Quarterly Theory Says the Day's Extreme Forms in Q2. It's the Least Likely Quarter.

Every couple of years the retail trading internet adopts a new framework for what price is secretly doing. Quarterly Theory is the current one. It comes from a trader called Jevaunie Daye, it sits downstream of the ICT world, and its core idea is that time is fractal: every unit of time divides into four quarters that always play the same four roles.

For a trading day the story goes like this. Q1 accumulates a range. Q2 manipulates, sweeping one side of that range to trigger stops. Q3 distributes, running in the opposite direction. Q4 continues or reverses. There are indicators for it in the LuxAlgo library, several on TradingView, and a dedicated strategy page on FX Replay.

Read a dozen of those pages, as we did, and you notice something missing. Not one of them publishes a sample size. No win rate, no trade count, no backtest window, no expectancy. The closest thing to evidence is a hedge: it works best in trending markets with institutional activity, and less well in low liquidity or heavily manipulated ranges.

So we measured it. Three FX majors, the three US indices the framework's audience actually trades, and gold. One minute data, 2019 to 2021, which is our in-sample window. We wrote no expert advisor and touched no out-of-sample data, because the framework makes claims you can test directly from raw prices. If the effect is not in the data, there is nothing to automate.

Pinning the rules down first

Frameworks like this are slippery to test because the rules move. We wrote ours down before running anything.

The quarters hang off a "True Day Open," and the sources disagree on whether that is 18:00 or 19:00 New York time. Rather than pick the one that flattered the result, we measured both and pre-registered that a disagreement between them would itself be the finding. A sweep means Q2 traded beyond the Q1 range and then closed back inside it, which is the stricter reading and the one the sources describe. Days where Q2 swept both sides, or neither, are skipped, since the framework gives no rule for them.

Then two claims, both falsifiable:

1. The clock claim. The day's high or low forms in Q2 more often than the 25% you would get if extremes fell wherever they liked.
2. The direction claim. After Q2 sweeps one side, Q3 runs the other way by enough to pay the spread and commission.

Claim one: the manipulation quarter is the quietest quarter

Share of daily highs and lows that form in Q2, by market, with 95% confidence intervals. The dashed line is the 25% you expect from chance.

Q2 holds the day's extreme 16.6% of the time across all seven markets, with a 95% confidence interval of 15.9% to 17.3%. Chance gives 25%. The interval sits entirely below chance, so this is not a near miss or a small sample wobbling around the null.

Pooled across the seven, the quarter the framework names as the one that manufactures the day's high or low is the quarter least likely to contain it. Every market sits below the 25% line. Of the fourteen cells we measured, across both True Day Opens, exactly one clears 25%, GBP/USD at the 19:00 anchor, and it does not survive the other anchor. One in fourteen at 95% confidence is what randomness produces.

Sitting under the null is not the same thing as being the quietest quarter, and the two come apart on two of the markets. On five of the seven, Q2 is also the lowest of the four quarters outright. On EUR/USD and GBP/USD the quietest quarter is Q1 instead, at 19.6% and 20.1%, and Q4 slips under Q2 as well, so Q2 ranks third of the four. It still lands under chance on both. On no market does it behave like the quarter that makes the day.

Where do the extremes actually form?

Share of daily highs and lows by quarter. Q2 lands below the 25% null on every market.

In Q3 on the currencies and gold, between 27.9% and 37.2%, and in Q4 on the indices, between 32.0% and 35.4%. In each case that is the stretch of the day when that particular market is busiest. Extremes cluster where the volume is, which is a fact about session hours rather than a hidden schedule.

Claim two: no direction that pays

The second claim is the tradeable one. Enter at the Q3 open against whichever side Q2 swept, stop just beyond the sweep, target twice the risk, flat at the end of Q3. That is how every source teaches it.

We report results in R, where 1R is the amount risked on the trade. Per trade dollar figures are useless for comparing an index against EUR/USD, and on a market like the Dow the trade to trade variation is so wide that a dollar average needs tens of thousands of trades before it means anything.

Net result per trade by market, in units of risk, with 95% confidence intervals.

Across 2,162 trades the setup loses 0.06R per trade, confidence interval 0.11R to 0.02R of loss. The interval is clear of zero, and no year of the three is positive. The 2R target, the one the sources put in every example chart, is reached 16.1% of the time.

