💰 quantitative finance

Epistemic Limits of Empirical Finance: Causal Reductionism and Self-Reference

This paper argues that the pursuit of unidirectional causal reduction in empirical finance is fundamentally flawed due to the self-referential nature of capital markets, suggesting instead that quantitative tools are best suited for ex post inference and that alternative frameworks acknowledging competing causal chains and reflexivity are necessary.

Original authors: Daniel Polakow, Tim Gebbie, Emlyn Flint

Published 2026-05-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Daniel Polakow, Tim Gebbie, Emlyn Flint

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Idea: The Map Is Not the Territory

Imagine you are trying to navigate a city. In a normal city, the streets are fixed. If you turn left at the bakery, you always end up in the park. Traditional science and financial models usually work this way: they assume a fixed cause (turning left) and a fixed effect (arriving in the park).

The authors of this paper argue that financial markets are not a fixed city. They are more like a living, breathing crowd that watches each other.

In a crowd, if everyone sees someone running toward the exit, they all start running. The "cause" (one person running) creates the "effect" (mass panic), but the mass panic then changes the behavior of that first person. The rules of the game change while the game is being played. The paper calls this reflexivity.

The Core Problem: The Fallacy of the "One-Way Street"

Modern financial science loves "causal inference." This is the scientific attempt to prove that A causes B.

Example: "Interest rates rose (A), so stock prices fell (B)."

The paper argues that in the financial world, this "one-way street" thinking is often a trap.

The market is self-referential: Markets are systems that talk to themselves. If a famous economist says, "Stock prices will rise," people believe it, buy the stocks, and make them rise. The prediction caused the reality.
The direction switches: Sometimes A causes B, but then B causes A again. Sometimes they cause each other simultaneously.
The "Emperor's New Clothes": The authors suggest that quantitative finance (QF) is dressed in the robes of "hard science" (like physics) and pretends to predict the future with certainty. But since markets consist of human psychology and feedback loops, they do not follow the same rigid laws as falling apples or orbiting planets.

The Toy Model: The Predator and the Prey

To prove their point, the authors use a metaphor from nature: predator and prey (like wolves and rabbits).

In a simple model, one might think: "More wolves (A) cause fewer rabbits (B)."
But in reality, if there are too few rabbits, the wolves starve and die, meaning there are fewer wolves, which allows the rabbits to grow again.

The authors built a mathematical model showing that even if you know the exact rules of how wolves and rabbits interact, you cannot reliably predict the future population by looking only at past data. The system is too chaotic and self-referential.

They apply this to the financial world:

Investors are the wolves and rabbits.
Trend followers (wolves) eat contrarians (rabbits), but then the contrarians adapt, or the trend followers become overcrowded and crash.
The paper shows that even with perfect mathematics, it cannot be clearly determined who causes whom by looking only at the numbers. The "cause" is often an illusion generated by the behavior of the crowd.

The Two Types of "Science" in the Financial World

The paper draws a sharp distinction between two ways we use financial models:

Model Estimation (The "Truth" Trap): This is when we look at past data and say, "We found the rule! A causes B, so we can predict the future." The authors say this is dangerous in the financial world because the market changes its rules based on what we think.
Model Calibration (The "Pragmatic" Approach): This is when we admit: "We don't know the true rule, but let's adjust our tools to help us make decisions now." This is like a pilot adjusting the plane's controls based on the current wind, rather than assuming the wind will always blow the same way.

The Conclusion: Why We Are Always Surprised

The paper concludes that financial crises happen because we continue to try to treat a reflexive, human system like a closed, mechanical system.

The Reality: Markets are "open systems." They are influenced by news, fear, greed, and the fact that people observe the models themselves.
The Danger: If we assume "A causes B" and build a massive investment strategy on it, we often create a situation where the market reacts to our strategy and breaks the rule "A causes B."
The Verdict: The authors argue that empirical financial science is currently a "pathological science" in many areas. It looks like science but fails to predict crises because it ignores that the market participants are reading the same scientific books and changing their behavior.

Summarizing Analogy

Imagine you are trying to predict the weather by looking into a mirror.

Traditional Science: Assumes the mirror only reflects the sky. If the sky is blue, the mirror shows blue.
The Paper's View: The mirror is actually a giant screen that changes the weather. If the screen displays "Sunny," people go outside, and their presence somehow changes the clouds.

The paper warns us: Stop predicting the weather by looking at the screen. Instead, admit that the screen is part of the storm, and be very humble about what we can actually know.

Technical Summary: Epistemic Limits of Empirical Financial Research

Problem Statement
The paper addresses a fundamental epistemological crisis in Quantitative Finance (QF) and empirical financial research: the uncritical application of scientific reductionism and unidirectional causal inference to capital markets. While a resurgence of interest in causal inference (supported by the work of Pearl, Angrist, and Pischke) is observable to correct "flawed" statistical methods, the authors argue that this movement risks "totalizing" disparate phenomena. The core problem is that capital markets are open, self-referential (reflexive) systems, whereas standard modeling of causal relationships presupposes closed systems with stable, unambiguous, and unidirectional causal chains (A causes B).

