TL;DR: the answer box
- A market regime is a label for today's trend and volatility. Trend from the 200-day average and its slope, volatility from 20-day ATR against 100-day ATR. Six regimes. The S&P 500 spent 39.5% of days since 1993 in Bull Quiet, the most common regime on all four markets.
- Regimes are short. On SPY the median spell lasted 2 to 14 trading days depending on the regime, and the label changed on 6.9% of days.
- Regimes step to a neighbour, nearly always. 0 of 564 SPY regime changes went straight between Bull and Bear. 82.2% of Bull Quiet spells ended by turning Bull Volatile.
- Volatility clusters. A volatile day was still volatile 20 days later 60.2% of the time, against 34.7% after a quiet day.
- The regime is a risk gauge, not a return forecast. Bear days' mean 20-day dip was 1.9x to 2.1x the Bull days' on all four markets. On return, 2 of 24 cells separated from the all-days mean.
- The "stay out" regime was not the worst one. SPY Sideways Volatile was followed by +2.33% over 20 days against +0.79% for all days, with a deeper dip (-4.07% against -3.50%).
What is a market regime?
A market regime is a label for the conditions the market is in right now. The version tested here is built from two things a chart already shows you: which way the trend points, and how violently price is moving. Cross three trend states with two volatility states and you get six regimes: Bull Quiet, Bull Volatile, Sideways Quiet, Sideways Volatile, Bear Quiet and Bear Volatile.
Think of it as a weather report. It tells you how rough the water is this morning. It does not tell you where the boat ends up. That one line is the whole article, and the numbers below are the proof.
Most pages on this topic stop at the definition, or teach a detection model and never say what happened after each label. I wanted the base rates: how often each regime shows up, how long it lasts, what comes next, and what the next month looked like from each one.
I started trading in 1998, at the dot-com peak. Nobody told me what kind of market I was walking into. A label would not have predicted the bust. It would have told me how rough the water was.
How I tested it
The rules, precise enough to rebuild.
| Piece | Rule |
|---|---|
| Bull | Close above its 200-day simple moving average, and that average above its value 20 trading days earlier |
| Bear | Close below its 200-day average, and that average below its value 20 days earlier |
| Sideways | Everything else: price and the average's slope disagree |
| Volatile | 20-day ATR above 100-day ATR |
| Quiet | 20-day ATR at or below 100-day ATR |
Three trend states times two volatility states: six regimes.
For trend I used the 200-day simple moving average, which StockCharts calls "the most popular long-term moving average" (StockCharts ChartSchool, moving averages). ATR, or average true range, is a standard gauge of recent daily movement that J. Welles Wilder introduced in his 1978 book (StockCharts ChartSchool, ATR). Comparing a short ATR with a long one asks a simple question: is this month rougher than the last five?
The timing. Each day's regime is decided at that day's close, using only bars up to that close. Every forward number is measured from the next day's open. That is Execution rule v2's timing: decide at the close, act at the next open. Nothing here reads a bar it could not have seen.
What I measured, per regime. Share of days. How long a spell lasts (an unbroken run of one label). Which regime comes next. The forward 5-day and 20-day return from the next open. And the 20-day dip: the lowest low over the entry day and the next 20 trading days, as a percent below that morning's open. The dip is the pain an entry that morning sat through, so every return here is paired with it.
The markets. SPY daily from 1993-12-14 to 2026-06-11, 8,179 classified days, 32.5 years. QQQ from 2000, DIA from 1998, and E-mini S&P 500 futures (ES) from 2007, using the ratio-adjusted file so percentage returns are real returns. 26,488 classified days in all.
The yardstick. Each regime is compared with the same measurement over every classified day of the same market. Consecutive days share most of their 20-day window, so 8,179 days are not 8,179 independent readings. Each regime's 90% interval resamples whole spells rather than single days. If that interval contains the all-days mean, the regime has not been told apart from an ordinary day.
This is measurement, not a trading system. No capital, no costs.
| Regime | S&P 500 (SPY) | Nasdaq 100 (QQQ) | Dow (DIA) | ES futures |
|---|---|---|---|---|
| Bull Quiet | 39.5% | 36.6% | 38.4% | 42.3% |
| Bull Volatile | 31.0% | 31.2% | 27.1% | 29.5% |
| Sideways Quiet | 4.8% | 6.5% | 7.7% | 5.1% |
| Sideways Volatile | 6.3% | 6.4% | 8.2% | 5.2% |
| Bear Quiet | 9.1% | 12.5% | 8.4% | 7.8% |
| Bear Volatile | 9.2% | 6.8% | 10.2% | 10.0% |
Share of classified days in each market regime.
On SPY, the two Bull regimes took 70.5% of classified days. Bear takes 18.4%. Sideways takes 11.1%.
How long do market regimes last?
At the median, days, not months. On SPY the median spell lasted 11 days in Bull Quiet, 7 in Bull Volatile, 2 in Sideways Quiet, 3 in Sideways Volatile, 4 in Bear Quiet and 14 in Bear Volatile.
