In recent years, factor investing has gained attention as a scientific approach to understanding stock market dynamics, aiming to explain why certain stocks consistently outperform others. However, many investors have expressed disappointment with the results over time. Growing research indicates that the underlying issues may stem from the construction of investment models rather than the data itself. A recent study introduces the concept of the factor mirage, highlighting how many models confuse correlation with causation.
Factor investing is based on the premise that investors can achieve superior returns by exposing themselves to specific, non-diversifiable risks such as value, momentum, quality, and size. This concept has attracted trillions of dollars into various financial products built around these principles. However, performance data reveals a troubling narrative. For example, the Bloomberg–Goldman Sachs US Equity Multi-Factor Index, which follows long-short strategies based on classic styles, has recorded a Sharpe ratio of only 0.17 since 2007, with no significant value added before costs.
The evidence
Many attribute the underperformance of factor investing to backtest overfitting or p-hacking, a process in which researchers sift through noise to identify apparent patterns. While this explanation holds some validity, recent findings from the ADIA Lab, published by the CFA Institute Research Foundation, suggest a more profound issue: systematic misspecification of models.
Typically, factor models adhere to an econometric framework that employs linear regressions, significance tests, and two-pass estimators. However, this methodology often conflates correlation with causation. Econometrics textbooks advise including any variable linked to returns, irrespective of its role in the causal pathway.
Understanding model misspecification
This methodological misstep can lead to significant biases in the coefficient estimates of factors. For instance, including a collider—a variable influenced by both the factor under review and the returns—can distort the results. Such bias can even reverse the sign of a factor’s coefficient, causing investors to incorrectly buy or sell securities. Even if the risk premiums are accurately estimated, a misspecified model can still result in persistent losses.
Consequences of the factor mirage
The phenomenon known as the factor zoo reflects numerous anomalies that fail to hold up in real-world applications. However, ADIA Lab researchers highlight a more insidious issue: the factor mirage. This illusion arises not from data manipulation, but from models that, while adhering to standard econometric practices, are fundamentally flawed.
Models incorporating colliders tend to report higher R² values and lower p-values, misleading researchers into believing they are more accurate than they truly are. In such cases, the return value is predetermined before considering the collider’s value, preventing investors from monetizing the perceived association. The investment gains promised by academic research are often illusory, leading to significant financial ramifications.
A practical example of misspecification
Consider two researchers analyzing a quality factor. One carefully controls for variables such as profitability, leverage, and size, while the other includes return on equity—a variable affected by both profitability and stock performance. The latter’s model creates a false correlation: high-quality stocks appear to correlate with past returns. While this model may seem superior during backtesting, it fails in real-world trading, resulting in capital losses.
Errors in model specification do not merely impact individual managers—they can skew capital allocation and create broader market inefficiencies. Recent findings from ADIA Lab assert that no investment portfolio can achieve efficiency without causal factor models. If the underlying factors are incorrectly specified, even accurate estimates of means and covariances will lead to suboptimal portfolio outcomes.
The need for a paradigm shift
To address the challenges facing factor investing, a transition from merely complex data analysis to causal reasoning is essential. The finance field must look to other disciplines, such as medicine, which shifted from correlation to causation decades ago, enhancing treatment efficacy based on evidence.
Factor investing is based on the premise that investors can achieve superior returns by exposing themselves to specific, non-diversifiable risks such as value, momentum, quality, and size. This concept has attracted trillions of dollars into various financial products built around these principles. However, performance data reveals a troubling narrative. For example, the Bloomberg–Goldman Sachs US Equity Multi-Factor Index, which follows long-short strategies based on classic styles, has recorded a Sharpe ratio of only 0.17 since 2007, with no significant value added before costs.0
Factor investing is based on the premise that investors can achieve superior returns by exposing themselves to specific, non-diversifiable risks such as value, momentum, quality, and size. This concept has attracted trillions of dollars into various financial products built around these principles. However, performance data reveals a troubling narrative. For example, the Bloomberg–Goldman Sachs US Equity Multi-Factor Index, which follows long-short strategies based on classic styles, has recorded a Sharpe ratio of only 0.17 since 2007, with no significant value added before costs.1