This guide is written for advanced traders running multiple Pine strategies, markets or timeframes, and it explains how to evaluate a multi-strategy portfolio backtest from the perspective of running everything together rather than comparing standalone results. It moves through the direct answer, standardizing before combining, the meaning and definition of correlation, allocation methods, combined equity and portfolio drawdown, clustering, a volatility calculator, and how to apply stress. If you are still unsure how to read a single backtest, it helps to first grasp the complete TradingView backtesting workflow so this article sits in context.

The 30-second answer

When you evaluate a portfolio of strategies, simply lining up good ones is not diversification on its own. The benefit of diversification is decided by when and how much each strategy loses at the same time. So the first thing to do is align every strategy’s profit and loss to the same time axis, the same currency and the same starting capital, and build a return series on a common calendar.

On that foundation you then check the correlation between strategies, the combined drawdown after allocation, and the simultaneous losses during a sharp sell-off. Even when correlation looks low in normal conditions, it can spike during stress and pull several strategies down together. Below, using fictional educational data for three strategies (trend-following, mean-reversion and breakout), we look concretely at how to separate apparent diversification from real diversification.

Standardize before combining

Before combining strategies, align the basis for comparison. Skip this step and the correlation and allocation that follow lose their meaning. At a minimum, standardize the following.

• Period and time zone: Trim each strategy’s coverage to a common start and end, and unify the time zone. If the date basis is misaligned, you will under- or over-state simultaneous losses.
• Currency: Convert profit and loss into a base currency. Adding raw amounts without converting lets exchange-rate moves leak into strategy correlations.
• Starting capital: Assume the same starting capital and align on return rate rather than amounts, so differences in size do not distort each strategy’s contribution.
• Trading costs: Unify spread, commission and slippage assumptions across strategies. Inconsistent assumptions push combined P&L away from reality. Modeling them is covered in cost sensitivity and stress testing.
• Return frequency: Decide on the frequency used for correlation and combination (daily, weekly and so on) and apply it to every strategy. Mixed frequencies make correlation coefficients incomparable.
• Missing data and overlapping positions: Decide how to treat gaps in the data and set a rule for aggregating positions when several strategies hold the same instrument in the same direction.

This standardization is unglamorous, but the reliability of portfolio testing is largely decided here. Proceed with a misaligned foundation and the correlation matrix and post-allocation drawdown you read later will reflect the roughness of your preprocessing rather than the nature of the strategies.

What correlation means and its limits

The correlation coefficient expresses, on a scale from −1 to +1, how much two strategies’ returns move together in the same direction. Closer to +1 means they rise and fall in step, closer to 0 means they are near-independent, and closer to −1 means they move oppositely. For diversification, the lower the correlation between combined strategies, the more the combined volatility and drawdown tend to fall below the sum of the individual parts.

But correlation has clear limits. First, it captures only linear co-movement. Non-linear links and structural dependence on the same driver are missed. Second, correlation depends on the period. Even when it is low in calm times, a sharp sell-off can push risk assets down together and drive many strategies’ correlations toward +1, a “correlation convergence.” In other words, low correlation in normal conditions is not guaranteed to hold during the very stress when you most need it. That is exactly why you should not judge diversification from a single correlation value; read it alongside simultaneous drawdowns and behavior by regime. For a view of the market backdrop, an external perspective such as the Macro Research Workbench can also help.

Which return series to correlate

When measuring correlation, the key implementation point is not to compute it trade-by-trade. Each strategy’s trades occur at different times and are held for different durations, so lining up per-trade profit and loss gives no shared time axis. Instead, convert everything into return series on a common calendar.

The implementation definition

This article defines correlation as follows. For each strategy, build a daily return series (that day’s percentage change in equity) over the common period. Days with no trade are set to a zero return so the series stays continuous. Then compute the Pearson correlation coefficient between two strategies’ daily return series. Frequency is unified at daily because mixed frequencies make correlations incomparable across strategies. If you want to match your operating granularity to weekly, align every strategy to weekly and recompute.

Under this definition, the correlation matrix in Figure 1 expresses “how much they moved together on the same day.” How to read standalone metrics such as trade count and win rate is covered in how to read backtest results, so understanding each strategy’s makeup before reading correlation keeps your interpretation stable.

Comparing capital allocation methods

Allocation is the step where you decide how much capital to assign to each strategy. Here we compare four representative methods descriptively. Each has trade-offs, and none is universally superior. This article does not recommend any particular allocation or a single optimum.

• Equal-capital: Assign the same amount of capital to each strategy. Simple to implement, but a high-volatility strategy tends to dominate the portfolio’s risk.
• Equal-risk (risk parity): Allocate inversely to each strategy’s assumed volatility, aiming to equalize risk contributions. It does not behave as intended if the volatility estimate is off.
• Fixed-weight: Fix the ratios at the manager’s discretion (for example 50/30/20). The stance is clear, but the rationale can become subjective.
• Capped: Set a ceiling on any one strategy’s weight (for example 40%) to limit concentration. Diversification is easier to preserve, but a mechanical cap can introduce inefficiency.

