Your EA Does Not Know What Time It Is
Key takeaways
- The hours an EA is allowed to trade change your cost per trade more reliably than most parameter tweaks, and the robot cannot notice the difference on its own.
- The London and New York overlap is valuable mainly because execution gets cheaper and more predictable there, not because the market gets easier.
- Gold concentrates most of its daily range in the liquid hours and shows its worst spreads outside them, so a 24 hour gold bot pays the most for the least.
- Rollover, Friday afternoon and the Sunday night open are three windows where spread widening, not price, is what reaches your stop.
- Session filters and news filters solve different problems and belong on top of each other, because high impact releases land inside the best hours.
- Do not choose a window from an article. Split your own trade log by server hour and read the net result after costs.
The hours are the one setting nobody audits
Ask a trader why an expert advisor underperformed and you will hear about the strategy, the broker, the market regime. You will almost never hear about the clock. That gap is expensive. Choosing the best trading sessions for an EA is not a cosmetic preference, it is one of the few inputs that changes what every single trade costs you, and a robot has no way of noticing that it is being charged three times more at 3am than at 3pm.
An EA sees price, the spread at the instant of execution, and whatever conditions you coded. It does not see context. It cannot tell that the last hour was thin because Tokyo had closed and London had not opened yet, or that the candle it just traded was a rollover artifact. It fills the order, pays the cost, and moves to the next signal. Across a thirty day evaluation those unnoticed costs compound into the distance between a profit target reached and one missed.
On a prop account that distance is not academic. You are trading someone else's rules with your own fee at stake, and the arithmetic has no room for a slow leak. Before you touch entry logic, look at the hours.
Best trading sessions for an EA: what Asia, London and New York actually do
Forex runs continuously from Sunday evening to Friday evening, but liquidity does not. It arrives and leaves in three waves, and each wave has a personality that shows up directly in your spread and in your fills.
- Asia, roughly 00:00 to 09:00 UTC and centred on Tokyo, is the quietest of the three for most majors. Ranges are tighter, spreads sit above the daily best, and moves that look like breakouts often fade. A breakout EA in these hours frequently pays full cost for half a move.
- London, roughly 07:00 to 16:00 UTC, is the largest volume block of the day for European pairs. Spreads tighten, depth improves, and the session tends to set the direction the rest of the day argues with.
- New York, roughly 12:00 to 21:00 UTC, is when US data lands and when dollar pairs and gold get their sharpest expansions. The second half thins out noticeably as London desks go home, and the last hours look more like Asia than like the open.
- Your broker's clock: those ranges shift when daylight saving changes in Europe and the United States, and they will not match your broker's server time, which runs on its own offset. Missing that offset by two hours is a common reason a session filter behaves nothing like it did in testing, because the window you thought covered the London open actually covers the last hour of Asia.
- Whatever window you settle on belongs in the parameters that travel with the robot rather than in your intentions, which is exactly what a [set file is for and why it matters more than most traders think](/blog/setfiles-explained). One window, written down, applied identically on every account you run.
The overlap is a cost window before it is an opportunity window
The London and New York overlap, roughly 12:00 to 16:00 UTC, is when the two deepest pools of liquidity are open at the same time. Most articles sell it as the window with the most opportunity. For an automated account that framing is backwards.
More participants means tighter spreads and, usually, less slippage on a market order, because there are more resting orders near the touch. That cost advantage is reasonably reliable. The opportunity advantage is not. More movement also means more failed breakouts, faster reversals and more stops getting run. Volatility is symmetric, it can pay you or take from you. Cost only ever subtracts.
So the honest reason to concentrate trading in the overlap is that execution gets cheaper and more predictable, not that the market gets kinder. If your strategy has an edge, the overlap lets you keep more of it. If it does not, the overlap just lets you lose faster. That cost side deserves study on its own, because [slippage and execution quality decide more prop firm outcomes than the entry signal does](/blog/slippage-execution-prop-firm-ea).
Gold is the clearest case of all
Gold makes this argument better than any currency pair, and it happens to be the instrument our own validated configuration trades, so it is the one we have looked at hardest.
Gold spends long stretches of the Asian session doing comparatively little, then produces most of its daily range between the London open and the first hours of New York. The spread follows the same shape in reverse: worst when almost nobody is trading it, tightest when everybody is. A robot that trades gold around the clock is therefore paying the widest spreads of the day in exchange for the smallest moves of the day.
The trap is that gold's daily range is large enough that a wide spread looks small in percentage terms. On a challenge account with a fixed profit target and a fixed drawdown, cost per trade is not a percentage, it is a subtraction from a budget you were given. Before you set a schedule, read what changes when the instrument is handed to [a gold trading bot on a prop account](/blog/gold-trading-bot-prop-firm) that trades it while you sleep.
Rollover, Friday afternoon, and Sunday night
Once a day, most brokers roll open positions to the next value date. Around that moment, commonly 17:00 New York time, liquidity providers step back for a short window. Spreads widen, sometimes dramatically, quotes can gap, and the candle printed on your chart may represent a price almost nobody traded in size.
A human sees that and waits. A robot running 24 hours a day sees a price. If a spike can trigger your entry logic, or your stop sits inside the range the spread briefly opens to, rollover can put you in a position you never wanted or take you out of one you were still in. It is the most avoidable technical loss in automated prop trading, and the fix is one setting: do not trade through it.
