Does Stress Cause the Decision? Causal Questions in Observational Neurofinance

The research question: association or intervention effect?

A study may find that physiological arousal and decision errors occur together. That observation does not immediately answer whether reducing arousal would reduce errors. The research question concerns a different comparison: what would happen under a defined intervention, compared with a relevant alternative? This distinction separates prediction from causation. A variable can help predict an outcome without being a useful intervention target, and an intervention can affect outcomes through pathways not captured by a simple association. Greenland, Pearl and Robins explained how causal diagrams can help identify the variables and assumptions needed for an unconfounded effect estimate. [1] Their contribution is a framework for reasoning about relationships rather than a mechanical recipe for declaring causation. A diagram represents assumptions about how variables relate. Drawing an arrow does not verify it, and omitting an important common cause can make a neat diagram misleading. Scientific knowledge and study design remain essential.

A common cause can create an apparent physiological explanation

Consider a hypothetical study of a difficult trading simulation. Task difficulty could increase both arousal and the chance of an error. An association between arousal and errors might therefore partly reflect the shared influence of difficulty. The example is an analytical possibility, not a reported NSI finding. Other structures are possible: an error could itself alter physiology, or an earlier state could influence both exposure to a difficult situation and the eventual response. Pearl's work on causal diagrams developed a formal approach to asking whether assumptions and observed distributions are sufficient to identify a causal effect. [2] This makes the assumptions inspectable, but does not eliminate them. A researcher should state which variables are measured before the exposure, which occur afterward and which may be influenced by both. Merely adding every available variable to a regression is not a general solution to confounding and can change the question being estimated.

Selection and adjustment can create new problems

Suppose a study includes only sessions in which either market turbulence or a participant's distress prompted a review. Conditioning on that selection could create a relationship between turbulence and distress that differs from the relationship in all sessions. This hypothetical example illustrates why the sampling process belongs in the causal account. Records that are easy to obtain may not represent the population needed for the intended claim. Missing or selectively recorded sessions should not be treated as irrelevant administrative details. Hernán and Robins organize causal inference around explicit questions, interventions and assumptions, including the challenges of observational and longitudinal data. [3] For neurofinance, a phrase such as 'the effect of stress' is often too vague. Stress could mean a manipulated task condition, a self-report, a physiological response or an ongoing life circumstance. Different interventions on those quantities need not have the same effect. Defining the exposure carefully is a prerequisite for interpreting an estimate.

What a cautious NSI study could establish

The discussion of volatility, stress and order flow already highlights the possibility of two-way relationships. A prospective study should preserve temporal ordering and record relevant task conditions before the decision outcome. It should distinguish a descriptive association from a causal estimate and explain which additional assumptions support the latter. If unmeasured confounding remains plausible, the conclusion should reflect that limitation rather than using causal verbs simply because the relationship is statistically significant. Randomization can strengthen a causal design when an intervention is feasible, ethical and clearly specified, but it does not automatically solve every measurement or generalization problem. A simulated task may differ from live financial decisions, and participants may respond to instructions or awareness of observation. A result can be credible within the experiment while transferring poorly to another setting. The article on acute stress and habitual action is useful context for separating evidence about a particular task from broader claims about trading behavior.

NSI is a proposed interdisciplinary framework, not a clinically or financially validated prediction system. A responsible research program would distinguish several aims: describing patterns, predicting outcomes and estimating intervention effects. Each aim requires its own design and evaluation. Combining them under one score can obscure whether the evidence supports any of them. A physiological indicator that predicts an error would not, by itself, show that changing the indicator prevents the error. A transparent report can provide the proposed causal structure in words, identify plausible competing explanations and explain how each design choice addresses them. It should also acknowledge assumptions that the available data cannot test. Sensitivity analyses may explore how conclusions change under alternatives, but they should not be presented as proof that no unmeasured cause exists. The strength of the report lies in showing where the inference depends on evidence and where it depends on an explicit assumption.

The conclusion is that causal language should follow a causal question and a defensible design. Ask what intervention is being compared, establish the sequence of events and examine common causes and selection. Observational physiology can offer valuable clues without proving that the measured state caused a decision. Preserving that distinction makes neurofinance research more useful because it prevents a descriptive relationship from being mistaken for an established remedy.