Within the world of finance, factor investing has gained prominence as a strategy aimed at explaining the consistent outperformance of certain stocks. This approach is based on the premise that investors can achieve rewards by exposing themselves to specific, undiversifiable risks such as value, momentum, quality, and size. However, after a period marked by disappointing results, researchers are beginning to reevaluate the foundations of these models. They are uncovering that the shortcomings may not stem from the data itself, but rather from the construction of these models.
A recent study has introduced the concept of the factor mirage, which highlights the risks associated with conflating correlation with causation in factor models. This misstep can lead to significant financial consequences, as many investors continue to allocate capital based on flawed assumptions.
Table of Contents:
The promise and perils of factor investing
The initial appeal of factor investing was its scientific approach, suggesting that certain risks were inherently linked to greater returns. However, the reality is more sobering; for instance, the performance of the Bloomberg–Goldman Sachs US Equity Multi-Factor Index has yielded a Sharpe ratio of only 0.17 since 2007. This indicates that these strategies have not provided substantial value to investors. This disappointing performance not only affects individual fund managers but also raises questions about the credibility of factor investing as a whole.
Understanding model mis-specification
One common explanation for the underperformance of factor investing strategies is backtest overfitting, where researchers manipulate data until it appears to show consistent alpha. However, a deeper issue has been identified: systematic misspecification of factor models. Most models are developed using traditional econometric methods, which often prioritize variables that correlate with returns rather than accurately understanding underlying causal relationships.
This methodological flaw can lead to significant biases in estimation. For instance, including a collider—a variable influenced by both the factor and the returns—can distort the coefficients, prompting investors to make poor decisions, such as buying what they should sell. Consequently, even if risk premia are estimated accurately, a misspecified model could result in persistent losses.
Consequences of overlooking causality
The phenomenon known as the factor zoo describes the many anomalies discovered in the literature that fail when tested in real-world scenarios. However, a more insidious issue is the factor mirage, which arises not from data manipulation but from flawed modeling approaches. These misspecified models often present misleadingly high R² values and lower p-values, misleading researchers into believing they are more valid than they truly are.
The impact of colliders and spurious relationships
When a model includes a collider, the relationship between returns and the collider variable can appear stronger than it actually is, leading to spurious correlations. For example, if one researcher accurately accounts for factors such as profitability and leverage while another incorporates return on equity—an outcome influenced by both profitability and stock performance—the latter’s model may falsely suggest a relationship between high quality and prior returns. While it may perform well in backtesting, this model can result in significant losses during live trading, eroding capital for investors.
The implications of model misspecification extend beyond individual portfolios. Research from ADIA Lab indicates that without accurate causal models, no investment portfolio can be deemed efficient. Even if the means and covariances are perfectly estimated, a misunderstanding of causal relationships can lead to suboptimal portfolio construction.
A call for a new approach
To truly leverage the potential of factor investing, a transition from correlation-based reasoning to an emphasis on causation is essential. This shift is not unique to finance; fields such as medicine and epidemiology have already recognized the importance of understanding causal relationships in their methodologies. The focus on causal inference can empower investors to make more informed decisions and allocate capital more effectively.
By developing models that accurately identify the true sources of risk and return, investors can enhance the reliability of their strategies. This approach will not only help explain performance more credibly but will also align with the broader goal of achieving practical results in financial markets.
A recent study has introduced the concept of the factor mirage, which highlights the risks associated with conflating correlation with causation in factor models. This misstep can lead to significant financial consequences, as many investors continue to allocate capital based on flawed assumptions.0
