missing interactions.
5. Keep a decision record
For each candidate, write down why it belongs in the research set, the evidence supporting that choice and what would cause you to reconsider. Keep code versions and settings with the results so a later change does not silently replace the system you evaluated.
- Can you explain each system's entry, exit and sizing rules?
- Are return periods and cost assumptions comparable?
- Have you checked simultaneous losses and open-position exposure?
- Does the combined study reflect account-level constraints?
- Are historical tests, reconstructed results and live records labeled separately?
What to do next
Use these checks to find the unresolved questions in a proposed portfolio. Diversification may change the pattern of risk, but it does not prevent substantial losses or guarantee better returns. Prismatic's portfolio v3 remains in development; this guide is not evidence of its performance.
For a technical reference, MQL5's portfolio-trading article explores implementation examples in MetaTrader 4; it is not a current operating manual for every MT5 setup.
