MyAlgoGuru Help
Education Glossary Open the app
Chapter 5 of 5

Reading a backtest report without flattering yourself

Total return is the least informative number on the page. Here is what to read instead, in the order that matters.

A backtest report gives you twenty numbers. Two or three of them decide whether the strategy is worth running, and the one everybody looks at first is not among them.

Start with the trade count

Before any performance figure, find out how many trades produced it.

This single number sets the weight everything else can carry. A 68% hit rate across 1,400 trades is a property of the strategy. A 68% hit rate across 19 trades is thirteen winners, which is roughly what you get from a coin a third of the time. Nothing else on the report can rescue a sample that small, and no amount of statistical sophistication changes it.

If the trade count is under about fifty, read the rest of the report as a sanity check on your rules — does it trade when you expect, does it exit where you expect — and not as evidence about profitability at all.

Then expectancy, not win rate

Win rate on its own tells you almost nothing, because it says nothing about size. A strategy that wins 80% of the time and loses five times as much when it is wrong is a losing strategy. One that wins 35% of the time with wins four times the size of losses is a good one.

The number that combines them is expectancy — the average rupees per trade:

Expectancy = (hit rate × average win) − (loss rate × average loss)

Positive expectancy is the minimum bar. It is also where most of the useful thinking happens, because it shows you the trade-off between the two things you can actually influence.

1× 2× 3× 4× 20% 40% 50% 60% 80% hit rate above the line — profitable below the line — losing 50% needs 1× 40% needs 1.5×
Every break-even combination of hit rate and payoff. The amber line is where expectancy is exactly zero: at a 50% hit rate you need wins the same size as losses; at 40% you need 1.5×; at 30% you need 2.33×. Anywhere above the line makes money before costs. The practical use is diagnostic — if your strategy sits just above the line, it has no room for the costs in the next module, and a small deterioration in either number puts it underwater.

A useful habit: mark your strategy's position on that curve and ask how far it is from the line. A strategy at 52% with a 1.05× payoff is technically profitable and practically not, because a single bad month moves it across. A strategy at 45% with a 2× payoff has room.

Then the drawdown, in two numbers

Maximum drawdown is the largest peak-to-trough fall in the equity curve. It is the number that tells you whether you could have held the strategy, which matters more than whether it made money, because a strategy you abandon at the bottom returns nothing regardless of what the curve eventually did.

Read it alongside a second number almost no one looks at: how long the drawdown lasted.

peak max drawdown the depth you had to sit through time under water — the number that actually ends careers
Depth and duration are different questions. A 14% drawdown that recovers in nine days is an incident. The same 14% taking seven months to recover is a different experience entirely: seven months of doing everything right and having nothing to show, which is how most traders talk themselves out of a working strategy. Both numbers are on the report. Only one of them is usually quoted.

Ask yourself the honest question in rupees rather than percentages. Not "could I tolerate a 22% drawdown" but "if my account is twelve lakh, could I watch two lakh sixty thousand go, over five months, without changing anything?" Most people answer the percentage question yes and the rupee question no.

Two related figures are worth finding: the worst single day, which is your real tail-risk estimate, and the longest run of consecutive losses, which is what tests your conviction. Eleven losers in a row is survivable arithmetic and brutal psychology.

What the report is quietly assuming

Every backtest report is the output of assumptions, and a report that does not show you its assumptions is less useful than one that does.

Three to check every single time:

Are costs applied, or only estimated? In this platform's backtester, transaction costs are off by default and are charged per trade rather than per day — so a rule set that stops out and re-enters three times on a four-leg structure pays three round turns, which is twenty-four orders of brokerage, STT and GST against a day the gross curve may show as a small win. Even with the deduction switched off, the bill is still computed and displayed beside the result, so you can always see what a round turn would have cost. Read that figure. The next module is entirely about why.

What slippage was assumed? A run with zero slippage is a run that assumes you always got the price you asked for. The platform applies slippage as a percentage of the fill, against you on both sides of every leg — and applying it to the fill rather than to the bottom line matters more than it sounds, because a slipped fill changes where your stop and target actually sit, so a slipped run exits at different minutes, not merely with a smaller number.

Where does the curve start? A cumulative P&L curve should lead from zero, so the first trade's contribution is visible as a step rather than hidden in the baseline. A curve that starts at its first trade's value flatters the early period and makes the first drawdown invisible.

Illustrative

Two reports for the same rules over the same two years.

Report A — gross, no slippage: 312 trades, 61% winners, net ₹4,18,000, max drawdown ₹46,000.

Report B — costs applied, slippage 0.15%: 298 trades, 57% winners, net ₹1,02,000, max drawdown ₹71,000.

Everything moved, and all of it moved the same way. The trade count fell because slipped fills tripped stops that the clean run never reached. The hit rate fell because marginal winners became marginal losers. The profit fell by three quarters. The drawdown grew by half.

Report B is not a pessimistic version of Report A. It is the same strategy with two assumptions corrected, and it is the only one of the two that describes something you could have earned. A strategy that survives Report B is worth forward testing. One that only exists in Report A was never a strategy — it was a description of a price series with the frictions deleted.

The numbers to distrust

Total return. Entirely determined by how much you chose to risk. Doubling your size doubles it and tells you nothing new.

Win rate quoted alone. Meaningless without the payoff ratio, and trivially inflated by moving the target closer and the stop further away.

Sharpe ratio on a short sample. A legitimate measure that needs far more data than most retail backtests contain. On 40 trades it is noise with a Greek letter attached.

The best month. Interesting, never relevant. Nobody has ever been ruined by their best month.

What to take from this chapter

  • Read the trade count first. It sets the weight every other number can carry; under about fifty, read the report as a sanity check on your rules and nothing more.
  • Use expectancy, not win rate. Win rate without payoff ratio is not a result.
  • Mark your position against the break-even curve and ask how much room you have. Just above the line is not profitable once costs arrive.
  • Read drawdown as depth and duration, and convert both to rupees before deciding you could have held it.
  • Check whether costs were applied or only estimated, what slippage was assumed, and whether the curve leads from zero.
  • Total return, bare win rate, Sharpe on a short sample and the best month are the four numbers most likely to flatter you.