How 2 — Measuring “Factor Alpha”

Categories Investing, OSAM Research

This post is a how-to guide for interpreting our proprietary attribution model for factor-based investing strategies. (Also see our Factor Alpha Newsletter for correlative charts, periodically updated.)

Since 1996, the O’Shaughnessy team has been investing according to several factor-based themes — Quality, Valuation, Momentum, and Yield — in conviction-weighted and high active share portfolios. But the conventional Brinson attribution framework, though useful for some purposes, does not split out the performance of specific factor exposures to distinguish them from allocation and selection decisions.

In order to explain portfolios that include factor-based themes, we could use the factor-specific attribution analysis made available by some prominent vendors, but even their off-the-shelf attribution tools fail to sync with the unique definitions we use for Quality, Valuation, Momentum, and Yield. For example, Northfield and Barra both use Price-to-Book to measure exposure to value. Instead of Price-to-Book, OSAM uses a multi-factor hybrid to define value (e.g., Sales, Earnings, and Cash Flows) — so the measurement gap in our exposures can sometimes result in counter-intuitive metrics.

Several years ago, we built a proprietary factor attribution tool based on similar frameworks for common regression tools but using our own factor definitions.

Example: OSAM’s Factor Attribution Tool1
YTD Factor Alpha vs. the Benchmark (As of 9/3016)
2016-11-16_01

Framework

Our attribution uses a multivariate regression framework. In the “Factor Return” column, we establish the factor return stream against which portfolio and benchmark exposures are measured (see the “Key Factor Inputs” section below for definitions). In the “Average Exposure” columns, we then measure the Beta exposure of the portfolio and its benchmark versus the universe, our factor themes, and industry groups — including countries and currencies for strategies with international/global exposure (the latter three are the only overlap with conventional Brinson attribution framework).

Betas are calculated on a trailing 12-month basis. Once the Factor Return and Average Exposures are established, we link the daily factor return contributions of individual securities. The “Contribution” column shows the results for the full time period for the portfolio and benchmark respectively. This column effectively serves as a decomposition of total return based on factor performance. Finally, we calculate the “Average Active Exposure” as the difference between the portfolio and benchmark on a daily basis (aggregated for the full time period) to derive the “Factor Impact”.

Interpreting the Results

The three most important columns of the table are the first column (Factor Return) and the rightmost two columns (Average Active Exposure and Factor Impact). Factor Return is the return of the custom-defined factor inputs over the time period. Average Active Exposure measures the portfolio’s over/underweight to the theme, relative to the benchmark. Factor Impact is the decomposition of excess return for the portfolio versus the benchmark based on the themes. A positive number suggests a positive contribution to excess return and a negative number a negative contribution to portfolio excess. The “Residual” line item is a plug, which can be interpreted as the amount of return that is unexplained by the themes’ attribution. It is our expectation that the Residual line item can be higher in shorter periods.

Key Factor Inputs

Our attribution tool uses a multivariate regression framework to relate the exposure of a portfolio to its benchmark based on our themes. Here are definitions for how we measure each of the variables incorporated into the framework:

  • Universe
    Our equal-weighted selection universe for the portfolio.
  • Size
    Size is the difference between the return of our selection universe on a market cap-weighted and equal-weighted basis. This helps to isolate the difference in performance associated with a benchmark’s market cap-weighted methodology versus an equal-weighted selection universe. A positive Active Exposure on size indicates that the portfolio is positioned larger than the benchmark on the market cap spectrum.
  • Value
    Value is the excess return of the highest ranking decile of the multi-factor OSAM Value composite relative to our selection universe. Our Value Composite consists of underlying constituents such as price relative to sales, earnings, and cash flows. As mentioned above, this definition of value differs from common factor attribution frameworks because it doesn’t include Price-to-Book.
  • Momentum
    Momentum is the excess return of the highest-scoring decile of the multi-factor OSAM Momentum composite relative to our selection universe. Whereas common frameworks typically measure momentum using raw 6-month or 12-month momentum, OSAM Momentum uses multiple measures of momentum and penalizes highly-volatile stocks.
  • Yield
    Yield is the excess return of the highest-scoring decile of Shareholder Yield (the sum of Buyback Yield and Dividend Yield) relative to our selection universe. (In some strategies, such as the table above, Dividend Yield is used by itself instead of Shareholder Yield.)
  • Earnings Quality
    Earnings Quality is the excess return of the highest-scoring decile of the multi-factor OSAM Earnings Quality composite relative to our selection universe. The composite consists of several underlying constituents, which measure the conservatism of accounting choices through accruals.
  • Financial Strength
    Financial Strength is the excess return of the highest-scoring decile of the multi-factor OSAM Financial Strength composite relative to our selection universe. Its underlying constituents assess balance sheet leverage and strength.
  • Earnings Growth
    Earnings Growth is the highest-scoring decile of the multi-factor OSAM Earnings Growth composite relative to our selection universe. OSAM Earnings Growth consists of multiple underlying constituents, which measure the consistency of earnings and profitability.
  • Industry, Country, Currency
    The excess return of each industry group relative to the selection universe. For strategies with exposure outside of the U.S., the excess return of each country relative to the selection universe and each currency relative to the portfolio’s base currency. These three variables have some overlap with conventional attribution frameworks.
  • Residual
    Residual is the amount of return that is unexplained by the regression framework. It is our expectation that this line item can be higher in shorter periods.

In our quarterly strategy commentaries, we use this proprietary factor analysis to measure the “Factor Alpha” that can result from unique actively-managed equity portfolios.


 

  1. Numbers may not add up due to rounding.