The tail matters as much as the middle. Bull Quiet's 90th percentile spell was 56.5 days and its longest 127. So you get many short flickers and a few long stretches.
A common claim is that the quiet regimes last longest. Half right. Bull Quiet does run long. Bear Quiet does not: its median SPY spell was 4 days. The longest median belonged to Bear Volatile, at 14 days.
The label also flickers. It changed on 6.9% of SPY days, and 34.4% of Sideways Quiet spells lasted a single day. If you switch strategies every time the label moves, you will be switching a lot.
What usually comes after each regime?
A neighbour. Regimes almost never jump across the board.
On SPY, 82.2% of Bull Quiet spells (120 of 146) ended by turning Bull Volatile. Bull Quiet went straight to a Bear regime 0 times out of 146. Going the other way, 0 of 91 Bear spells went straight to Bull. Across all 564 SPY regime changes, zero went directly between Bull and Bear. The other markets were nearly as strict: 1 of 476 on QQQ, 3 of 582 on DIA, 0 of 287 on ES.
On SPY and ES, every move between Bull and Bear went through Sideways first. Across all four markets the skip happened 4 times in 1,909 regime changes. The Sideways label is short and rare, but it is nearly always the doorway. It is not a warning on its own, though: of 149 SPY Sideways spells, 88 turned Bull next and 39 turned Bear (26.2%).
Here is what that looked like through 2020.
Does high volatility always lead to low volatility?
Eventually, yes. Next, no. Volatility comes in clusters, which Benoit Mandelbrot described back in 1963: large changes tend to be followed by large changes, and small by small (Mandelbrot, "The Variation of Certain Speculative Prices", Journal of Business).
| Market | Volatile 20 days after a volatile day | After a quiet day | Median volatile run | Runs over 60 days |
|---|---|---|---|---|
| S&P 500 (SPY) | 60.2% | 34.7% | 18.5 days | 11.9% |
| Nasdaq 100 (QQQ) | 61.3% | 31.9% | 10 days | 9.3% |
| Dow (DIA) | 58.1% | 35.5% | 15.5 days | 10.6% |
| ES futures | 58.6% | 34.2% | 21 days | 11.1% |
Volatile = 20-day ATR above 100-day ATR, read at the close.
Every volatile run in the sample did end. On this definition it has to, because the short ATR always falls back under the long one. But on SPY the run lasted anywhere from 1 day to 124. If you are waiting for calm to come back "next", you could be waiting six months.
Can a market regime predict returns?
Rarely. This is the finding that matters most, so here is the whole table for SPY.
| Regime | Days | Mean 20d | Up | Dip | 5% dip | Ret/Dip | 90% interval | Dip interval |
|---|---|---|---|---|---|---|---|---|
| All days | 8,178 | +0.79% | 64.1% | -3.50% | 22.9% | 0.23 | n/a | n/a |
| Bull Quiet | 3,232 | +0.92% | 67.4% | -2.55% | 12.9% | 0.36 | +0.61% to +1.21% | -2.85% to -2.29% |
| Bull Volatile | 2,539 | +0.74% | 65.0% | -3.31% | 21.7% | 0.23 | +0.18% to +1.22% | -3.81% to -2.84% |
| Sideways Quiet | 390 | +0.89% | 64.1% | -3.17% | 21.5% | 0.28 | -0.22% to +1.96% | -4.18% to -2.40% |
| Sideways Volatile | 515 | +2.33% | 67.0% | -4.07% | 28.7% | 0.57 | +1.33% to +3.36% | -4.86% to -3.30% |
| Bear Quiet | 746 | -0.65% | 46.5% | -5.71% | 48.5% | -0.11 | -2.00% to +0.64% | -6.51% to -4.69% |
| Bear Volatile | 756 | +0.68% | 63.0% | -5.82% | 41.0% | 0.12 | -1.08% to +2.62% | -7.91% to -3.89% |
SPY, forward 20 trading days from the next open. Up = share of days with a positive 20-day return. Dip = mean 20-day dip. 5% dip = share of days whose dip reached 5%. Ret/Dip = mean return per unit of mean dip. The two intervals resample whole spells.
On SPY, two regimes separate from an ordinary day on return: Sideways Volatile above it and Bear Quiet below it. On QQQ, DIA and ES, none do. That is 2 of 24 regime and market cells.
The dip is different. 12 of 24 cells had a dip interval that excludes the all-days dip, and Bull Quiet (shallower) and Bear Quiet (deeper) did so on all four markets.
So the regime label, on this definition, is a weak return forecaster. If someone shows you a regime table and says "trade more in this one, it returns more", ask for the interval.
Which half of the label carries the information?
For the dip, the trend half. The gap in mean dip between Bear and Bull days was larger than the gap between Volatile and Quiet days on all four markets.
| Market | Bull dip | Bear dip | Bull 5% dip | Bear 5% dip |
|---|---|---|---|---|
| S&P 500 (SPY) | -2.88% | -5.77% | 16.7% | 44.7% |
| Nasdaq 100 (QQQ) | -3.91% | -8.37% | 26.2% | 54.9% |
| Dow (DIA) | -2.73% | -5.24% | 15.6% | 36.7% |
| ES futures | -2.84% | -5.82% | 15.7% | 43.2% |
Mean 20-day dip below the next open, by trend label alone.
