Every model of a financial market is a counterfactual. It describes not the market as it is, but the market as it would be if the simplifying assumptions held. Sometimes those assumptions hold well enough to be useful. The problem arises when participants mistake the model for the territory, and then act with confidence on predictions that were never more than conditional approximations.
What a Counterfactual Actually Is
A counterfactual is a statement about what would happen under a specified set of conditions. When a physicist models the flight path of a ball, the formula works because the relevant variables, mass, velocity, angle, gravity, are stable and measurable, and the system is sufficiently isolated from interference. Watch the same throw enough times and a reliable predictive model emerges. The formula holds.
Now introduce a dog into the park. The ball’s flight path is the same, but the system is no longer isolated. The dog introduces a variable that the model has no mechanism to account for: unpredictable, reactive, and capable of invalidating the forecast entirely. Add more dogs, more owners, a gust of wind, a sore throwing arm, and the gap between the model and reality widens with each additional variable. Each park is unique. Each situation carries its own combination of factors that the counterfactual cannot anticipate. What works in a controlled setting fails in the open system, because the open system contains precisely the variables that the model was built to exclude.
Financial markets are the open system. The dogs are always present.
They are complex adaptive systems (CAS) in which the variables are not stable, the relationships between them are non-linear, and the act of modelling the system can itself influence the system being modelled. A trading strategy that works in one regime can alter participant behaviour in ways that erode the very edge it was designed to exploit. A risk model calibrated on historical data will perform well until the regime it was calibrated on ceases to exist.
The counterfactual assumption embedded in most financial models is that the measuring instrument is independent of what it measures. This is what allows practitioners to apply historical statistical relationships with confidence: the past distribution is assumed to be a reliable guide to the future one, because the system generating the data is assumed to be stable. In a world of deep correlations and feedback loops, this independence is not a simplification. It is a category error.
Irreducible Complexity and the Limits of Shortcuts
Stephen Wolfram’s work on computational irreducibility is directly relevant here. Some systems cannot be short-circuited. The only way to know where the system will be at time T is to run the system forward to time T. There is no formula, no compressed description, no analytical shortcut that produces the same answer faster. The system’s evolution must unfold step by step, because each step depends on the precise state produced by the previous one.
Financial markets exhibit this property during regime transition events. The dynamics of a transition cannot be derived from the statistical properties of the prior stable regime, because the transition is precisely the process by which those properties break down and are replaced. No model built on prior regime data can reliably forecast the path, magnitude, or endpoint of the transition. The complexity is irreducible, and the attempt to reduce it through statistical shortcuts produces confident predictions that fail at exactly the moments when the stakes are highest.
This is the deeper problem with counterfactual reasoning in markets. It is not merely that the models are imprecise. It is that the class of events that matters most for long-run geometric return outcomes, the fat-tail shocks that define Outlier Hunting, are structurally outside the reach of models calibrated on prior stable distributions. The counterfactual says: if conditions remain as they have been, then this is what will happen. Reality answers: conditions do not remain as they have been, and the most important events are precisely those in which they change most dramatically.
Path Dependence and the Single Realisation Problem
There is a further issue with counterfactual thinking that is especially significant for systematic traders: the single realisation problem. Statistical models of financial markets are typically estimated on a single historical path through time. The market did not run a thousand parallel experiments across different starting conditions and report back the average. It followed one path, shaped by the specific sequence of events that actually occurred. Conclusions drawn from that single path carry a margin of uncertainty that most practitioners do not adequately account for.
This matters because path dependence is a structural feature of complex adaptive systems. The market at any given moment is not simply in a state. It is in a state that was produced by a specific sequence of prior states. A market that arrives at a given price level after a prolonged drawdown is a different system from one that arrives at the same price level after a steady advance. The history of the path shapes the current distribution of participant positions, the warehoused risk embedded in the system, and the likely dynamics of future moves. Counterfactual models that treat the current state as sufficient for forecasting, without accounting for how it was reached, discard precisely the information that path dependence makes most relevant.
The Outlier Hunter’s Position
The Outlier Hunter’s response to the limits of counterfactual reasoning is not to build better counterfactuals. It is to operate in a way that does not require them.
Cutting losses short and letting profits run does not depend on a model of what the market will do. It depends on a rule about how to respond to what the market is doing. The position is sized to limit the loss if the move fails to develop. The trailing stop preserves participation if it does. No forecast of the endpoint is required. No assumption about the stability of the prior distribution is embedded in the process. The system is designed to remain functional across the full range of regime states, including the transition events that sit outside every counterfactual model’s scope.
The park with one ball and no dogs is the model. The park as it actually exists, crowded, unpredictable, and full of variables no formula anticipated, is the market. The Outlier Hunter does not build a better formula for the dogs. The process is designed to work regardless of where they run.