Quantitative finance continues to debate the reliability and limitations of model-driven investment strategies. A key query is how much weight investors should give to backtesting.
In Marcos López de Prado, PhD, and Vincent Zoonekynd, PhD, we explain why investors should transcend simply assuming historical performance and give attention to understanding why a model works. This is a worthwhile contribution to strengthening the rigor of quantitative investing – and one which invites further reflection on the structure of this argument.
It could also be helpful to view the issue not as a binary selection between correlation and causation, but as a multifaceted problem wherein different types of pondering play different roles.
In practice, the selection isn’t between easy correlation and fully specified causation. Most investment evaluation falls somewhere in between. Sometimes we will describe and test a mechanism directly. Sometimes we won’t. The system could also be moving too quickly, key variables could also be only partially observable, or there will not be the time and resources needed to construct a more comprehensive model.
In these situations, association-based pondering still has value. This just isn’t a financial deficiency; It is a typical feature of decision making under uncertainty.
Association under duress
People often depend on associations when there is no such thing as a time to create a whole causal explanation. This is not necessarily irrational; it will possibly be adaptive. A fast association can guide motion before slower, more sophisticated reasoning is feasible.
The same applies to investment practice. When relevant drivers can’t be directly observed or the causal structure is just partially understood, association signals should still contain useful information.
Association just isn’t a proof. The query just isn’t whether the association has value, but whether it’s sufficient. For institutional investors, this distinction has practical implications for due diligence, including how managers justify the inclusion and exclusion of variables in systematic models. When fuller structural knowledge exists, it just isn’t sophistication to disregard it; it’s a loss of knowledge. The association has its place, however it mustn’t develop into a stopping point.
The call for greater causal discipline in finance just isn’t recent. The more interesting query is how one can integrate this discipline without oversimplifying the character of markets themselves.
Epidemiology as a model of structured pondering
An epidemiologist wouldn’t analyze an epidemic as a purely statistical pattern, divorced from what is thought about transmission. If susceptible individuals will be infected and infected individuals can get well or be removed, this data becomes a part of the model structure.
Compartmental models similar to SIR (susceptible, infected, recovered) and SEIR (susceptible, exposed, infected, recovered) formalize these transitions. Statistical methods remain essential for estimating parameters and testing suitability. But the evaluation doesn’t start from scratch; It begins with a longtime causal structure.
The financial industry can learn an identical lesson from this. Where enduring mechanisms are reasonably well understood, they must be presented explicitly. When leverage increases forced sales, refinancing conditions shape default risk, inventories influence pricing power, passive capital flows influence demand, or network structures transmit distress, these are greater than just recurring correlations. These are mechanisms that will be modeled, tested and questioned.
Dynamic models will be particularly useful here. A regression captures the co-movement; A dynamic model represents stocks, flows, delays and feedback. In finance, this may mean balance sheet capability, financing conditions, capital flows or acceptance dynamics. Such models help to make clear how the state of the system evolves and the way today’s conditions influence tomorrow’s results.
Reflexivity and adaptive markets
Finance is different from epidemiology.
Markets are reflexive. Beliefs influence prices, and costs in turn change beliefs, incentives, and financing conditions. A narrative can attract capital; Capital flows can move prices; Rising prices can reinforce the unique narrative. What appears to be an ongoing relationship may, for a time, reflect a self-reinforcing loop.
Causal reasoning stays vital, however the relevant structure may itself involve feedback between beliefs, processes, and outcomes.
A 3-layer framework
Investment research can occur at three different but interrelated levels:
- Association: What looks like a prediction, even imperfect?
- Causal: What mechanism could plausibly produce this relationship?
- Reflexive: How might using the signal itself alter behavior, overwhelm trading, alter flows, or reshape the modeled environment?
Seen this manner, the controversy just isn’t about selecting correlation over causation. It’s about knowing when association is sufficient, when mechanisms must be explicitly modeled, and when reflexive feedback makes the system more adaptable than each approaches assume.
Few serious quantitative researchers would defend correlation without testing. Solid practice already includes stress testing, economic intuition and structural pondering. The query just isn’t whether causality matters, but moderately whether we’re clear about which layer is doing the work—and the way those layers interact.
Towards a more disciplined quantitative practice
We should use causal knowledge when it is on the market and test causal hypotheses when we’ve it. When a phenomenon involves accumulation, delay, or feedback, dynamic models could also be more appropriate than static statistical matches.
Associational pondering retains a vital role, particularly under time and observability constraints. But where established structures exist, it just isn’t subtle to disregard them; it’s a loss of knowledge.
The opportunity for quantitative finance just isn’t to interchange one methodological slogan with one other. It’s about becoming more disciplined and transparent about how different types of pondering contribute to sound investment research – when patterns are enough, when mechanisms are vital, and when reflexivity demands that we treat markets as adaptive systems, shaped partially by our own participation.
Therefore, it’s unlikely that the long run of investment research can be purely correlational or narrowly causal. It can be more plural, more dynamic, and more explicit concerning the difference between patterns that merely appear stable and mechanisms that may maintain them.
References
Delli Gatti D, Gusella F, Ricchiuti G. Endogenous vs. exogenous fluctuations: Uncovering the results of heterogeneous expectations. Macroeconomic dynamics. 2025;29:e125. doi:10.1017/S1365100525100345
Gigerenzer, Gerd and Daniel G. Goldstein. “Thinking Fast and Frugally: Models of Bounded Rationality.” Psychological Review 103, No. 4 (1996): 650–669.
Kermack, WO and AG McKendrick. “A contribution to the mathematical theory of epidemics.” Proceedings of the Royal Society of London. Series A 115, No. 772 (1927): 700–721.
Greenwood, Robin, Samuel G. Hanson, and Lawrence Jin. “Reflexivity in Credit Markets.” NBER Working Paper No. 25747, April 2019.