- Rollover: block entries for a few minutes on either side, and understand that a stop can be reached by spread widening rather than by price moving.
- Friday afternoon: the last New York hours of the week thin out, and anything left open runs into gap risk over the weekend. [Weekend holding and swap treatment on a prop account](/blog/weekend-holding-swap-prop-account) is commonly structured differently from firm to firm, and some do not allow it at all, so confirm with yours in writing.
- Sunday night: the first minutes after the week opens carry the widest spreads you will see for the next five days, on almost every instrument. There is no strategy reason to be first in line.
What running 24 hours a day actually costs
Numbers make the point faster than argument. Everything below is an invented worked example. It is not a result, not a projection, and not a claim about any broker or any firm.
Say a strategy takes four trades a day on gold. Inside the liquid hours the round trip cost is 25 dollars per trade in spread. Outside them, on the same size, it is 55 dollars. Left running around the clock, the robot takes two trades inside the good window and two outside, so the daily cost is 2 times 25 plus 2 times 55, which is 160 dollars. Restrict it to the liquid window and the daily cost falls to 50 dollars, but you give up whatever the two removed trades were worth.
The real question is whether those two extra trades earned more than 110 dollars a day between them. Over a twenty day evaluation that is 2,200 dollars of extra cost needing justification. When traders run this split on their own history, the dead hour trades are usually not the ones carrying the expectancy, they are the ones diluting it. Do not take that on faith, run the split yourself. The numbers above were invented to show the shape of the calculation.
News windows sit on top of session windows
Session filters and news filters solve different problems, and a serious setup wants both. A session filter says which part of the day your costs are acceptable. A news filter says which minutes are untradeable regardless of session, because a high impact release pulls liquidity out of the book in exactly the hours that are otherwise the best available.
The two are layered, not in conflict. You select the overlap because execution is cheap there, then pause inside the release because for those minutes it stops being cheap. A robot that respects the first rule and ignores the second collects its worst fills of the month inside its best window of the day.
Prop firms handle this in their own way and the rules genuinely differ: some restrict opening or closing positions inside a window around high impact news, some do not, some apply it only to certain account types. Read how [news trading rules at prop firms are commonly structured](/blog/news-trading-prop-firm-rules), then confirm the current wording on your firm's own site, because this is one of the rules that gets revised.
How to pick your window, and why it still matters after you pass
You do not need a theory about sessions, you need your own trade log split by hour, which takes an afternoon and pays off only if you know [what each number on that log actually answers](/blog/prop-firm-account-metrics-explained).
PraxAI includes a news filter that pauses around high impact events, and the validated gold configuration takes one position at a time with no martingale and no grid, because concentrating into liquid hours only helps if position count stays sane. If you are still comparing tools rather than tuning one, the wider survey of [AI trading bots for prop firms in 2026](/blog/best-ai-trading-bot-prop-firms-2026) is the place to start.
The clock does not stop mattering once the evaluation is behind you. Funded accounts are lost to the same slow costs and the same 3am fills that sink challenge accounts, usually more quietly, which is a large part of why [funded traders lose the account](/blog/why-funded-traders-lose-the-account) after clearing the hard part.
- Tag every closed trade with the server hour it opened, group by hour, and read net result after costs rather than gross profit.
- Compare the average spread you actually paid inside and outside your candidate window, on the same instrument and the same size.
- Count the trades the window would have removed and add up what they contributed. If the answer is close to zero, the window is free money.
- Re-check the window after each daylight saving change, because it silently moves relative to the market while your server clock stays put.
Frequently asked questions
What are the best trading sessions for an EA on a prop firm account?
For most instruments the London session and the London to New York overlap, roughly 07:00 to 16:00 UTC, give the tightest spreads and the most predictable fills, which is what an automated account needs most. That is a starting point, not an answer. The right window is the one your own trade log supports once you group results by server hour and read them net of costs. Verify the offset between your broker's server time and UTC before you set anything.
Should I turn my EA off completely outside the overlap?
Not necessarily, and the decision is measurable rather than philosophical. Split your history into trades taken inside the window and trades taken outside, then compare net result after spread and commission. If the outside trades add expectancy, keep them. If they only add cost and exposure, the filter pays for itself. Strategies built specifically for Asian session ranges are a real exception, so test rather than assume.
Does restricting my robot to certain hours break any prop firm rule?
Trading fewer hours is not usually a problem in itself, but two commonly structured rules interact with it. Minimum trading day requirements need the account to trade on a set number of separate days, and a narrow window can make that harder if the strategy is selective. Consistency rules can also react to results being concentrated. Rules vary by firm, by account type and over time, so confirm the current version on your firm's own site.
Why does my EA behave differently live than in testing when I use a session filter?
The most common cause is the server time offset. Historical data and your live broker may sit on different clocks, and daylight saving in Europe and the United States shifts the relationship twice a year, so a window set once quietly points at the wrong hours later. The second cause is that spread in testing is often modelled as fixed, which hides exactly the cost difference between sessions that the filter exists to avoid.
What is rollover and why does it matter more for a robot than for a person?
Rollover is the daily moment when brokers roll open positions to the next value date, commonly around 17:00 New York time, and liquidity thins for a short window while spreads widen. A person sees the widened quote and waits. A robot sees a tradable price and acts, which is how spread widening rather than real price movement ends up triggering entries and reaching stops. Blocking a few minutes on either side removes the problem.
Want the bot that runs this discipline for you?