The table below applies each method to the three illustrative strategies, together with the combined annualized volatility and combined maximum drawdown as fictional educational data. Note that under equal-risk, low-volatility Strategy B (assumed 12%) receives a heavier weight and high-volatility Strategy C (assumed 24%) a lighter one.

The numbers in the table are fictional educational data meant to explain the allocation concepts. Combined annualized volatility and combined maximum drawdown change substantially if the correlation and volatility assumptions change. They do not show that any particular allocation is good or safe.

Combined equity and portfolio drawdown

Once weights are set, weight each strategy’s return series by its allocation and add them together to build the combined equity curve. From that curve, recompute the portfolio maximum drawdown. The important thing here is not to simply add up each strategy’s maximum drawdown. When correlation is low, one strategy holds up while another sinks, so the combined drawdown tends to be shallower than the sum of the individual ones. Conversely, when highly correlated strategies sink together, it can be deeper than expected.

The effect of combining is tested not only by calm-period smoothness but during simultaneous losses. The next figure decomposes each strategy’s monthly contribution during a sharp sell-off month (fictional educational data) and the combined profit and loss of the equal-capital portfolio. You can read the structure: daily trend-family strategies A and C swing negative together, while mean-reversion Strategy B moves slightly the other way and cushions the fall.

Clustering to spot duplicate strategies

You can line up several strategies, but if they behave essentially the same, you have merely concentrated on one bet while believing you diversified. In the correlation matrix of Figure 1, strategies A and C correlated highly at 0.62. Both are daily trend-family strategies: they win together in the same trending phase and lose together at the same reversal. Grouping highly correlated strategies surfaces fewer clusters than the apparent count of strategies.

In this example, strategies A and C form one cluster (trend-family) and Strategy B forms another (mean-reversion). So even though it looks like “three strategies,” the number of independent sources of movement is effectively two. Clustering and similarity diagnostics help visualize such duplication numerically and check whether the allocation is skewed toward a particular cluster.

Portfolio volatility calculator

How correlation changes combined risk can be felt intuitively with a simple two-strategy formula. Enter two strategies’ assumed annualized volatility, correlation coefficient and allocation weights into the calculator below and it displays a simplified portfolio volatility. The assumptions and formula stay visible at all times, and it does not output an optimal allocation. Values are computed entirely in your browser and never sent externally. If JavaScript is disabled, see the static worked example directly below.

This calculation is a conceptual simplified formula for showing how correlation affects risk; it does not indicate an optimal allocation or future risk. Real portfolios involve three or more strategies, time-varying correlation and non-linear joint risk that a single formula cannot fully capture. If the assumed volatility or correlation is wrong, the result changes too.

Stress-testing the historical optimum

An allocation that minimizes drawdown or maximizes Sharpe on past data is merely a result fitted to that period’s correlations and volatilities. Change the market regime and that allocation breaks down easily. So evaluate an allocation not as a single optimal point but as a distribution of outcomes when you shake the assumptions. How to spot overfitting itself is covered in how to detect backtest overfitting with parameter stability, and validation across time splits in walk-forward analysis and out-of-sample testing.

The angles for stressing an allocation are as follows.

• Out-of-sample (OOS): Check the same allocation’s combined results over a period different from the one used to set it.
• Correlation shift: Recompute combined drawdown using correlation restricted to sell-off phases rather than calm-period correlation.
• Trading costs: Set spread, commission and slippage on the heavier side and see whether the ranking of contributions changes.
• Constraints and caps: See how combined drawdown and contributions move once you cap any single strategy or cluster.
• Rebalancing frequency: Vary the interval (monthly, quarterly and so on) and check that results are not overly sensitive to frequency.

The ordering of simultaneous losses and the impact of rare losing streaks are invisible in a single scenario. Combining Monte Carlo testing of risk of ruin, which randomly reshuffles losing streaks and sequence risk, with reading maximum drawdown for combined depth and recovery, lets you check the robustness of an allocation from several angles. A capital-allocation simulation is ultimately a distribution of outcomes and does not guarantee future returns or optimal management.

Practical checklist

• Did you unify period, time zone, currency, starting capital, cost and return frequency across all strategies?
• Did you compute correlation on common-axis return series with an explicit definition, not on individual trades?
• Did you check not only calm-period correlation but also correlation and simultaneous drawdown during sell-offs?
• Did you compare equal-capital, equal-risk, fixed-weight and capped descriptively rather than depending on a single optimum?
• Did you compute combined maximum drawdown from the combined equity curve rather than summing the individual ones?
• Did you group highly correlated strategies into clusters and separate the apparent count from the effective count?
• Did you stress the allocation with OOS, correlation shift, cost, constraints and rebalancing frequency, and view results as a distribution?

Frequently asked questions

Related reading and next steps