RESOURCES / BACKTESTING FOUNDATION
Portfolio Backtesting Guide
A useful backtest begins with a decision, not with a search for the highest historical return.
Questions about backtesting guide.
What should I define before running a portfolio backtest?
Start with a decision you want to investigate, not a ticker you hope will win. Define the reference portfolio, each asset and target weight, the initial capital, the observation window, the rebalancing rule, and whether contributions or withdrawals occur. PortfolioBacktest turns those choices into a visible calculation record before presenting a result. This matters because an apparently precise return can be misleading when the portfolio definition is incomplete. A complete reference gives every later variant a stable baseline and makes it possible to explain what actually changed.
Why does PortfolioBacktest change one variable at a time?
A controlled comparison changes one meaningful decision while holding the other rules constant. For example, you can move 10% from an equity ETF to gold, change annual rebalancing to quarterly, or alter the start date—but not all three at once. The engine then reports the difference in ending value, annualized return, drawdown, volatility, recovery, and consistency. This does not prove that the changed variable caused every observed movement, but it gives the evidence a defensible interpretation. When several variables move together, even a large improvement is difficult to attribute or reproduce.
How should I read a backtest result instead of chasing the highest return?
Read the outcome as a trade-off. First compare ending value and annualized return, then inspect the path that produced them: maximum drawdown, time below the prior high, volatility, down-market capture, and relative performance versus the reference. A portfolio that finishes higher may have demanded losses that an investor could not hold through. The Portfolio Contribution and Alpha & Relative sections help identify which asset or period created the difference. The useful conclusion is not simply ‘A won’; it is ‘A improved this objective while accepting these costs.’
What does testing different start dates tell me?
Start-date testing reveals whether the conclusion is broad or concentrated in one historical episode. A single window can begin just before a crash, a commodity surge, or an unusually strong technology cycle. Re-running nearby years and market regimes shows how often the direction of the result survives. PortfolioBacktest keeps the rules, capital, and comparison method fixed while the date changes, so the test isolates window dependence. This is not a forecast of future returns; it is a check against presenting one lucky historical path as a universal answer.
How does the tool make a backtest reproducible?
A useful result records the inputs that produced it: symbols, weights, dates, rebalancing, contribution assumptions, data treatment, common observation dates, calculation version, and the resulting output. AI may help translate a plain-language question into those fields, but deterministic code performs the price alignment, portfolio updates, and risk calculations. The report separates observed historical calculations from causal claims and future forecasts. Reproducibility therefore means another reader can inspect the same definition, use the same data snapshot, and understand why the displayed result looks the way it does.