tickdistill-learn

Why Market Days Cluster: Volatility Clustering, Regime Persistence, and What “Bull Follows Bull” Really Means

By TickDistill — order-flow microstructure signals. Educational content, not financial advice.

The intuition is right — but be precise about what clusters

Experienced traders notice that markets seem to come in streaks: calm begets calm, turmoil begets turmoil, and a strong trend often continues longer than a coin-flip world would allow. This is not folklore. It is one of the most thoroughly documented facts in quantitative finance. But the literature is precise about which quantity clusters, and getting that precision right is what separates a real edge from a comforting story.

Three distinct phenomena hide under “days cluster”:

  1. Volatility clustering — the size of moves persists. (Very strong, universal.)
  2. Regime persistence — markets sit in bull / bear / quiet states that are sticky. (Strong, well-modeled.)
  3. Return-direction autocorrelation — whether an up day literally makes another up day more likely. (Weak and horizon-dependent — this is where naive intuition overreaches.)

1. Volatility clustering: the rock-solid fact

Benoît Mandelbrot (1963) first observed that “large changes tend to be followed by large changes, of either sign, and small changes by small changes.” Note “of either sign”: it is the magnitude, not the direction, that persists.

This was formalized by Robert Engle (1982) with the ARCH model and generalized by Bollerslev (1986) as GARCH — work that earned Engle the 2003 Nobel Prize in Economics. The mechanism: today’s variance depends on recent variance, so volatility shocks decay slowly rather than resetting each day. Rama Cont (2001) catalogued this as a core “stylized fact” present across stocks, indices, currencies, and commodities, alongside fat tails and the near-absence of return autocorrelation.

The practical reading: a violent day makes another violent day far more likely — but it does not, by itself, tell you the direction.

2. Regime persistence: bull/bear states are sticky

James Hamilton (1989) introduced the Markov regime-switching model, originally for business cycles, now a workhorse for markets. The idea: the market occupies one of a few hidden states (e.g. a calm bull regime and a volatile bear regime), each with its own mean and variance, and it transitions between them with estimated probabilities.

The empirical signature of “clustering” lives in the transition matrix: the diagonal probabilities are high — often ~0.95–0.98 — meaning that, conditional on being in a regime today, the market most likely stays in it tomorrow. Hamilton & Susmel (1994), and later surveys like Ang & Bekaert on regime changes in financial markets, confirm this across equity returns. A common asymmetry: calm bull regimes are more persistent and longer-lived; high-volatility bear regimes are sharper and shorter.

This is the rigorous version of “a bull day is more likely followed by a bull day”: not because up-returns mechanically beget up-returns, but because the market is in a persistent bullish regime, and regimes are sticky.

3. The honest caveat: naive daily sign-following is weak

Here is where careful quants part ways with naive intuition. If you simply ask “does a positive daily return predict a positive next-day return?”, the answer is: only weakly, and it depends on the horizon and the instrument.

Lo & MacKinlay (1988), using the variance-ratio test, found positive autocorrelation in weekly/monthly index returns (a momentum signature) but negative autocorrelation in individual securities. The broader, robust picture: short-horizon momentum coexists with longer-horizon mean reversion (Poterba & Summers; Jegadeesh & Titman 1993 on 3–12 month momentum). At the daily scale, raw directional predictability is small and unstable — far smaller than the volatility/regime persistence above.

Takeaway: what clusters strongly and reliably is volatility and regime; what clusters weakly is naive day-to-day direction. A trustworthy product respects that distinction instead of blurring it.

In crypto, the effect is stronger — and tradeable as context

Crypto markets show pronounced volatility clustering and clear regime structure. Studies fitting Markov-switching GARCH (MS-GARCH) to Bitcoin returns consistently find two volatility regimes and report that regime-switching specifications outperform single-regime models for risk forecasting (e.g. one-day-ahead Value-at-Risk). GARCH parameter sums near 1.0 confirm high volatility persistence in BTC and ETH. Being open 24/7 and sentiment-driven, crypto tends to exhibit longer, cleaner trend and volatility regimes than mature equity indices — which makes regime context especially useful for order-flow interpretation.

Why this matters for order-flow signals

The same order-flow reading means different things in different regimes. A burst of one-sided committed buying during a persistent low-volatility bull regime is continuation; the identical burst at the tail of an exhausted high-volatility move can be capitulation. Regime is the context that disambiguates the microstructure. Treating every signal as regime-agnostic throws away information that the data plainly contains.

Where TickDistill is going with this

We are studying a future regime-state package — a signal that classifies the expected day type (bull / bear / consolidation) from volatility-clustering and regime-switching methods, delivered the same way as everything else: point-in-time-correct, backtestable, and tunable.

Two honesty rules govern it, consistent with how we build everything:


Sources

TickDistill sells clean, computed order-flow inputs — not trading advice or guaranteed alpha. Backtests are illustrative and not a promise of future results.