Backtesting

Time-of-day filters in prop backtesting

Time-of-day filters change volatility, spread, fills, and drawdown clustering. A prop backtest must prove the clock helps.

The clock can make a strategy look smarter than it is. A time-of-day filter can remove dead hours, avoid ugly spreads, and keep risk out of the worst overlap. It can also be one more fitted parameter pretending to be discipline.

For a prop trader, the useful question is not "what is the best session to trade?" The useful question is whether the time filter improves expectancy and drawdown after realistic costs, without simply deleting the inconvenient trades that made the backtest honest.

What a time-of-day filter actually changes

A time-of-day filter tells the strategy when it is allowed to open, manage, or close trades. It might trade only during London, only during New York, avoid rollover, stop before the weekend, or skip a thin overnight window.

That sounds simple. In a backtest, it changes several moving parts at once:

Time filter issueWhat changesWhy a prop trader cares
VolatilitySome windows move more than othersStops and targets may behave differently by session
SpreadThin hours can widen execution costSmall edges can disappear when the bid/ask gap expands
Fill qualityFast windows can slip moreThe signal price and executable price can separate
Trade clusteringTrades may bunch into one sessionA bad window can pressure the daily loss rule
ExposureHolding across rollover or weekend changes riskSwap, gaps, and floating equity matter before the exit

Why session filters can help

Some strategies depend on activity. A breakout rule may need enough movement for price to escape the prior range. A mean-reversion rule may need enough two-way flow for overshoots to snap back. A spread-sensitive scalping idea may be unusable when the market is thin.

In those cases, a time filter can be legitimate. It can remove periods where the strategy has no reason to trade. It can also make the backtest closer to how the trader intends to run the system live.

But the filter has to earn its place. The test is not whether London looks better than Asia in one old sample. The test is whether the rule still works when the sample changes, costs are charged, and nearby time windows do not collapse.

That last point is where many session filters fail. If 08:00 to 10:00 works and 07:45 to 10:15 does not, the clock probably found a historical accident. Robust session logic usually has a reason and a band, not one magic minute.

Why time filters can also overfit

Time is a parameter. That means it can be abused like any other parameter.

The common workflow is dangerous: test every possible session, keep the best one, and call the result "market structure". That is still curve-fitting if the filter was chosen because it made the old equity curve prettier.

The warning signs are practical:

Warning signWhat it suggests
One exact start minute carries the resultThe strategy may be fitted to old trade timestamps
The filter removes most losing trades but has no market reasonThe rule is probably a hindsight delete button
The session works in-sample and fails out-of-sampleThe clock did not generalize
The result depends on zero or idealized costsThe session edge may be execution fantasy
The filter creates a few crowded risk windowsDaily drawdown pressure may increase, not decrease

This is the same family of problem as backtest overfitting for prop traders. A time filter can be real. It can also be the cleanest way to hide selection bias because every chart has a "best" hour after the fact.

The prop-firm problem: clustering by clock

Prop rules care about when losses hit, not just whether the month ends green. A session filter can accidentally concentrate risk into the same part of the day.

That matters if several symbols or strategies all wake up in the same window. The trade count may look diversified, but the account may still be making one time-of-day bet. If the New York open is bad for the whole book, losses can arrive together and pressure the daily limit before the long-run expectancy gets time to work.

This is why time filters should be read beside correlation and trade frequency. A filter that improves average trade quality but bunches losses into one session may be worse for a prop account than a slightly weaker but smoother distribution.

The related checks are why correlated trades fail prop accounts and trade frequency in prop backtesting. The clock can diversify risk, but it can also synchronize it.

What to test before trusting the clock

A useful time-of-day filter should pass more than one pretty equity curve.

Check these before trusting it:

CheckStronger signWeak sign
Market reasonThe window matches volatility, liquidity, or rule mechanicsThe window was picked only because it won the sweep
Stability bandNearby start and end times behave similarlyOne exact minute makes or breaks the test
Out-of-sampleThe filter works on data not used for selectionThe filter only works where it was discovered
Cost sensitivityResults survive real spread, commission, swap, and slippageThe edge disappears after execution costs
Daily loss pathLosses do not bunch into one dangerous windowBad sessions cluster losses near the rule floor
Broker portabilityThe logic survives different session data and costsThe backtest changes too much across brokers

The broker point is not cosmetic. Session candles, spread behavior, rollover treatment, and symbol trading hours can differ. A time filter tested on one broker's data may not behave the same way elsewhere. That is why the same cTrader backtest can change across brokers.

Entry filter, exit filter, or risk filter?

Not every clock rule does the same job. Be precise about what the filter controls.

Filter typeWhat it controlsRisk in the backtest
Entry filterWhen new trades may openCan remove trades without proving exits are better
Exit filterWhen positions must closeCan distort average win, loss, and holding time
Risk filterWhen size is reduced or trading pausesCan improve survival but reduce opportunity
Maintenance filterWhen rollover, weekend, or platform risk is avoidedCan protect execution while changing the strategy path

An entry-only filter is the easiest to flatter because it simply refuses trades the old sample did not like. An exit filter is more invasive because it changes the trade distribution. A risk filter may be useful for prop survival, but it must be tested as part of position sizing, not bolted on after the equity curve.

The backtest should say which one is being used. "We trade London" is not enough. Does the system open only in London? Close before London ends? Reduce size outside London? Hold existing positions through the next session? Those are different strategies.

Where realbacktesting draws the line

realbacktesting is a trading-software studio for cTrader: prop-firm cBots, indicators, and plugins built around verifiable backtests rather than vague session stories.

That matters because time filters are easy to dress up. The public methodology uses intrabar M1 execution, cTrader broker M1 bars + tick-measured spread from 2021-2026, real per-symbol spread, real commission, swap, 1 bps slippage, and an 80,000 EUR model base. The drawdown ceiling is the worst floating-equity low and is checked against a 30% out-of-sample hold-out. The method is laid out on the methodology page.

Those details do not prove that any session is special. They make the session claim harder to fake. If a time filter only works with ideal fills, no spread expansion, no swap, and no out-of-sample check, it has not proved enough for a prop account.

The funding side is the same argument under stricter rules. A session filter has to be read as account-survival math, not just trade selection. The broader model is explained on the funding page.

Frequently asked

Are time-of-day filters useful in backtesting?

They can be. A time filter is useful when it removes periods where the strategy has a clear reason not to trade and still survives out-of-sample with realistic costs. It is weak when it only selects the historical hours that looked best.

What is the best trading session for prop firms?

There is no universal best session. The right window depends on the strategy, instrument, broker costs, and prop rules. A breakout system, mean-reversion system, and carry-sensitive swing system can need different clocks.

Can a session filter reduce drawdown?

Yes, but it can also concentrate losses. A filter that removes bad hours may reduce drawdown. A filter that pushes every trade into the same volatile window may increase daily loss pressure.

Should I optimize start and end times?

Test them, but do not worship one exact timestamp. A stronger result holds across nearby start and end times and has a market reason beyond the backtest winner.

The stubborn takeaway

The clock is not an edge by itself. A time-of-day filter earns trust only when it improves the path after costs, out-of-sample, and under the rule pressure of the account.

Published Jul 15, 2026 · realbacktesting · Educational content and market commentary — not financial advice. Trading involves risk; past performance does not guarantee future results.