Why is it a mistake to describe a market as draws from one stable distribution? This note explains the mistake, using the February 2018 jump in U.S. equity volatility as the public example, and says what a regime is for.
One shape does not survive contact with a calendar
A distribution is a summary of how a quantity has varied: how often large moves appear, how the ordinary days look, how the extremes behave. It is a description of a sample. Treating it as the market itself is the mistake. The sample has a start and an end. Outside that window, or inside a different slice of it, the frequency of large moves can change. A rule fit to the summary is a rule fit to a period. The period is not a law.
The practical symptom is a model that looks calm on a long average and then meets a cluster of days the average had diluted. The cluster is not a surprise to someone who plotted the days instead of the summary. Averages hide regimes by construction. A regime, in this note, means a stretch of time during which the behavior of interest is different enough that a single summary misleads. The label is a reading aid. It is not a claim that the world switches among a known set of states with clean borders.
A regime is a slice of history in which the behavior under study differs enough that one distribution, fitted to another slice, is the wrong summary to trust.
February 2018 as a public break
February 2018 is a public episode in which measures of U.S. equity volatility, including the VIX index, moved far outside the quiet pattern of the months before. Products and rules that had been living on that quiet pattern met a different market within a few sessions. The episode is useful here because it is widely documented and because it does not require a private dataset to remember the shape of the lesson. A distribution estimated on the preceding calm did not describe the week that arrived.
Nothing in that sentence says what anyone should have held. It says that a single distribution fitted to the calm was the wrong object to trust as if it were the mechanism. A researcher who splits history only after seeing which split flatters a rule has not found regimes. That researcher has fit the labels to the result. The labels have to be motivated by something stated in advance, such as a level of a public volatility index, a change in policy, or a calendar of known events, and then the rule has to live with the slices it did not choose in hindsight.
Other periods would teach the same point. A year in which stocks and bonds fall together teaches it about correlation. A year of steady drift teaches it about how easy it is to mistake a trend for a law. February 2018 is only the example in this note, chosen because the break in volatility was public and sharp. The note does not recount a profit or a loss. It recounts a change in behavior that one distribution cannot hold.
What to do with the observation
The constructive habit is to report a rule on more than one slice and to let a slice say no. If the rule requires the calm distribution in order to look acceptable, the write-up should say that, in those words. Readers can then decide what they think of a rule that only works in one weather. Hiding the other weathers inside a full-sample average is how the single distribution sneaks back in.
The market is not one distribution because the process that generates prices changes, and because any summary is a choice of window. February 2018 is a reminder a reader can look up. It is not a template for the next break, and it is not a forecast. The useful remainder is the refusal to let one shape stand in for the calendar.
This piece does not claim that regimes can be predicted, that February 2018 will repeat, or that any rule survives a change in volatility. It argues only that one distribution is a summary of a window, not a description of the market.
