Portfolio backtesting: combine strategies and check the drawdown

To backtest a portfolio of strategies, add up each strategy's daily profit and loss in money, on one shared calendar, with the contract sizes you would really trade. Then judge the combined curve by its worst drawdown, how long it took to recover, its monthly results, and whether the strategies lost money at the same time.

How do you combine the equity curves of several strategies?

You combine strategies by adding their profit and loss in money, day by day, rather than averaging their percentage returns or their summary statistics. A portfolio backtest is a test that runs several strategies over the same history and measures the one account that holds them all. An equity curve is the line of an account's running total of profit and loss over time. A drawdown is the fall of the equity curve from its last high to its lowest point before a new high.

A free spreadsheet is enough for a first portfolio backtest. You need one column per strategy, one row per day and one column for the sum. Build it in this order:

1. Get daily profit and loss for each strategy, in account currency, with commission and slippage already included. Costs added later change the ranking. See how to model commission and slippage in backtests.
2. Use the contract count you plan to trade for each strategy. Results at one contract each are a starting point, not the final portfolio.
3. Put every strategy on one calendar. A day with no trade gets 0, not a blank. Assign night-session trades to trading dates with the same rule for every strategy.
4. Use daily open equity, not closed-trade equity. Closed-trade equity is a curve that changes only when a trade is closed. It hides losses on positions that are still open. Daily open equity values open positions at each day's close, so overlapping losses show up when they happen.
5. Use the same date range for all strategies. If one strategy's history starts later, start the portfolio on that date. Otherwise the early years show fewer strategies than the later ones.
6. Sum each row and keep a running total. The running total is the portfolio equity curve.

Why do correlation and overlapping drawdowns matter?

Correlation tells you whether strategies usually move together. The overlap of their drawdowns tells you whether they hurt you at the same time, and that second check matters more. Think of two ice-cream stands on the same beach. They are different businesses, but one rainy week empties both.

Correlation is a number from −1 to +1 that shows how closely two series move together. +1 means they always move in the same direction. 0 means no straight-line relation. −1 means they always move in opposite directions. Compute it on daily or weekly profit and loss, not on the equity curves. Two rising equity curves always look correlated, even when the daily results are unrelated.

These bands are rough readings, not laws. One correlation number has a weakness. It is computed over all days, and most days are quiet. Two strategies can be unrelated on quiet days and lose together on the few days that move the market hard. For example, two trend strategies on ES and NQ tend to be hit by the same reversal, because those two stock index futures often move in the same direction.

So check the overlap directly:

1. List the three worst drawdowns of each strategy, with start and end dates.
2. Mark which of them fall in the same weeks.
3. Compute the correlation again using only the portfolio's worst months. If it rises sharply, the diversification disappears when you need it.

How should you size each strategy and split the capital?

Size each strategy so that its bad periods are similar in money. Giving every strategy the same number of contracts usually lets one strategy dominate the portfolio. Position sizing is the choice of how many contracts each strategy trades. Capital allocation is the share of the account's risk each strategy is allowed to use.

A simple equal-risk rule uses the worst drawdown per contract:

Contracts = risk budget ÷ worst drawdown per contract

The risk budget is the amount of money you accept losing on that strategy in a bad period. The worst drawdown per contract is the largest peak-to-low loss that strategy had in the backtest while trading one contract. Round the result down to a whole number.

Three practical points help here:

• Plan for a deeper drawdown than the backtest shows. A common practice is to use 1.5 to 2 times the historical worst drawdown as the planning figure, because the future worst period has not happened yet.
• Check margin when every strategy holds a position at once. Add up the margin your broker charges today for all positions open on the same day.
• Keep the size fixed during the test. Compounding mixes the sizing rule into the result and makes the strategies harder to compare. Test fixed sizes first, and add a sizing rule as a separate step.

What should you check in the portfolio drawdown and periodic results?

Check the portfolio's worst drawdown against what you can sit through. Then read the monthly and yearly table to see whether the result is spread out or carried by a few periods. Drawdown duration is the time from the high before a drawdown until the curve makes a new high.

The diversification check is the most useful single number. With fixed sizes, the portfolio's worst drawdown can never exceed the sum of the individual worst drawdowns. It reaches that sum only when all worst periods line up exactly.

Before you combine anything, review each component on its own. In AlphaPrime (Valynox), the Performance Report shows equity, drawdown, trades and periodic performance for a strategy, including long and short analysis. That lets you see each strategy's bad months before you line them up against the others.

Why can a portfolio of in-sample picks look better than it will trade?

A portfolio built from strategies chosen for their best backtests inherits each strategy's overfitting. It also adds a new layer: the choice of the combination itself. Overfitting is fitting rules to random noise in past data, so that the backtest looks better than the method really is. In-sample data is the history used to choose or tune a strategy; out-of-sample data is history the choice never saw. Selecting the best of many backtests raises the in-sample result while the out-of-sample result tends to fall behind (Bailey et al., 2014). The same logic applies when you choose 5 strategies out of 50 because their past correlations were lowest. You have searched a very large number of possible combinations and kept the luckiest one. Low correlation measured on the same data is part of that luck. For the research behind this, see why the best backtest is usually inflated. One response from the literature is to discount backtest results more heavily the more variants were tried (Harvey & Liu, 2015). Another is to estimate how likely the chosen combination is overfit (Bailey et al., 2017).

A more honest portfolio test follows a fixed order:

1. Split the history into a selection period and a later holdout period.
2. Choose strategies, sizes and the combination rule using only the selection period. Write the rule down before you look further.
3. Run the fixed portfolio on the holdout period without changing anything.
4. Use out-of-sample segments as inputs where you can. Each strategy's joined out-of-sample curve from walk-forward analysis is a fairer input than its full in-sample curve.
5. Compare holdout drawdown and correlation with the selection period. A deeper drawdown and higher correlation in the holdout are normal. A collapse means the portfolio was fitted, not built.

Record how many candidates you tried, because that count tells you how much selection bias to expect. A strategy databank with batch management and revalidation keeps each batch and its candidates together, so the count is not lost. Each strategy can also pass strategy validation, including walk-forward analysis, before it is considered for the portfolio at all.

What this means if you use AlphaPrime: validate each strategy and review its drawdown and monthly results before it enters any portfolio. Then keep the full batch in the databank, so you know how many candidates your final picks were chosen from.