Using Multifactor Models

Welcome to Multifactor Models!

In Level I, you learned about the Capital Asset Pricing Model (CAPM), which suggests that only one thing matters: the "market." But the real world is a bit messier. Think of it like this: if you’re trying to predict how fast a car goes, the "market" model only looks at the size of the engine. A multifactor model, however, looks at the engine size, the wind speed, the weight of the driver, and the quality of the tires. It gives us a much more detailed picture! In this chapter, we’ll explore how to use these models to understand investment returns and manage risk like a pro.

1. Arbitrage Pricing Theory (APT)

The APT is the foundation for multifactor models. It’s based on a simple idea: in a rational market, you shouldn't be able to make a "riskless profit" (arbitrage). If two assets have the same risk but different returns, investors will buy the cheap one and sell the expensive one until the prices balance out.

Unlike CAPM, which says only market risk matters, APT says there are multiple sources of systematic risk. The formula looks like this:

\( E(R_i) = R_f + \beta_{i1}(\lambda_1) + \beta_{i2}(\lambda_2) + ... + \beta_{ik}(\lambda_k) \)

Where:
\( R_f \) = The risk-free rate.
\( \beta_{ik} \) = The sensitivity of asset i to factor k (also called "factor loadings").
\( \lambda_k \) = The risk premium for factor k (the extra return you get for taking on that specific risk).

Quick Review: The Three Assumptions of APT

1. Returns are described by a multifactor model.
2. There are enough securities to diversify away idiosyncratic (specific) risk.
3. No arbitrage opportunities exist among well-diversified portfolios.

2. Macroeconomic Factor Models

In a Macroeconomic Factor Model, we assume the return of a stock is driven by "surprises" in economic data. It's not the actual inflation rate that moves the needle—it's the difference between what everyone expected and what actually happened.

The Equation:
\( R_i = E(R_i) + b_{i1}F_1 + b_{i2}F_2 + ... + \epsilon_i \)

Where:
\( E(R_i) \) = Expected return based on analyst forecasts.
\( F_k \) = The surprise in the factor (Actual value minus Expected value).
\( b_{ik} \) = Sensitivity of the stock to that surprise.
\( \epsilon_i \) = The firm-specific surprise (things like a CEO scandal or a factory fire).

Example: If the market expects inflation to be 2% and it comes in at 3%, the "surprise" is +1%. If a stock has a sensitivity (\( b \)) of -2.0 to inflation, the stock price would likely drop by 2% ( -2.0 \(\times\) 1% surprise).

Key Takeaway: In macroeconomic models, the factors are external to the companies (like GDP, interest rates, or inflation).

3. Fundamental Factor Models

These models look inside the company. We use characteristics like the P/E ratio, market cap (size), or financial leverage. This is where things get a little flipped from the macroeconomic model, so pay close attention!

In Fundamental Factor Models:
- The "beta" or sensitivity is the actual characteristic (e.g., the company's P/E ratio).
- The "factor" is the return associated with that characteristic (e.g., how much "Value" stocks outperformed the market today).

To make this work, we usually use standardized betas (Z-scores):
\( \beta_{standardized} = \frac{(Value - Average)}{Standard Deviation} \)

Don't worry if this seems tricky! Just remember: In Macro models, we measure how much a stock reacts to a surprise. In Fundamental models, we look at the stock's attributes (like "Is it a small-cap stock?") to explain its return.

Comparison Table: Macro vs. Fundamental

Macroeconomic Model: Factors are "surprises" in variables. Factor sensitivities (\( \beta \)) are estimated via regression.
Fundamental Model: Factors are "returns" to a trait. Factor sensitivities (\( \beta \)) are calculated from company data (like P/E or Size).

4. Statistical Factor Models

These models are for the data scientists! Instead of picking factors like "inflation" or "P/E ratio," we let a computer program (using techniques like Principal Component Analysis) look at historical returns and find patterns.

Pros: It finds every possible relationship in the data.
Cons: The factors have no names. The computer might say "Factor 1" is important, but it won't tell you if Factor 1 is "interest rates" or "consumer confidence." This makes it very hard to explain to clients!

5. Using Models for Return and Risk Attribution

This is the "So what?" part of the chapter. Why do we do all this math? To see where our money is coming from.

A. Return Attribution

Active return is the difference between a portfolio's return and its benchmark. We can break it down:
Active Return = Factor Return + Security Selection Return

Factor Return: Did we win because we tilted the portfolio toward "Small Cap" or "Value"?
Security Selection: Did we win because we picked better individual stocks than the benchmark within those categories?

B. Risk Attribution

Just like return, we can break down Active Risk (also called tracking error):
1. Active Factor Risk: Risk coming from being "different" from the benchmark's factor exposures (e.g., having a higher beta to oil prices than the index).
2. Active Specific Risk: Risk coming from picking specific stocks that behave differently than their peer groups.

Did you know? Most professional fund managers try to maximize their "Information Ratio," which is just Active Return divided by Active Risk. It measures how much extra return you get for every unit of "uniqueness" you take on.

6. Portfolio Construction and the Multifactor Approach

When building a portfolio, managers use these models to "target" specific exposures. If you think the economy is going to grow faster than expected, you might use a multifactor model to find stocks with the highest sensitivity to GDP growth.

Factor-Based Investing: This involves creating portfolios that track specific factors, like a "Value" factor or a "Low Volatility" factor. This is often called Smart Beta.

7. Summary and Key Takeaways

1. APT is the theory that says multiple systematic risks drive returns, and no-arbitrage keeps prices in check.
2. Macro Models use economic surprises. The sensitivities are calculated via regression.
3. Fundamental Models use company traits (P/E, Size). The sensitivities are the traits themselves (Z-scores).
4. Statistical Models are data-driven but hard to interpret.
5. Active Return comes from either your factor tilts or your ability to pick specific winners (security selection).
6. Active Risk can be decomposed into factor risk and specific risk.

Common Mistake to Avoid: On the exam, don't confuse the factors in Macro vs. Fundamental models. In Macro, the factor is the surprise. In Fundamental, the factor is the return to the attribute. If a question mentions "surprises," think Macro!