The heaviest systematic selling in a market break often begins before a single trend signal has flipped. The trigger is a spike in volatility. Trend-followers are known for buying strength and selling weakness, but the second rule in their models, volatility targeting, can force them to cut positions while their trend view is still fully long. That rule makes their flows unusually predictable, and it also explains why the recovery after a washout tends to be slow rather than sharp.
The week calm stopped being calm
On February 5, 2018, the S&P 500 fell 4.1% in a single session. The VIX, the market's gauge of expected 30-day volatility, closed at 37.32, up from 17.31 the prior Friday. That was the largest one-day percentage jump in the index's history.
The fundamental backdrop barely changed that day. Earnings were strong and the economy was growing. The S&P 500 had just finished one of its calmest stretches on record, with January 2018 posting a large gain. Wage data on February 2 had nudged bond yields higher, but nothing in that report justified a near-doubling of implied volatility.
The selling came from structure. Inverse-volatility products blew up overnight. Commentary at the time widely attributed the follow-through over the next several sessions to systematic funds de-risking, with CTAs prominent among them. The precise dollar figures in those estimates varied by source and are not reproduced here. The direction of the flow is not in dispute: funds that had been running maximum equity exposure, because the market was both trending up and unusually quiet, were cutting.
The episode carries a useful lesson. Many of these funds still had positive trend signals that week, because the market was only a few percent off its high and the longer lookbacks still pointed up. They sold anyway. Before February 2018, the risk was hiding in the calm, not in the price.
The two rules that set a CTA's position
A commodity trading advisor (CTA) is a fund, usually systematic, that trades liquid futures in equity indices, bonds, currencies and commodities. Most run some version of trend-following: hold a position in the direction of recent price movement, without forming any opinion about value.
The position comes out of two independent inputs. One is a trend signal, a number between −1 and +1 that measures how strongly the instrument is trending. A common construction averages the sign of returns over several lookbacks, for example one, three and six months. The other is a volatility scalar, which sizes the position so the fund's expected risk matches a fixed target.
Combined:
Position = Capital × (Target volatility / Realized volatility) × Signal
Realized volatility is the annualized standard deviation of recent daily returns, often measured over 20 to 60 trading days. The formula has a clear consequence. A position can shrink without the signal moving at all, provided realized volatility rises.
Take a hypothetical fund with $10 billion of capital, a 10% volatility target on its equity sleeve, and a full long signal of +1.
- Realized S&P volatility at 12%: Position = $10B × (10 / 12) × 1 = $8.3B long
- Volatility jumps to 30%, signal still +1: Position = $10B × (10 / 30) × 1 = $3.3B long
That is $5 billion of selling with no change in trend view. Now suppose the decline runs long enough for the signal to flip to −1 while volatility stays at 30%:
- Position = $10B × (10 / 30) × (−1) = $3.3B short
The total move from the calm long to the stressed short is $11.6 billion. Close to half of it came from the volatility leg alone, and that leg moved first.
When you read a sell-side note on CTA "trigger levels," you should split it into these two components. The trigger levels describe the signal. The volatility scalar is often the larger and faster force.
Why the flow cannot be argued with
A discretionary manager selling into a decline can be talked out of it by a cheap valuation, a policy response, or a change of heart. A rule-based fund cannot. Its model takes price history as input, so its behavior over the next few sessions can be roughly estimated by anyone who knows the price path and makes a sensible assumption about lookbacks.
That predictability cuts two ways.
The first effect is that other traders position ahead of the flow. If desks estimate that a break below a given S&P level flips the medium-term signal, discretionary sellers have a reason to sell before that level and buy the CTA supply afterward. The mechanical selling can then arrive earlier and in a more concentrated form than the model alone would produce.
The second effect is an asymmetry in supply. Once a fund has cut to a small or short position, it has little left to sell. The mechanical seller exhausts itself. Late-stage selloffs often run into a market whose most rule-bound participants have already finished.
The third effect, and the less obvious one, is the shape of the recovery. Realized volatility uses a trailing window, so it decays slowly after a shock. A 30% reading does not fall back to 12% the day the market stabilizes. It falls as the violent days roll out of the window, which can take weeks. The volatility scalar therefore rebuilds exposure in increments. Re-leveraging becomes a grind of modest daily buying, not a single burst. The practical consequence is that a market recovering from a volatility shock often has steady systematic demand underneath it for weeks, even while the headlines remain nervous.
