Everyone checks the spread in pips and almost nobody checks the only thing that matters: what share of the move you pay to get in. A 0.5-pip spread is cheap on EUR/USD at 14:00 UTC and expensive at 21:00, because at 21:00 the pair barely moves. We measured the real bid/ask spread of 13 markets across 67,661,092 one-minute bars of tick data and divided it by the average range of each hour. The ranking that comes out does not match any broker's pip table.
The number almost nobody publishes
The spread on a broker's website is a marketing figure: a "from", measured at the best moment of the day. These are something else: the median bid/ask spread measured tick by tick, hour by hour, over each market's full history. Median rather than mean on purpose — the daily rollover and weekends inject spikes of tens of points that describe a moment when nobody is trading.
Read the third column: real cost = median spread divided by that hour's average range. It says how much of the available move the spread takes before you are right about anything.
| Market | Median spread | Average real cost | Cheapest hour | Priciest hour |
|---|---|---|---|---|
| USDJPY USD/JPY | 0.400 | 2.8 % | 14:00 (1.1 %) | 21:00 (11.5 %) |
| EURUSD EUR/USD | 0.500 | 3.1 % | 14:00 (1.3 %) | 21:00 (9.3 %) |
| GBPUSD GBP/USD | 1.100 | 5.9 % | 14:00 (2.7 %) | 21:00 (14.9 %) |
| DJ Dow Jones | 3.310 | 7.0 % | 14:00 (1.8 %) | 04:00 (16.0 %) |
| SP500 S&P 500 | 0.525 | 7.1 % | 14:00 (2.8 %) | 04:00 (12.9 %) |
| NQ Nasdaq 100 | 2.202 | 8.3 % | 14:00 (1.9 %) | 04:00 (16.3 %) |
| AUDUSD AUD/USD | 1.000 | 8.9 % | 14:00 (5.1 %) | 21:00 (23.0 %) |
| XAUUSD Gold | 37.000 | 9.4 % | 13:00 (4.8 %) | 21:00 (19.8 %) |
| EURGBP EUR/GBP | 0.900 | 9.9 % | 08:00 (4.2 %) | 21:00 (23.5 %) |
| USDCHF USD/CHF | 1.000 | 10.4 % | 14:00 (4.0 %) | 21:00 (31.3 %) |
| XAGUSD Silver | 31.000 | 17.9 % | 14:00 (11.3 %) | 23:00 (24.6 %) |
| BTCUSD Bitcoin | 75.700 | 23.8 % | 14:00 (15.7 %) | 05:00 (30.3 %) |
| ETHUSD Ethereum | 5.500 | 27.9 % | 14:00 (19.4 %) | 05:00 (35.1 %) |
Median spread in each market's own quote points (not comparable across asset classes: one EUR/USD point is 0.0001, one Bitcoin point is a dollar). Real cost is comparable, because it is a percentage.
1. There is a 10x gap between the cheapest and priciest market
USDJPY takes an average 2.8% of the hour's range. ETHUSD takes 27.9%. Ten times more. And it is not that crypto moves little — it moves a lot — it is that its spread grows faster than its range.
The practical consequence is uncomfortable: an intraday strategy that works on USDJPY may not survive the same edge on Ethereum, even with an identical signal. Not because the signal fails, but because the toll is different.
2. 14:00 UTC is the cheapest hour in 11 of 13 markets
That is the London/New York overlap, and it is not only about liquidity: the hour's range grows more than the spread does. That combination is what drives real cost down, which is why the cheap hour is nearly the same for gold, the Nasdaq and Bitcoin.
The other end splits by asset. In forex the worst hour is 21:00 UTC — the rollover, with the market already empty — where USD/CHF costs 31.3% of the hour's move. For indices and crypto the worst window is the European small hours, 04:00-05:00 UTC, with Ethereum at 35.1%.
3. Forex got much cheaper; crypto did not
EUR/USD had a 2003 median of 1.0 points and sits at 0.3 in 2026 — more than halved in two decades, and the same holds across the majors. Careful reading that series in points when the price level changed scale: gold went from $340 to above $4,000, so its spread in points rose while, as a share of price, it fell sharply.
What these numbers are NOT
- Not what your broker will charge. This is an institutional feed. A retail broker adds its markup, and a market maker can double it. Use it as a floor, not an expectation.
- No commission, no slippage. Total cost is spread + commission + slippage. Only the first is here.
- The median hides the tail. The hourly tables also publish the 90th percentile, which is what you pay during news. That gap is often wider than the gap between brokers.
- Years without ask data are excluded. The feed carried no ask for index CFDs before 2013; including them would have published a zero spread for the Dow, which is false.
The full data
Every market has its hour-by-hour and year-by-year breakdown, as tables and CSV, at /stats. The raw M1 bars behind all of it are at /data.