A Profitable Backtest Isn’t Enough: 5 Checks Before You Trust It

Key takeaways

Correct arithmetic is a starting point. Without written rules, dated contracts, source data and entry/exit records, we cannot reproduce the trades.

An invented $5 allowance per completed trade reduces $1,000 gross to $900. That assumed calculation is not verified executable profit; fees, modeled fills and unresolved candle sequences still need examination.

Three trades contribute $900 before costs. Temporarily setting them aside leaves $100 gross across seventeen trades, or $15 after the same allowance. This diagnoses concentration; the full record stays intact.

Ask how the rules were selected and what observations informed that choice. Repeating a calculation on the same data adds no new evaluation period.

Match the next decision to the evidence: obtain the missing materials, attempt reproduction, then specify a further test and its challenge conditions. A positive report does not establish a trading method.

Full transcript

Open the fictional record

A trading report shows twenty trades and a thousand dollars in profit.

Twelve winners. Eight losers. A positive result.

Would you trust the strategy behind it?

Before answering, let’s examine the report. We’re going to open the trade record, check what the calculation includes, and look at what we would still need to know before taking its result further.

This entire example is invented for teaching. It is not my research, an actual trading record, or a forecast.

Here is what our fictional record contains. Nine trades made one hundred dollars each. Three made three hundred dollars each. Eight lost one hundred dollars each.

That gives us eighteen hundred dollars in winning trades, minus eight hundred dollars in losing trades. The total is one thousand dollars before costs.

The arithmetic is correct. Now we need to examine what sits behind it.

Check 1: Can we reproduce the trades?

Our first check is whether the record can be traced back to clear rules and usable data.

The report says, “Buy the opening-range breakout.” That sounds like a rule, but it leaves questions unanswered.

Which market and dated contract? What time zone? When does the opening range begin and end? Does touching its high qualify, or must price trade above it? Do we enter immediately, wait for a candle to close, or use the next candle’s opening price?

Where does the stop go? What ends the trade? Can another entry occur that day?

Those choices determine what we are testing.

For this fictional report, we have been given the trade amounts, but no written entry rules, dated contracts, or source bars. We can check its arithmetic. We cannot reproduce its trades.

That is a specific missing piece. Before accepting the record as a backtest we can inspect, we would ask for the rule version, data source, date range, contract mapping, and the entries and exits behind each row.

Check 2: Costs and execution

Our second check is what happened between the modeled prices and the reported profit.

The thousand-dollar total does not include costs.

For a simple illustration, suppose we deduct five dollars per completed trade. That is an invented combined allowance for this exercise, not a broker quote or a realistic estimate for every market and order.

Twenty trades multiplied by five dollars gives us one hundred dollars. Our report now shows nine hundred dollars after that allowance.

It is still positive. Including an assumed cost does not automatically make the result trustworthy.

We would need to know which commissions and fees apply, how entries and exits were modeled, and whether the assumed prices could reasonably have been obtained. We should also check whether any execution adjustment is already included, so we do not deduct it twice.

Imagine one minute’s candle reaches both the stop and the target. Its opening, high, low and closing prices alone may not tell us which level came first. Assigning the target without resolving that uncertainty could change the result.

In our fictional report, those details are missing. The nine hundred dollars is therefore the answer to our assumed calculation. It is not verified executable profit.

Check 3: Where did the profit come from?

Our third check is where the profit came from.

Remember the three trades that made three hundred dollars each? Together, they contributed nine hundred dollars before costs.

Let’s temporarily set them aside and examine the remaining seventeen trades.

The other nine winners made nine hundred dollars. The eight losers lost eight hundred. That leaves one hundred dollars before costs.

Applying the same five-dollar allowance to those seventeen trades deducts eighty-five dollars. The remainder is fifteen dollars.

That does not mean we should delete the biggest winners from the actual record. Some approaches depend on occasional large gains. Their presence alone does not invalidate a strategy.

This check tells us where to investigate next. Were those three trades legitimate applications of the same rules? Were their fills modeled consistently? Did they occur under a particular condition we need to understand?

The report’s result depends heavily on them. We should examine them closely rather than treating the total as evenly distributed evidence.

Check 4: How were the rules chosen?

Our fourth check is how the rules were chosen.

Suppose the creator tried many entry times, stop distances and targets, then showed us only this version because it finished ahead.

The report would leave out part of the experiment: the search that produced the selected result.

We would ask what alternatives were tested and which data were used to choose the rules. We would also ask whether the final version was written down before it was evaluated on additional observations.

For our fictional report, that development history is unavailable. We cannot tell whether these twenty trades evaluated a previously specified idea or helped select the idea being presented.

Running it again on the same observations would not answer that question. It would check repeatability of the calculation, not supply a new evaluation period.

Check 5: What decision does the evidence support?

Our fifth check is what decision the evidence supports now.

Here is our assessment of this example.

The supplied amounts add up to one thousand dollars before costs. An invented five-dollar allowance per trade reduces that to nine hundred dollars. Three trades account for most of the reported gross profit.

But the rules, source data, execution assumptions and development history remain missing.

We have enough information to inspect part of a report. We do not have enough to treat it as an established trading method.

The next step would be to obtain those missing materials and attempt to reproduce the record. If that work holds up, we could specify a further test, record the rule version in advance, and define which observations would challenge it.

For example, we could track how often an apparent signal cannot be executed as modeled, whether unresolved candle sequences change the result, and whether a cost estimate based on the intended setup materially changes the assessment.

That gives us questions we can investigate. It does not give us a promise of future profit.

Questions to take into the next report

When you next see an attractive backtest, ask to see the rules, the record, the assumptions, and how the test was developed.

Then ask what the evidence actually supports doing next.

That is how we move from looking at a result to examining the work behind it.

Revisit the concepts behind the checks

Use the worked example alongside the reference lessons on testing, interpreting results and specifying a plan.

Strategy Testing · Reading trading results · Build a Trading Plan

Next: Understand the CPI Release

Examine inflation figures, expectations and the questions to check before interpreting a market reaction.