Forex pullbacks: how deep before a trend stops resuming
Loading the current reading…
The latest trade on every instrument, on Recent Signals
A trend pulls back. Does it carry on, or has it turned? The percentages above are the only version of that question that can be answered from data: of all the pullbacks that ever got this far, what share went on to make a new extreme. Measured on 15 markets, 96,341 pullbacks since 1999, and checked on years the measurement never saw.
How to read it
All four tables are updated hourly from one broker's live feed and use the same method on different bars: intraday reads hourly bars, short term four-hour, med term daily, long term weekly. A market can be shallow on one and deep on another — that is not a contradiction, it is the difference between where this hour sits and where this year does.
θ is set per timeframe as well as per market, and the coarser views use a finer setting: on weekly bars the hourly one would define a pullback as a move of around a quarter, of which there have been fifteen since 1999. So a deep pullback on the weekly table is a smaller swing, in bars, than a deep one on the hourly table. What does not change is the method, or where the line crosses: half-way at 0.50θ on all four.
State is how far the pullback has already retraced — shallow, half-way, or deep. A market that is not in a pullback shows its trend state instead, and its pullback columns are left blank.
Sure? is the percentage: of all the past pullbacks that got as deep as this one, the share that went on to resume the trend. It is a count of what has already happened across every market at once — not a degree of confidence about the market you are looking at.
The percentage is not a forecast. It is read off the single curve described in the next section — the one measured across every market at once, because the whole finding is that the same curve fits all of them. It is not fitted to that market's own history, and it does not know anything about today.
The three prices are the boundaries of the measurement itself, not levels to act on: target is the extreme the trend would have to retake for the pullback to count as resumed, reversal is where the retracement becomes large enough that the trend is counted as over, and abandon is the half-way mark between them. They are where the label in the State column changes from one word to the next. Nothing here says what to do at any of them.
The measurement
Every pullback in 27 years of hourly bars was labelled by how deep it went before it resolved, and by which way it resolved. Depth is expressed as a fraction of θ so that a slow market and a fast one can be compared at all: θ is set per market, from its own typical bar, and a retracement of 1.0θ is the point at which the trend is declared over.
Counted that way, the share that resumed falls almost in a straight line as the pullback deepens, and it passes half-way at 0.50θ — exactly half of the retracement that would end the trend. Shallower than that, resuming is the ordinary outcome. Deeper, it stops being so. Each market's own crossing point sits close to the same place.
A straight line has no steps in it, and that is itself a result: there is no depth at which the odds drop away faster than at the depth beside it. The levels a Fibonacci tool draws — 38.2%, 50%, 61.8% — are the opposite claim about the same pullbacks, measured against the move before them instead of against θ, and they can be counted the same way. We did: do Fibonacci retracements work puts the question to the corrections behind this page, and the levels do not survive it.
The same rule on 15 different markets
This is the part worth being sceptical about, so here is the check. θ and the curve were fitted on the first half of each market's history and scored only on the second — years the measurement never saw. Splitting each market at its own midpoint rather than on one shared date is what lets Bitcoin be in the test at all: its history starts long after the currency pairs, and any date early enough to give them a real out-of-sample half leaves Bitcoin with nothing to fit on. Each market was scored on its own rather than pooled, because pooling is how one market that behaves differently gets hidden inside an average.
Scored that way, every one of the 15 markets crosses half-way within 2.8 percentage points of 50% on hourly bars, and within 8.7 on the weekly view, which is the loosest of the four. Gold and Bitcoin are in that set on purpose — they are not currency pairs and they trade differently — and the crossing lands in the same place. That is what licenses the single curve the table at the top reads from.
It is worth being clear about what that does and does not demonstrate. A market in which this line came out anywhere else would be a market where price at these scales is not close to a fair coin — so the check confirms that none of these 15 is grossly mispriced at this scale, which is a weaker and more useful statement than "a surprising pattern survived a hard test". It is also why the box below can say there is no signal here: the straight line and the absence of an edge are the same fact.
The 15 markets overlap heavily — the euro, sterling and yen crosses share most of their moves, so the independent content is nearer four or five markets than 15. Turnmarks publishes turn markers for thirteen of them; the rest are measured here and have no instrument page.
What this does not tell you
It is not a signal. Nothing here is an instruction to buy or sell, and no combination of these readings adds up to one: they are counts of what happened after past pullbacks, not a rule for acting on the next one. What they are worth as a trading proposition is not something this page works out for you — that arithmetic needs your costs, your sizing and your entry, none of which are in the measurement.
It says whether, not when. How long a pullback has lasted turns out to carry almost no information once depth is known; age was mostly depth in disguise. There is no reading here that says a move is due.
Descriptions of what price did in the past on one broker's feed — not a prediction and not a recommendation. Turnmarks is chart analytics, not advice. Trading carries risk, including the loss of your capital.
How well calibrated is it
A percentage is worth what its calibration is worth, so that is measured too, weekly, and it is the number to hold this page to: across the last 1 year, readings were out by an average of 1.9 percentage points, and leaned slightly high, overstating how often pullbacks resumed. The audit re-fits as the engine does, because a model that refits itself and is then audited frozen will look wrong when it is not.
Where the markers fit
This page is context, not a call. What Turnmarks itself publishes is narrower: the bar where a push stops going anywhere, on the hourly chart, with every past marker's outcome measured per instrument. It does not tell you to buy or sell and it is not a signals group. There is a free plan and it needs no card.
The same question, one market at a time
Each market has its own forecast page: where it stands in a pullback right now, the probability its trend resumes from that depth, and what that number does not claim. EUR/USD · GBP/USD · AUD/USD · USD/CHF · USD/JPY · EUR/JPY · GBP/JPY · AUD/JPY · CHF/JPY · CAD/JPY · XAU/USD · BTC/USD
Every past marker, per market: EUR/USD · GBP/USD · AUD/USD · NZD/USD · USD/CHF · USD/JPY · EUR/JPY · GBP/JPY · AUD/JPY · CHF/JPY · CAD/JPY · XAU/USD · SPX500 · NatGas · WTI · BTC/USD · US30
Two measurements sit beside this one, on the same price series: the average daily range of each market, year by year since 1999 — how far a day travels, in pips — and which hours of the day each one actually moves in, measured on fifteen-minute bars.