On SPY, 44.7% of Bear days saw a 5% dip within 20 trading days. For Bull days it was 16.7%. The mean dip itself ran deeper on Bear days on all four markets: 2.0x the Bull days' on SPY, 2.1x on QQQ, 1.9x on DIA, 2.0x on ES.
The volatility half adds less. On SPY the mean dip was -3.15% on quiet days and -3.91% on volatile days, a gap of 0.76 points against 2.88 for Bear minus Bull.
Bull Quiet is the calm one. It had the shallowest mean dip on 4 of 4 markets, in both halves of the SPY sample, and in 8 of 8 market and parameter combinations I tried (SPY and QQQ, four parameter sets each). On SPY its return per unit of dip was 0.36 against 0.23 for all days.
Is Sideways Volatile really the regime to stay out of?
Not on these numbers. It is the regime that sounds the most dangerous: no trend and wide swings. On SPY it was followed by +2.33% over 20 days, against +0.79% for all days, and its interval (+1.33% to +3.36%) sits wholly above the all-days mean. It had the highest mean 20-day return on SPY, QQQ and ES, and in 8 of 8 SPY and QQQ market and parameter combinations.
Higher risk, yes. Its mean dip was -4.07% against -3.50%, and 28.7% of its days saw a 5% dip against 22.9%. It is also rare (6.3% of SPY days) and short (a 3-day median spell), so the evidence comes from 88 spells.
The honest reading: on SPY, Sideways Volatile paired a higher average 20-day return with a deeper average dip than an ordinary day. More pain on the way, more gain at the end, on average. Whether your strategy survives the dip to collect the end is a question about your strategy, not about the regime.
Bear Quiet looked like the real no-go on SPY: -0.65% over 20 days, up only 46.5% of the time. But it does not hold up. Before 2010 it averaged -1.16%. From 2010 it averaged +0.96%. It was the lowest-return regime in only 5 of 8 market and parameter combinations.
The verdict, and the honest limits
The definition from the original playbook holds: trend plus volatility, six regimes. The claim that regimes "do not predict, they frame risk" holds too, and now it has numbers behind it.
Where the common advice breaks: on SPY the median spell ran 2 to 14 trading days, with a long tail (Bull Quiet's longest was 127). Volatility does not reliably flip to quiet "next": 60.2% of the time it is still volatile a month later. And Sideways Volatile is not the worst regime on forward return. On SPY it had the highest.
Limits.
- One definition. Different lengths move the shares: Bull Quiet took 33.0% to 39.5% of days across the 8 SPY and QQQ combinations. I re-ran SPY and QQQ under 4 parameter sets. Bull Quiet's shallow dip and Sideways Volatile's high return held in 8 of 8 combinations, Bear Quiet's low return in only 5 of 8.
- Many comparisons. With 24 cells and 90% intervals, about 2.4 of 24 would clear the all-days mean by luck alone, so 2 of 24 on return is close to what chance alone produces. The dip came out at 12 of 24, five times that rough baseline, though I did not run a formal adjusted test.
- Overlapping windows. Day counts overstate the evidence. The 90% intervals resample whole spells to correct for it, but 88 Sideways Volatile spells is still a thin base.
- US equity indices only, all highly correlated. ES starts in 2007. No dividends, no costs.
- This measures what followed each label. It does not test a strategy that trades on it, a sizing rule that shrinks in Bear regimes, or a rule that waits for a regime to persist. Those are tests to run on your own system.
What this means for you
- Read the regime as a risk gauge, not a forecast. Bear days dipped about twice as deep. That makes Bear-regime sizing a test worth running on your own system. This study did not run it. Do not expect the label to tell you the month's return.
- Expect the label to change often. It changed on 6.9% of SPY days, so a rule that switches strategy on every change would switch on every one of them. Whether waiting for a regime to persist helps is a test to run on your own system.
- Watch Sideways, without panicking. On SPY no regime change went straight between Bull and Bear, and across all four markets 4 of 1,909 did. Sideways is the doorway both ways, and on SPY it led back to Bull more often than on to Bear.
- Test your own strategy by regime. These are base rates for the index, not for your system. If you want to build and test regime filters inside your own strategies, that is a module of the Algo Trading Masterclass.
The regime is a weather report. Pack for the weather. Don't bet on the destination.
If you want the next study like this before anyone else, it goes out in the Overfit newsletter. StatOasis.com/Overfit
Methodology: StatOasis research engine 1.21.0. Regimes from the engine's admitted 200-day simple moving average and simple-average ATR, decided at each close. Forward returns and dips measured from the next open by the engine's forward-outcome module. SPY 1993 to 2026, QQQ 2000 to 2026, DIA 1998 to 2026, ES futures 2007 to 2026. Measurement only: no trades, no costs, no dividends. Past behaviour of an index is not a promise about its future. Research, not investment advice.
