Spread series: the dataset nobody studies
Traders spend hours on price and almost none on spread. Yet the spread is a time series with structure of its own.
A week of recorded spreads for a pair shows a daily rhythm: tight during the busy sessions, wide in the dead zones, with spikes at rollover and around news. It also shows how quickly it recovers after a spike and how often it makes short unexplained jumps.
Useful statistics include the median, the 95th percentile, and the duration of wide spells. The median says what trading usually costs. The 95th percentile says what it costs when things go wrong.
Strategies are often tested with the median or a fixed spread, missing the fat tail. The trades that matter most for the bottom line, entries at the worst moments, occur in the tail.
Recording the spread at every tick and storing it with the price makes tests far more realistic. Some brokers provide this history, and a platform's own recording is also fine.
An easy chart: median spread by minute of the day, with the 5th and 95th percentile bands. It takes a few lines of code and makes the structure of costs visible at a glance. Many traders keep it pinned above the desk after seeing it once.
What is the 95th percentile spread on the main pair during the hours traded?
Spread series: the dataset nobody studies
Spread series: the dataset nobody studies
It’s Fairman 
-
FXS Prop Desk
- Posts: 1689
- Joined: Wed Oct 07, 2026 9:52 am
Re: Spread series: the dataset nobody studies
And the spread profile is also the cheapest risk tool available. Knowing that a pair's spread usually triples at a given hour lets a trader avoid placing tight stops then, without needing any other indicator.
FXS Official Team | The Idea Lab & Strategy
-
Amanhasnoname
- Posts: 2060
- Joined: Fri Oct 30, 2026 9:45 am
Re: Spread series: the dataset nobody studies
I stopped trusting the average spread after comparing my EUR/USD fills with a fixed-spread backtest. During the London-morning window I actually trade, my recorded 95th-percentile spread was about 1.9 pips, although most quotes were far tighter. The wide readings arrived in short bursts near rollover and scheduled releases, so the average hid precisely the conditions that hurt tight-stop entries.
I logged bid and ask changes from the platform, kept the sample limited to my broker and account type, then grouped observations by time of day. I also marked how long each wide spell lasted; a single spike is different from a spread that stays elevated for several minutes.
For a useful comparison, report the median beside the 95th percentile and note the session and news windows. I use the upper tail to set a no-trade threshold and to avoid judging a setup from a backtest that assumes the normal spread during an abnormal minute. Those values will shift across brokers, so the trader's own feed is what matters.
I logged bid and ask changes from the platform, kept the sample limited to my broker and account type, then grouped observations by time of day. I also marked how long each wide spell lasted; a single spike is different from a spread that stays elevated for several minutes.
For a useful comparison, report the median beside the 95th percentile and note the session and news windows. I use the upper tail to set a no-trade threshold and to avoid judging a setup from a backtest that assumes the normal spread during an abnormal minute. Those values will shift across brokers, so the trader's own feed is what matters.