Do Streaks Exist? Daily Continuation Across 30 Futures and 117,560 Sessions

The line everyone repeats and almost nobody has counted

"It's been up three days, it's due for a pullback." That is probably the most widespread idea in short-term analysis, and it has one advantage over most of the others: it is easy to check. Count how often, after a run of N days in the same direction, the next day carried on in that direction.

Here is that count across thirty futures contracts from seven families: the five index futures, five metals, two energies, three grains, eight currencies, five rates contracts and two cryptocurrencies. 117,560 classified days, most of the series starting in 2010.

The result in one line: after three or more days in the same direction, the set continues 48.9% of the time. After two days, 48.7%. After one, 49.0%.

So yes, there is a very slight tendency to reverse, and it is exactly the same after one day as after three. Three is not a threshold. It is a number that sounds good when you say it out loud.

How it is measured, and why 50% is the benchmark

Each day's direction is measured from its own open to its own close, not against the previous close. A three-day streak is three consecutive days closing in the same direction. The percentage in the table is how often the next day also closed in that direction instead of turning around.

The benchmark is not a number anyone picked: it is 50%. Above it the market carries on more often than it turns; below it, the other way round. The belief being tested predicts clearly below 50% in the three-or-more column, and well below, or it would be of no use to anyone.

One detail shaves a fraction off every column equally: a day that closes exactly at its open counts as not continuing. There are few of them, but they explain part of why nearly every figure lands just under 50 rather than just over.

The series do not all start together. Most begin in June 2010; the RTY in 2017, SOFR in 2018, bitcoin in 2017 and ether in 2021. Their samples are smaller because of it, and the days column is in the table precisely so it can be discounted while reading.

Seven families, and not one leaves the band

Grains 49.9%, index futures 49.6%, rates 49.4%, currencies 48.9%, metals 48.7%, crypto 48.1% and energy 45.6%, all in the three-or-more column.

Six of the seven families fit inside two points. The seventh, energy, is the only one that pulls away, and it is just two contracts: crude oil and natural gas. At 8,440 days it is the smallest sample in the study alongside crypto.

What does not show up anywhere is a family where streaks work. Not even the cryptocurrencies, the market where momentum gets talked about most: 48.1% after three days, across 3,705 days.

That is what makes this hard to wave away. If the effect appeared in one family and not another, the measurement would be worth arguing about. Getting the same answer in grains, in Treasury futures and in bitcoin points at there being nothing to measure.

Contract by contract: 21 of 30 below 50%

Twenty-one of the thirty contracts continue less than half the time after a run of three days or more. Nine are above. That split alone says the tendency to turn is real but small: if it did not exist, the split would be fifteen and fifteen.

The full range runs from natural gas at 45.3% to SOFR at 55.6%. Leave those two out and everything sits between the Russell's 46.0% and the Dow's 51.9%.

Ten points from top to bottom looks like a lot until you read the days column. SOFR has 2,149 measured days and the Russell 2,388, against more than 4,200 for the contracts with a full series. The ends of the table are, as usual, the contracts with the least sample.

The other two columns behave the same way. After one day, 46.0% in bitcoin to 51.6% in the pound. After two, 45.6% in copper to 51.5% in the Dow. No contract sits outside the band consistently across all three columns, which is what it would take to call it a behaviour of its own.

The two exceptions, and why they are no use either

Three-month SOFR posts 55.6% after three days or more, the only figure clearly above 50 in the whole study. It is a short-term interest rate contract whose price tracks expectations about central bank policy, and those expectations move in steps that last weeks. Stringing runs together makes sense. It also has the shortest series here, from 2018.

Natural gas sits at the other end with 45.3%, across 4,220 days. It is the most volatile of the thirty and the one that collects the most weather headlines. That it turns more often than the rest is plausible; that four and a half points off a coin flip is tradable is another matter.

And here is the deeper problem with both exceptions: to trade them you would have had to pick them in advance. Across thirty contracts, two landing four or five points from the centre is what chance produces. Pointing at them after seeing the table is picking the winner with the race already run.

What this study does not say

It says nothing about size. A day that continues by a tick and one that continues by 3% count the same. It is entirely possible that continuations are larger than reversals, or the reverse; this measurement cannot see it.

It says nothing about what happens inside the day. A day that opens against the streak, travels 2% and closes in the streak's direction counts as a continuation, however untradable it was.

It does not separate runs longer than three days. The last column groups three, four, seven and fifteen days into one bucket, and there could be differences inside it. That is the cut the report provides.

And it does not say momentum does not exist. It says that, measured this way, daily and open to close, it does not show up. Other timeframes and other definitions are a different study.

What to do with this when you plan the session

First, stop using the streak as an argument. Neither for nor against: three days up make neither the fourth nor the reversal any more likely, in none of the thirty contracts measured.

Second, be suspicious of any rule with a round number inside it. Three days, five candles, two touches. If the threshold were real it would show in the table as a step between the two-day column and the three-day one. There is no step: there is 48.7% and 48.9%.

Third, and this is the only actionable part: if a strategy depends on the market turning after a run, this study caps it. The most edge available, in the best contract and before costs, is about four points on a coin flip. Any commission and any spread eats that.

For a single contract, each instrument page carries its own streak table next to range, gap and inside day.

Frequently asked questions

After three days up, is the market more likely to fall?

Slightly, and only slightly. Across the 30 contracts measured, after a run of three days or more the market continues in that direction 48.9% of the time, across 117,560 days. That is 1.1 points below a coin flip, and the same deviation shows up after a single-day run (49.0%).

Is there any market where streaks do work?

Three-month SOFR continues 55.6% of the time after three days or more, the only figure clearly above 50 in the study, and it has the shortest series (2,149 days, from 2018). Natural gas marks the other end at 45.3%. With thirty contracts on the table, two values that far from centre is what chance produces.

How is a streak defined here?

Consecutive days closing in the same direction, with each day measured from its own open to its own close. The three-or-more bucket includes longer runs. A day that closes exactly at its open counts as not continuing.

Do cryptocurrencies behave differently?

Not in this measurement. Bitcoin and ether together give 48.1% continuation after three days or more, across 3,705 days, essentially the same as index futures (49.6%) or metals (48.7%). The market where momentum gets discussed most does not show it on a daily scale.

So momentum does not exist?

That is not what this study says. It says that, measured in days and open to close, yesterday's direction carries no information about today's in any of the 30 contracts. Momentum on other timeframes, or with other definitions of direction, is a different question this table does not answer.

These numbers, instrument by instrument

• es
• nq
• rty
• ym
• nkd
• cl
• gc
• 6e