Six of the seven markets lose. The exception is the Dow, at +0.10R, and we are not going to bury it in a footnote: on one of seven markets, at one of two anchors, this setup made money in sample. One cell in seven clearing is close to what chance produces, and it is not enough to carry a framework, but it is a real number and it is on the chart.

A losing net result can always be blamed on the tester rather than the idea, so we removed everything that could be our fault. No stop, no target, no commission, just the signed move from the Q3 open to the Q3 close in the direction the framework predicts. That gives 0.14R against the claim, with a confidence interval of 0.25R to 0.04R against, clear of zero.

That cost-free number is now the strongest evidence in the study. Whatever you think of our cost model, the raw move the framework predicts goes the wrong way.

We then spent a day trying to prove ourselves wrong

A negative result is worth very little if you stop at the first version that agrees with you. Three attacks, all of which we expected might work:

First, is the framework simply backwards? A negative fade result is a positive continuation result. If Q3 runs with the sweep instead of against it, the rules would be inverted but the clock would be real, and that is a genuine finding. This is the one number in the study that flickers: continuation shows 0.14R in its favour with an interval that clears zero. It still does not hold up. Against the threshold you need when running eight comparisons it falls short, the other True Day Open gives a weaker result, and traded as an actual 2R setup it loses 0.05R with an interval that contains zero.

Second, did we test the wrong timeframe? The framework is explicitly fractal, and a lot of the teaching happens at the 90 minute session quarters, which we had deliberately left out. So we ran the identical rules one scale down, on roughly four times the events. On 8,416 setups the raw drift is 0.0045R with a confidence interval straddling zero, the cleanest null in the whole study, and the traded version does worse than the daily version because you pay the spread four times as often. The fractal claim fails harder at the finer scale.

Third, did the data have a hole in it? On Mondays the Q1 range is built during Sunday hours, when our data feed quotes prices but the venue does not accept trades. Dropping every Monday moves the pooled result from 0.06R to 0.04R of loss. Nothing turns on it.

A correction, 2026-08-19

The first version of this post reported the traded loss as 0.23R per trade rather than 0.06R, and said every market was negative. Both came from a cost model that was wrong.

We were charging index CFDs a commission of 7 index points per trade, which came from applying our FX assumption of $3.50 per side per lot to instruments that do not work that way. IC Markets charges no commission on index CFDs on either account type; the cost is built entirely into the spread. Two other studies on this site already modelled that correctly. This one did not.

On the S&P 500, where the structural stops are tightest, that phantom commission was 81% of the risk on the trade. Removing it moves the S&P from 0.75R of loss to 0.02R, and moves the Dow from a small loss to a small profit.

We have also added gold, which belongs in the market set and was missing.

What this does not touch: claim one is a counting exercise with no costs in it, so 16.6% against a 25% null is unchanged in substance. Nor does it touch the cost-free drift, which is measured with no stop, no target and no commission at all. Both of those still say the same thing. What changed is the size of the traded loss, and one market's sign.

What this joins

This is now the fifth independent measurement on this site saying the same thing about intraday sweep and continuation structures at retail cost. The 9:30 opening range breakout does not continue on any of three indices. The ICT Silver Bullet loses 0.24R per trade across 2,200 setups on six markets, and its liquidity sweep, the component that makes it smart money rather than a gap fill, changes nothing measurable. The London failed breakout fade nets negative after a 1.6 pip toll. The pivot retest continuation is negative on all thirteen instruments we own.

The pattern across all of them is the same. These frameworks describe real things: ranges do get swept, stops do sit above highs, and price does reverse after taking them. What they do not do is tell you which way the next six hours go, or hand you enough edge to pay a spread.

The honest limits

Three years is three years. A framework that only works in some later regime is not tested here, though the framework's own sources make no such claim. And we tested the daily and 90 minute scales, not the 22.5 minute micro layer, where the cost per trade would be worse again.

The Dow result is the honest awkward one. It is positive in sample, on one market, at one anchor, with an interval that is not clear of zero. We would not trade it and we would not have published a framework on the strength of it, but it exists and you should know it does.

What we can say is bounded and specific. On seven markets, over three years, at both published day opens, at two timeframes, in both directions, and with all costs stripped away, the second quarter does not make the day's extreme and the third quarter does not run in a predictable direction. The clock is not doing what the framework says it does.