Objective research to aid investing decisions

Value Investing Strategy (Strategy Overview)

Allocations for October 2024 (Final)
Cash TLT LQD SPY

Momentum Investing Strategy (Strategy Overview)

Allocations for October 2024 (Final)
1st ETF 2nd ETF 3rd ETF

Combining Fundamental Analysis and Portfolio Optimization

| | Posted in: Fundamental Valuation, Strategic Allocation

Can stock return forecasts from fundamental analysis make conventional mean-variance stock portfolio optimization work? In their December 2018 paper entitled “Optimized Fundamental Portfolios”, Matthew Lyle and Teri Yohn construct a portfolio that combines fundamentals-based stock return forecasts and mean-variance optimization and then compare results with portfolios from each employed separately. To suppress implementation costs, they focus on long-only portfolios reformed quarterly. Their fundamentals return forecasting model uses cross-sectionally normalized versions of book-to-market ratio, return on equity, change in net operating assets divided by book value and change in financial assets divided by book value. They update fundamental variables quarterly at the end of the reporting month. They generate stock return forecasts via a complicated multivariate regression of cross-sectionally normalized versions of the variables based on five years of rolling historical data. They then form a portfolio of the tenth (decile) of stocks with the highest expected returns, either value-weighted or equal-weighted. They consider several portfolio optimization methods, including minimum variance (requiring no return forecasts); mean-variance optimization with target expected return; and, Sharpe ratio maximization. Their combined approach employs fundamental stock return forecasts as inputs to those portfolio optimization methods that require returns. They use data from 1991-1995 to generate initial model inputs and 1996-2015 for out-of-sample testing. Using end-of-month data for a broad but groomed sample of U.S. common stocks with at least three years of historical data during January 1991 through December 2015, they find that:

  • Preliminary tests confirm that future monthly gross stock return relates positively to firm book-to-market, return on equity and growth in financing, and negatively to growth in assets and size.
  • Combining return forecasts from fundamentals with mean-variance optimization produces portfolios with higher out-of-sample gross Sharpe ratios, information ratios, factor model alphas and average mean-variance utilities than comparable use of fundamentals or portfolio optimization alone. For example, using Sharpe ratio maximization as the optimization approach:
    • Quarterly gross Sharpe ratio for combined fundamentals-optimization is 0.47, compared to 0.25 for the value-weighted top-decile fundamentals portfolio and 0.13 for the optimization top-decile portfolio.
    • Quarterly gross 5-factor (market, size, book-to-market, profitability, investment) alpha for combined fundamentals-optimization is 2.30%, compared to 0.95% for the value-weighted top-decile fundamentals portfolio and 0.26% for the optimization top-decile portfolio.
    • Excluding the 20% of stocks with the lowest market capitalizations lowers combined fundamentals-optimization quarterly gross Sharpe ratio (5-factor alpha) to 0.44 (1.66%).

In summary, evidence indicates that using fundamental analysis to forecast stock returns as an input to conventional mean-variance portfolio optimization may be attractive.

Cautions regarding findings include:

  • Reported results are gross, not net. Trading frictions associated with turnover from quarterly portfolio reformation would reduce all returns (more for equal-weighted than value-weighted alternatives). The authors do use a long-only approach to avoid shorting costs/constraints and quarterly rebalancing to limit trading frictions. They also test alternatives that exclude the smallest 20% of stocks. However, they do not explicitly address turnover or frictions. There may also be material data acquisition/grooming and model execution costs.
  • The study may inherit data snooping bias from methodology features taken from many cited sources. Direct bias derives from selection of the best of three return forecasting alternatives and testing of many alternative portfolios, such that the best-performing alternatives overstate expectations. There may be further direct bias from parameter selections used in fundamental analysis and portfolio optimization. 
  • The approach in the paper is beyond the reach of most investors, who would bear fees for delegating the work to a fund manager.
Login
Daily Email Updates
Filter Research
  • Research Categories (select one or more)