How to Read a Backtest Report: Costs, Drawdown, Trade Log

TL;DR: Read a backtest in order of how easily each number can be inflated: first the assumptions (data, fills, costs), then a selection-adjusted statistic such as the Deflated Sharpe Ratio and PBO, then drawdown depth and duration, then trade count and concentration, and only then the headline return. The trade log is where claims get checked. A result that survives that order is worth paper trading; one that does not is still useful, because it saved you money.

Why does the order you read a backtest matter?

Most backtest reports put the biggest, friendliest number at the top: total return. It is also the number most sensitive to leverage, lucky timing and missing costs. If you read it first, everything after it gets interpreted in its light.

A better habit is to read from the least flattering evidence to the most. Start with what the simulation assumed, then check whether the edge survives a correction for luck, then look at the pain, and finish with the gain. The principles behind the first step are covered in our guide to point-in-time backtesting.

Step 1: What did the backtest assume?

Before any metric, answer four questions:

• Data. Was every input knowable at each bar's close? Were any inputs simulated or missing?
• Fills. Do entries fill at the next bar's open or at the signal close? What happens when a take-profit and stop-loss share a bar?
• Costs. Are fees included? Is the result gross or net of slippage and funding?
• Window. Which period does it cover, and does that period include both rising and falling markets?

On Trigr these answers are part of the result. Every run includes an audit trail that flags which inputs were real and which were simulated. Fees are always applied, and a result without slippage and funding is labeled gross. Signals fill at the next bar's open. When a take-profit and stop-loss share a bar, the engine replays 5-minute candles if the strategy's timeframe is coarser, and the stop wins if that still cannot decide; a stop gapped through fills at the worse open. All of this is explained in next-bar-open fills. If you see a gross result, rerun it with the costs described in slippage and funding in perp backtests before going further.

Step 2: Is the edge more than luck? DSR and PBO

The next question is whether the result could plausibly come from trying many ideas and keeping the best one.

The Deflated Sharpe Ratio (DSR), introduced by Bailey and López de Prado, is the probability that the strategy's true Sharpe is above the best Sharpe you would expect from luck across all the variants tested. Values close to 1 are strong evidence; values near 0.5 or below mean the result is hard to distinguish from search. The full intuition, with a table by trial count, is in the Deflated Sharpe Ratio explained.

The Probability of Backtest Overfitting (PBO), from Bailey, Borwein, López de Prado and Zhu, estimates how often the configuration that looks best in-sample ranks below the median out-of-sample. Lower is better; a PBO around 0.5 means your selection process is roughly as good as picking at random.

On Trigr, both statistics come from Optimize with ML runs, which report the DSR with the run's own trial count and PBO from a separate combinatorial validation grid, next to out-of-sample Sharpe. A standard backtest does not compute them: its headline row shows net return, win rate, raw Sharpe and max drawdown, and the "Deep dive" section lists DSR and PBO as "Not yet computed." Until they exist, read the raw Sharpe as optimistic, because it has not been corrected for how many variants you tried.

Step 3: How bad did it get? Drawdown and underwater time

Drawdown is the fall from a previous equity peak. Two numbers matter:

• Maximum drawdown depth. A −40% drawdown needs a +67% gain to recover; a −60% drawdown needs +150%.
• Longest time underwater. How many months passed between a peak and the next new high. Long underwater periods are where people abandon strategies.

Ask yourself honestly whether you would have kept running the strategy through its worst stretch. If not, its average return is irrelevant to you. Also remember that a backtest's worst drawdown is a sample of one; live trading can be worse. Backtests are not guarantees; perps are leveraged and can lose more than expected.

Step 4: Is there enough evidence? Trade count and concentration

Statistics need observations. A strategy with 25 trades over three years has little to say, no matter how good its numbers look.

Look at:

• Trade count. A few dozen trades leave wide error bars.
• Concentration. What share of total profit comes from the top three trades? If removing them turns the result negative, the strategy may depend on rare events it cannot reliably repeat.
• Win rate with payoff. A 35% win rate is fine if winners are much larger than losers; a 90% win rate is fragile if a single loss can erase months of gains.
• Profit factor. Gross profits divided by gross losses. Close to 1 means the strategy barely pays for its losers.

Step 5: Read the trade log and monthly returns

The trade log is where every other claim is checked. Open it and sample a few trades by hand:

• Do entry prices match the next bar's open after the signal?
• On bars where both the target and the stop were in range, does the exit match the first level touched on finer data, or the stop when that cannot be decided?
• Are there clusters of trades in one short period that account for most of the result?

Monthly returns show whether performance is spread across time or bunched into one regime. A strategy that made all its money in a single bull quarter is really a bet on that quarter repeating.

Finally, compare the equity curve with buy-and-hold of the same asset. A leveraged long strategy that tracks the market is not an edge; it is exposure you could get more cheaply.

The trailing 25% shading

Trigr shades the trailing 25% of a full-history backtest so you can compare recent behavior with the earlier curve. Use it to spot degradation. Do not treat it as an out-of-sample test: once you inspect it while choosing or tuning a strategy, it is part of your research data. For genuine out-of-sample evidence, use the anchored, purged walk-forward results from ML optimization, or freeze the strategy and forward-test it on a paper agent.

A worked example: reading one hypothetical result

Here is a hypothetical 4H ETH trend strategy, read in the order above. The numbers are invented for illustration, and the DSR and PBO assume an ML optimization run has been done on it:

Read top-down, this is a strategy to keep researching, not to deploy. Read bottom-up, a 1.4 Sharpe and +85% return look attractive. That difference is the whole point of reading in order.

A one-page backtest reading checklist

Next steps

Take the last backtest you were excited about and read it again in this order, starting with the cost labels. To compare against strategies that were published with a stored, verified full-history backtest, browse the Trigr marketplace, or read the backtesting docs for every assumption Trigr makes.