Why trend-following earns a return at all
A rule that buys after prices rise and sells after they fall should, in principle, be arbitraged away. It survives because several real economic actors move prices gradually instead of all at once.
Hedgers pay for insurance. Producers and consumers of commodities, and corporations managing currency exposure, hedge for business reasons, not to make trading profits. They accept a cost to transfer risk, and some of that cost accrues to speculators who take the other side and hold through the move.
Central banks move slowly by design. Policy rates change in steps over quarters, not in single leaps. When a tightening or easing cycle begins, bond and currency markets tend to reprice in stages as each meeting confirms the path. 2022 is the clearest recent case. Persistent short positions in government bonds worked for most of the year as policy rates kept rising.
Information spreads unevenly. Investors underreact to new information at first, then extend the move as more participants recognize it. Institutional mandates, committee approvals and rebalancing schedules all add delay. That staggered adjustment is what produces serial correlation in returns, and serial correlation is exactly what a trend signal measures.
None of these sources is guaranteed. Each depends on prices adjusting in stages. When prices instead jump to a new level and stay there, or reverse sharply, the edge disappears. This means you should treat trend as a premium earned for providing patience during staged adjustments, not as a law about how prices move.
Where the mechanism breaks
Trend-following fails in recognizable conditions, and the volatility rule can turn a small failure into a large one.
Range-bound, choppy markets. When prices swing back and forth without direction, the signal flips repeatedly. The fund buys near local highs and sells near local lows, paying transaction costs each time. This is the strategy's steady drain, and it can last for years. Trend-following as a category had a long stretch of weak returns in the 2010s, when policy suppressed the kind of staged macro moves the strategy depends on.
V-shaped reversals. The rule needs time to recognize a turn. After a sharp drop and an equally sharp recovery, a CTA can end up short near the bottom and slow to rebuild, because both the signal and the volatility scalar lag. March 2020 had this shape: a crash measured in weeks, followed by a policy-driven rebound that outran the lookbacks of most models.
Crowding. When many funds run similar lookbacks and similar volatility targets, their trades cluster at the same levels. A volatility shock then pulls the same lever across the whole category at once. In August 2024, the VIX touched about 65 intraday on August 5 and the Nikkei fell more than 12% in a session, as a crowded yen carry trade unwound. Systematic de-risking was widely cited as an accelerant. The exact split between CTAs, volatility-targeting funds and discretionary deleveraging is not verified here, but the mechanism matches the formula above: higher volatility forces smaller positions, whatever the signal.
Volatility measurement that lags reality. A fund using a 60-day window will be slow to cut and slow to rebuild. A fund using a 20-day window reacts faster but trades more. Neither choice avoids the core problem. Trailing volatility is lowest just before a shock, so leverage is highest at the moment of greatest fragility.
The practical point is that the risk of a CTA-driven break is usually greatest after a long, quiet advance, not after an obvious warning sign.
Reading the tape through the flow
The opening claim holds up against the evidence. Volatility, not the trend signal, often drives the first and sharpest systematic selling, and the slow decay of trailing volatility sets the pace of the recovery. Several observations follow for anyone reading markets.
A selloff that begins in a calm market has more mechanical fuel behind it than one that begins in an already volatile market, because exposure starts out larger. When realized volatility on a major index is near multi-year lows and the trend has been up for months, systematic funds are likely close to maximum long, and the next jump in volatility will be met with selling that no argument can halt.
After a shock, ask how much has already been sold. Once the volatility scalar has done its work and medium-term signals have flipped, the forced seller has finished. Later declines then depend on discretionary conviction, which is much easier to reverse.
During the recovery, track the trailing volatility window rather than the headlines. As the worst days drop out of the 20- and 60-day windows, mechanical buying resumes in steady increments, usually well before sentiment improves.
In short, CTAs size positions to price and price volatility, so their flows can be estimated from the tape. The useful question is not what the trend says. It is how much exposure the current volatility allows them to hold, and how quickly that allowance is changing.