The authors assume that the prevailing paradigm in QF treats markets as if they conform to epistemic norms suitable for natural sciences. However, in reflexive systems, the act of observation and prediction alters the system itself (the "Oedipus effect" or "self-fulfilling prophecy"). Consequently, the assumption that causal chains are fixed and can be unambiguously extracted from data to support ex ante predictions is highly problematic. When markets switch between different causal directions or exhibit competing causal chains, standard models fail, frequently leading to unforeseen financial crises.

Methodology
The paper employs a critical-theoretical analysis rather than empirical data testing. The methodology consists of three main components:

Epistemological Critique: The authors recapitulate the history of economic thought and contrast orthodox (neoclassical) and heterodox schools. They analyze the limits of the "Rational Expectations" framework (Muth, 1961), which assumes that reflexivity has no impact on prediction accuracy. Relying on the work of Soros, MacKenzie, and Lawson, they define markets as "open systems" where correlations are not invariant and the system can modify its own structure.
Theoretical Modeling (Toy Model): To illustrate the instability of causal inference in reflexive systems, the authors adapt a predator-prey model from ecology (May, 1976; Sugihara et al., 2012). They construct a system of logistic difference equations (LDE) comprising two coupled variables, $x(t)$ $x (t)$ and $y(t)$ $y (t)$ .
- The model simulates two competing causal chains, where $x$ and $y$ interact via coupling parameters ( $a_{y \to x}$ and $a_{x \to y}$ ).
- The system is discretized to demonstrate how unidirectional causal relationships can generate complex, nonlinear dynamics (bifurcations).
- The authors demonstrate that even with complete knowledge of deterministic causal rules and parameters, the system's behavior becomes unpredictable (chaotic) under certain growth rates ( $r$ ) and carrying capacities ( $K$ ).
Conceptual Synthesis: The paper synthesizes these findings to argue that in a reflexive environment, the distinction between "model estimation" (assuming the model is true) and "model calibration" (managing uncertainty with potentially misspecified models) is critical.

Main Contributions

Reflexivity as a Causal Breaker: The paper formalizes the argument that reflexivity destabilizes unidirectional causality ( $A \to B$ ). In self-referential markets, $B$ can cause $A$ , or the causal direction can change based on market participants' beliefs, rendering the causal graph non-unique and indistinguishable from observational data without intervention.
The Distinction of "Pathological Science": The authors propose a nuanced classification of empirical financial research. They argue that QF is not merely "at risk" of becoming a pathological science (as suggested by López de Prado regarding false claims); rather, in contexts where reflexivity is negligible (allowing ex post inference), it is categorically a pathological science, but in contexts where reflexivity dominates (making ex ante prediction impossible), it is obviously not a science.
Causal Predation and Competition: By adapting ecological models, the paper introduces the concept of "causal predation," where competing causal chains coexist and interact. This challenges the notion that a single causal graph can represent market dynamics, suggesting instead that multiple, competing versions of reality coexist and vie for dominance.
Epistemic Limits of Current Tools: The paper claims that current QF tools, under standard assumptions, are limited to ex post causal inference. When these assumptions are challenged by reflexive contexts, the prescription of unidirectional causality fails.

Results and Findings

Indistinguishability of Feedbacks: Analysis of the LDE model shows that feedback dynamics embedded in a unidirectional causal graph (Equation 1) are effectively indistinguishable from those in a reversed causal graph (Equation 2), unless a top-down intervention occurs.
Unpredictability in Deterministic Systems: Even in a deterministic model with known causal rules, the introduction of coupling and self-interaction leads to bifurcation and instability. This implies that stable causal graphs cannot be unambiguously derived from market data, even if the underlying causal mechanisms are known.
Failure of Ex Ante Prediction: The study concludes that the conditions required to extend ex post understanding to ex ante prediction are severely constrained in reflexive systems. The market's adjustment mechanism and the malleability of causal chains by actors prevent the establishment of invariant laws.
The "Demi-Regularities" of Markets: Drawing on Lawson, the paper suggests that market regularities are "demi-regularities"—probabilistic and susceptible to changes in underlying structural shifts, rather than being invariant laws.

Significance and Claims
The paper claims that the search for causality in highly reflexive systems may be "doomed to fail" if it relies on standardized reductionist frameworks. Its significance lies in:

Challenging the "Science" Framework: It argues that QF often dresses itself in the "robes of causality" without acknowledging that the scientific method cannot unambiguously recognize its own limits in self-referential systems.
Redefining Risk Management: It suggests that risk management must go beyond standard model validation to include "out-of-the-box" challenges to causal assumptions and the acceptance of the risk of a "false model."
Call for Epistemic Humility: The authors argue that empirical financial research frequently seeks answers under the sole epistemic light it possesses (unidirectional causality), which is insufficient for reflexive markets. They call for a deeper understanding of reflexivity to avoid the "Tragedy of the Commons" in factor investing and prevent the misapplication of statistical instruments.
Necessity of a Paradigm Shift: The paper concludes that a viable alternative paradigm is required to account for market anomalies and reflexivity, noting that without such a shift, QF remains trapped in a "spiral of tautological self-reference."

The authors maintain a modest tone regarding specific algorithmic solutions and explicitly avoid the "data science problem" or the merits of specific machine learning algorithms. Instead, they focus on the fundamental nature of the problem: the adoption and proliferation of models in a reflexive environment that assumes an independent life apart from computational tools.