Factor investing has been recognized as a systematic approach to understanding market dynamics, aiming to clarify why certain stocks consistently outperform others. However, recent insights suggest that the challenge lies within the construction of the models rather than the data itself. A significant study indicates that many traditional factor models confuse correlation with causation, leading to what has been termed a factor mirage.
The concept of factor investing originates from a compelling hypothesis: markets inherently reward exposure to specific risks that cannot be diversified away, such as value, momentum, quality, and size.
This framework has driven the allocation of trillions of dollars into various investment products based on these principles.
Table of Contents:
Examining the performance metrics of factor investing
A closer look at the data reveals a concerning trend. For instance, the Bloomberg–Goldman Sachs US Equity Multi-Factor Index, which measures the long-short performance of established style premiums, has yielded a Sharpe ratio of merely 0.17 since 2007. This statistic (t-stat=0.69, p-value=0.25) suggests that performance levels are statistically indistinguishable from zero when accounting for costs. In simpler terms, factor investing has frequently failed to deliver substantial value to investors, resulting in prolonged periods of underperformance and eroded confidence among fund managers.
Identifying the root causes of underperformance
Common explanations for these disappointing results often cite issues like backtest overfitting or the phenomenon of p-hacking, where researchers sift through random noise in search of alpha. While these factors contribute to the dilemma, recent research by the ADIA Lab, published by the CFA Institute Research Foundation, highlights a more profound flaw: systematic misspecification.
Many factor models are constructed using traditional econometric methodologies, relying on techniques such as linear regressions and significance tests. This reliance often leads to conflating mere association with genuine causation. Econometric principles advocate for including any variable correlated with returns, without considering the actual role these variables may play in the causal framework.
Consequences of methodological errors
This methodological misstep can yield significant biases in coefficient estimates. For instance, if a model incorrectly includes a collider (a variable influenced by both the factor and the returns) or fails to account for a confounder (a factor that impacts both the factor and returns), it distorts the underlying relationships. Such distortions can flip the sign of a factor’s coefficient, prompting investors to make poor decisions—buying assets they should sell and vice versa. Even under ideal conditions where risk premiums are stable and accurately assessed, a poorly specified model can lead to substantial and systematic losses.
The danger of the factor zoo and mirages
The notion of a factor zoo refers to the proliferation of published anomalies that fail to hold up under out-of-sample testing. However, the ADIA Lab’s research introduces a more nuanced issue: the factor mirage. This phenomenon arises not from data mining but from inherent shortcomings in model specifications, even when adhering to established econometric principles.
Models that incorporate colliders can create misleadingly high R² values and lower p-values, often mistaken for indicators of accuracy. In such scenarios, the return values are determined prior to the collider’s influence, rendering any purported profits from academic findings illusory. The implications of these methodological flaws can be staggering, potentially costing investors billions.
For example, consider two researchers assessing a quality factor. One controls for profitability, leverage, and size, while the other includes a variable like return on equity, which is affected by both the quality factor and stock performance. The latter approach generates a false correlation, leading to the erroneous belief that high quality equates to high past returns. While the backtested model might appear superior, real-world trading often reveals its shortcomings, draining capital instead of enhancing it.
The necessity of causal reasoning in finance
ADIA Lab’s findings assert that without causal factor models, no portfolio can achieve true efficiency. If foundational factors are incorrectly specified, even the most precise estimates of means and covariances will yield suboptimal outcomes. This indicates that investing transcends mere prediction; increased complexity in models does not inherently lead to better results.
The concept of factor investing originates from a compelling hypothesis: markets inherently reward exposure to specific risks that cannot be diversified away, such as value, momentum, quality, and size. This framework has driven the allocation of trillions of dollars into various investment products based on these principles.0
The concept of factor investing originates from a compelling hypothesis: markets inherently reward exposure to specific risks that cannot be diversified away, such as value, momentum, quality, and size. This framework has driven the allocation of trillions of dollars into various investment products based on these principles.1
The concept of factor investing originates from a compelling hypothesis: markets inherently reward exposure to specific risks that cannot be diversified away, such as value, momentum, quality, and size. This framework has driven the allocation of trillions of dollars into various investment products based on these principles.2
