Arbitrage Pricing Theory and Multi-Factor Models › Building a Multi-Factor Risk Model

Factor models serve two purposes: explaining expected returns and decomposing risk. Let's build a complete factor risk model for a portfolio.

For a portfolio with weights $w_i$ across $n$ assets, the portfolio's factor exposure is the weighted average of individual exposures:

$$\beta_p^{j} = \sum_{i=1}^{n} w_i \beta_i^{j}$$

where:

• $\beta_p^{j}$: portfolio sensitivity to factor $j$
• $w_i$: weight of asset $i$ in the portfolio
• $\beta_i^{j}$: sensitivity of asset $i$ to factor $j$
• $n$: number of assets

This linearity makes factor models computationally tractable for large portfolios. Rather than tracking the correlations among thousands of individual securities, we need only track each security's exposure to a handful of factors. The portfolio's risk characteristics are then determined by its aggregate factor exposures, a dramatic simplification that enables practical risk management for institutional portfolios.

The portfolio maintains a market beta near 1.0 but has specific net exposures to size and value factors based on the underlying holdings.

The total variance of a portfolio in a factor model decomposes into factor risk and idiosyncratic risk. This decomposition is one of the most powerful applications of factor models because it separates diversifiable risk from non-diversifiable risk.

Using the factor covariance matrix $\boldsymbol{\Sigma}_F$ and idiosyncratic variances $\sigma_{\epsilon,i}^2$:

$$\sigma_p^2 = \boldsymbol{\beta}_p' \boldsymbol{\Sigma}_F \boldsymbol{\beta}_p + \sum_{i=1}^{n} w_i^2 \sigma_{\epsilon,i}^2$$

where:

• $\sigma_p^2$: portfolio variance
• $\boldsymbol{\beta}_p$: vector of portfolio factor exposures
• $\boldsymbol{\Sigma}_F$: covariance matrix of factor returns
• $w_i$: weight of asset $i$
• $\sigma_{\epsilon,i}^2$: idiosyncratic variance of asset $i$

The first term is systematic risk from factor exposures. This term depends on how the factors co-move with each other and how much exposure the portfolio has to each factor. Even with perfect diversification across many securities, this systematic risk cannot be eliminated because it arises from economy-wide forces.

The second term is idiosyncratic risk, which decreases as the portfolio becomes more diversified. Notice that individual idiosyncratic variances are multiplied by squared weights. When weights are small (as in a well-diversified portfolio), squared weights become very small, causing the idiosyncratic component to shrink rapidly.

The decomposition shows that systematic factor risk dominates, accounting for most of the portfolio variance. The idiosyncratic component is small because diversification across five stocks reduces stock-specific risk. With more holdings, idiosyncratic risk would decrease further.

Understanding which positions contribute most to portfolio risk helps with risk management. The marginal contribution to risk (MCTR) measures how much total volatility would change with a small increase in each position's weight:

$$\text{MCTR}_i = \frac{\partial \sigma_p}{\partial w_i} = \frac{\text{Cov}(R_i, R_p)}{\sigma_p}$$

where:

• $\text{MCTR}_i$: marginal contribution to risk of asset $i$
• $\sigma_p$: portfolio volatility
• $w_i$: weight of asset $i$
• $R_i$: return on asset $i$
• $R_p$: return on the portfolio

Intuitively, the MCTR shows that an asset's contribution to risk depends not on its standalone volatility, but on its covariance with the portfolio. Assets that move in sync with the portfolio increase risk, while those that move inversely can reduce it.

This insight changes how assets should be evaluated for portfolio construction. A highly volatile stock that is negatively correlated with the rest of the portfolio might reduce total portfolio risk when added. Conversely, a low-volatility stock that is highly correlated with existing holdings might substantially increase risk. The MCTR captures these portfolio-level effects that are invisible when examining securities in isolation.