Skip to content
Quantoros

Article

While you trade, the market learns

A look at a paper on stealth trading and legal risk: how execution leaks your information to the market and why constraints in the objective reshape the optimal strategy.

Say you know an asset is undervalued. All that's left is to build a position and wait for the market to agree with you. Except that while you're buying, everyone else is watching your trades and revising the price. It's a slightly frustrating setup: through your own execution, you help the market figure out exactly what you were planning to profit from.

A paper on stealth trading and dynamic legal risk looks at precisely this situation. The foundation is classic, the Kyle-Back model: an informed trader, noise traders, and a market maker trying to extract information from the aggregate order flow. The market maker doesn't need to know your forecast. It's enough to notice that some of the buying might be driven by something meaningful.

This point is easy to miss when people talk about alpha. You found a pattern, tested it, launched the strategy. But an informational edge still has to be monetized, and the execution itself can gradually erode it. While you're building the position, the market learns from your actions, and your next trades happen on different terms.

Stealth and the identification problem

Stealth is also more interesting than "trade less and nobody will notice." In certain regimes, the informed trader can increase volume and still become less detectable, as long as the surrounding noise trading grows faster. What matters is relative scale. That said, it would be too bold to carry this over to any liquid market: the result depends on the structure of the noise and on what exactly the market maker observes.

This leads to an identification problem. A weak price response does not prove that nobody is trading on an informational edge. The trader may simply be choosing the timing and method of execution well. People try to detect them from their footprints, while the trader has every incentive to make those footprints as uninformative as possible. So the absence of an obvious effect may well be a consequence of strategic behavior.

Legal risk and the objective function

The paper adds another constraint: the risk of prosecution depends on the trading history. If the authorities intervene, trading stops, accumulated profits are confiscated, and additional penalties apply. You have to account for both how much you make and the probability of keeping it. A profit may already show up in your calculations, but that doesn't mean it will still be yours at the end of the period.

Then comes a useful nuance. Two strategies can earn the same profit, but one builds the position gradually while the other concentrates its trades in a short window. A penalty tied only to profit may barely distinguish between these trajectories. If instead the sanctions depend on trading intensity, the optimal trade schedule changes too.

It's a familiar objective-function problem. People discuss the quality of the optimizer before agreeing on what behavior they actually want. Then it turns out the solution is formally correct but unpleasant to use: overly aggressive trading, awkward risk concentration, sensitivity to constraints. The optimizer didn't violate anything. The constraints that stayed in the researcher's head simply weren't part of the problem.

Robustness to the choice of horizon

In some of the paper's specifications, trading accelerates toward the end of the given period. This immediately raises the question of how robust the result is to the choice of horizon. If the model stops accounting for consequences after a certain date, activity near that boundary can look more attractive than it should. Would the acceleration survive if the prosecution risk continued to exist after trading ends? Without that check, it's hard to separate the economic mechanism from an artifact of the problem setup.

Questions like these are what make theoretical papers useful for your own research. Gradually it becomes clear which part of the result comes from the participants' behavior, which from the information available to them, and which from constraints introduced to keep the model tractable. And before carrying a neat conclusion over into a trading system, it's worth checking whether it rests on exactly the assumption that will be the first to break in reality.

Sources

New articles by email