The Backtest Problem Most Traders Ignore: Start-Date Sensitivity

A trading system can look exceptional in a backtest and still tell you surprisingly little about what would have happened to somebody who actually started using it on a different day.

That is one of the biggest weaknesses in the way trading systems are often presented.

A conventional backtest normally has one beginning, one ending and one resulting equity curve. It might show an impressive return, an acceptable drawdown and a healthy profit factor.

But a prop firm trader does not get to choose the perfect point in the historical sequence.

They start today.

And that changes the question completely.

The problem with one perfect starting point

Suppose a trading system is tested from January 2025 to July 2026.

The final result may look excellent. But someone starting the system in January 2025 experiences a completely different sequence of trades from somebody starting in March 2026.

The strategy itself has not changed.

The market sequence has.

This creates what I think is one of the most important considerations in prop-firm system testing:

start-date sensitivity.

For an evaluation account, the objective is not simply to be profitable eventually. The path matters.

A system can have a strong long-term expectancy and still fail an evaluation if a cluster of losses arrives before the profit target is reached.

That is why drawdown and sequence risk can be more important than the final number at the end of a backtest. Our own Low Risk analysis makes exactly this distinction: a profitable long-run system can still fail an evaluation if losses cluster before the target is reached.   

A different way of testing

Rather than asking:

“What happens if the strategy starts once?”

we wanted to ask:

“What happens if the exact same strategy begins at many different points in the same historical period?”

For the Moreland models, we created fresh $100,000 evaluation-style simulations beginning at the first recorded trade of each calendar month from January 2025 through July 2026.

That produced 19 separate starting points.

The historical return path was then replayed from each individual start date toward an illustrative +10% evaluation target.   

The important point is that the trading logic does not get altered to suit each starting date.

Only the point at which the trader enters the historical sequence changes.

And the results show why that matters.

The same model can produce very different journeys

With the Moreland Low Risk Model, all 19 rolling monthly simulations eventually reached the illustrative +10% target during the available historical window.

But the time required was far from consistent.

The average was 53.7 days, while the median was 39.8 days. The fastest historical simulation reached the target in 15.7 days, while the slowest required 122.3 days.   

That is an enormous range.

If I simply advertised the average and said:

“Expected completion time: 53 days”

I would be hiding a large part of the information that actually matters.

The distribution tells the more useful story.

The system encountered periods where progress was rapid, periods where progress was slow, and periods where a trader would have needed considerable patience before reaching the same target.

Increasing risk changes more than speed

The Medium Risk Model provides an even clearer example of why start-date testing matters.

Using the same rolling-start concept, 18 of the 19 historical starts reached +10% before the illustrative closed-balance failure thresholds used in the analysis.

One did not.

The unsuccessful start was March 2026.   

Among the successful simulations, average completion time fell to 30.1 days, with a median of 20.3 days. The fastest reached the target in just 2.4 days, while the slowest successful start required 77.9 days.    

That demonstrates the trade-off very clearly.

Higher risk historically produced faster successful progress.

But it also produced substantially greater drawdown and introduced a historical start point that failed before reaching the target.

The Medium Risk configuration recorded a 16.47% maximum closed-balance drawdown and 18.85% maximum equity drawdown, compared with 7.14% and 8.51% respectively for Low Risk.   

That is much more informative than simply comparing the final returns of the two systems.

This is really a sequence-risk problem

A trading system is not experienced as an average.

It is experienced as a sequence.

Win.

Loss

Loss.

Win.

Drawdown.

Recovery.

Another loss.

Another recovery.

The order matters enormously when an account has fixed loss limits and a fixed target.

Imagine two traders using exactly the same strategy.

One begins immediately before a favourable market regime.

The other begins immediately before the system's worst historical drawdown.

Over a sufficiently long period, both might be exposed to the same underlying statistical edge.

But their experience of a prop-firm evaluation could be completely different.

That is path dependency.

And it is one reason I think terminal profit is often one of the least interesting numbers in a prop-firm backtest.

A high win rate is not the answer either

Another common mistake is assuming that robustness means winning most of the time.

It does not.

The Low Risk Model recorded a historical win rate of 41.7%, while its average winner was approximately 2.32 times the size of its average loser. Its historical edge came from payoff asymmetry rather than an unusually high hit rate.  

The Medium Risk Model was similar: a 41.3% historical win rate with an average winner approximately 2.28 times the magnitude of the average loser.   

That matters psychologically as well as mathematically.

A system with a positive historical expectancy can still produce long losing sequences.

If your testing methodology assumes that a losing streak means the strategy has “stopped working”, you may misunderstand the behaviour of the system you are actually trading.

Robustness is not the same thing as certainty

There is an important distinction here.

Testing many starting points does not prove that a system will behave the same way in the future.

It simply asks a harder historical question.

Instead of:

“Did this strategy work over this backtest?”

we ask:

“How dependent was the outcome on when somebody happened to start?”

That is a much more useful question.

But it still has limitations.

For example, the rolling evaluation analysis in these reports is based on closed trade results rather than a continuous mark-to-market equity ledger. That means closed-balance target progress can be studied, but every intratrade floating-equity rule used by every prop firm cannot be perfectly reconstructed.    

Historical simulations are not live verified performance, forecasts or guarantees.

And that is exactly why I prefer showing the distribution rather than pretending there is one definitive result.

What I look for in a serious system test

When assessing a strategy intended for evaluation accounts, I am less interested in finding the prettiest equity curve and more interested in questions like:

• What happens if the trader starts during an unfavourable period?

• How long can target completion realistically take?

• How deep are the historical drawdowns?

• How badly can losses cluster?

• Does the strategy rely on an unusually favourable starting point?

• What happens when the same methodology is tested across multiple market regimes?

• How much additional drawdown is being accepted in exchange for faster progression?

  
  

Those questions tell me considerably more about a system than one headline return percentage.

The takeaway

A backtest beginning on one carefully selected date can show you whether a strategy performed well through one historical sequence.

It does not tell you how sensitive that outcome was to the starting point.

For prop-firm evaluations in particular, that distinction matters because the journey to the target matters just as much as whether the strategy was profitable eventually.

That is why we use rolling start-date simulations as part of Moreland's testing process.

Not because 19 starting points can predict the future.

They cannot.

But because a system that is only convincing when you begin the backtest on exactly the right date is probably telling you less than you think.