Earnings forecasting in a global stock selection model and efficient portfolio construction and management

https://doi.org/10.1016/j.ijforecast.2014.10.003Get rights and content

Abstract

Stock selection models often use analysts’ expectations, momentum, and fundamental data. We find support for composite modeling using these sources of data for global stocks during the period 1997–2011. We also find evidence to support the use of SunGard APT and Axioma multi-factor models for portfolio construction and risk control. Three levels of testing for stock selection and portfolio construction models are developed and estimated. We create portfolios for January 1997–December 2011. We report three conclusions: (1) analysts’ forecast information was rewarded by the global market between January 1997 and December 2011; (2) analysts’ forecasts can be combined with reported fundamental data, such as earnings, book value, cash flow and sales, and also with momentum, in a stock selection model for identifying mispriced securities; and (3) the portfolio returns of the multi-factor risk-controlled portfolios allow us to reject the null hypothesis for the data mining corrections test. The earnings forecasting variable dominates our composite model in terms of its impact on stock selection.

Introduction

Expected returns on assets are a key input in the mean–variance portfolio selection process. One can estimate models of expected returns by using earnings expectations data, price momentum variables, and reported financial data. In this analysis, we construct and estimate a global stock selection model by using these data for the period from January 1997 to December 2011. Earnings expectations information has been being rewarded in global stocks for the past fifteen years or so, and we expect it to continue to be the primary variable driving global stocks. Despite the recent volatility of the momentum factor, momentum is still associated statistically with security returns, and can be used with other factors to rank stocks for purchase. A composite model of earnings expectations information, value, and momentum factors is estimated for global stocks in order to identify potentially mispriced stocks. In addition, the regression-weighting of factors enhanced the information coefficients relative to equally-weighted factors. Analysts’ forecast and momentum variables are dominant in the regression-based composite model of expected returns. We create portfolios for the period January 1997–December 2011, and simulate portfolio returns which we compare with a set of global stock benchmark returns.

We begin with a review of the literature on stock selection models in Section  2. In Section  3, we discuss the testing of a composite model of stock selection, incorporating earnings forecast information. We use an APT-based multi-factor risk model to create efficient portfolios in Section  4. In Section  5, we present and estimate the data mining corrections test. In Section  6, we discuss the relevance of the “alpha alignment factor” and show its relevance. Section  7 presents our summary and conclusions.

Section snippets

A literature review of expected returns modeling and stock selection models

There are many different approaches to security valuation and the creation of expected returns. One seeks to select expected returns inputs that are associated statistically with stock returns. The correlation coefficient between the strategy and the subsequent returns is referred to as the information coefficient, IC (Grinold & Kahn, 1999). The expected returns input normally consists of variables that are denoted anomalies, which can be used as inputs to the portfolio construction process in

Building and testing stock selection models

How does one develop, estimate, and test a global stock selection model? We use the Guerard et al. (2013) database of global stocks included on the FactSet database during the period January 1997–December 2011. The number of stocks grows to approximately 16,000 during the period 1997–2011. Guerard et al.’s (2013) universe was restricted to global stocks that were covered by at least two analysts, which reduced the number to approximately 7000–8000 stocks. One can survey the academic and

Efficient APT portfolio construction

The mean–variance (MV59) portfolio construction and management can be summarized as: minimizewTCwλμTw, where μ is the expected return vector, C is the variance–covariance matrix, w is the portfolio weights, and λ is the risk-return tradeoff parameter. The estimation of C is usually done by a multifactor model, in which the individual stock return Rj of security j at time t, dropping the subscript t for time, may be written like this: Rj=k=1Kβjkfk̃+ej̃.

The nonfactor, or asset-specific, return

A further test of data mining corrections

In the practical world of Wall Street, it is conventional wisdom to cut your historical backtested excess returns in half; that is, if your backtested excess return (the portfolio geometric mean return less the geometric mean of the benchmark) was 6%, or 600 basis points, an investment manager and/or a client might expect 3% excess returns in the future. How do we justify this cutoff?

Markowitz and Xu (1994) proposed three statistical models for estimating the cutoff, which are close to half.

In

The alpha alignment factor: an application to global earnings forecasting

Several practitioners have decided to perform a “post-mortem” analysis of mean–variance portfolios, attempted to understand the reasons for the deviation of ex-post performances from ex-ante targets, and used their analysis to suggest enhancements to mean–variance optimization inputs, in order to overcome the discrepancy. Lee and Stefek (2008) and Saxena and Stubbs (2012) define this as a factor alignment problem (FAP), which arises as a result of the complex interactions between the factors

Conclusions

Investing based on analysts’ expectations, fundamental data, and momentum variables is a good investment strategy in the long-run. Stock selection models often use momentum, analysts’ expectations, and fundamental data. We find support for composite modeling using these sources of data, as well as evidence supporting the use of APT multi-factor models for portfolio construction and risk control. We develop and estimate three levels of testing for stock selection and portfolio construction. The

Acknowledgments

The authors appreciate the comments of Martin J. Gruber, the referee, Anureet Saxena, and Robert Stubbs. Research conversations with Bernell Stone and Mustafa Gultekin enhanced our analysis. Vishnu Anand and Chris Martin of Axioma were extremely helpful in helping the authors use Axioma, particularly Chris’ work on limiting the numbers of securities (names) in the portfolios. Any errors remaining are the responsibility of the authors.

John B. Guerard is Director of Quantitative Research at McKinley Capital Management, in Anchorage, Alaska. He earned his AB in Economics from Duke University, his MA in Economics from the University of Virginia, MSIM from the Georgia Institute of Technology, and Ph.D. in Finance from the University of Texas, Austin. Mr. Guerard has published several monographs, including Quantitative corporate finance (Springer, 2007, with Eli Schwartz) and Introduction to financial forecasting in investment

References (57)

  • A.E. Beaton et al.

    The fitting of power series, meaning polynomials, illustrated on bank-spectroscopic data

    Technometrics

    (1974)
  • Berger, A. L., Israel, R., & Moskovitz, T. J. (2009). The case for momentum investing. AQR White...
  • J.M. Blin et al.

    Earnings forecasts, revisions and momentum in the estimation of efficient market-neutral Japanese and U.S. portfolios

  • B. Bruce et al.

    The handbook of corporate earnings analysis

    (1994)
  • J.S. Brush

    A flexible theory of price momentum

    Journal of Investing

    (2007)
  • J.S. Brush et al.

    The predictive power in relative strength and CAPM

    Journal of Portfolio Management

    (1983)
  • L.K.C. Chan et al.

    Fundamentals and stock returns in Japan

    Journal of Finance

    (1991)
  • E.J. Elton et al.

    Expectations and share prices

    Management Science

    (1981)
  • E.F. Fama et al.

    Cross-sectional variation in expected stock returns

    Journal of Finance

    (1992)
  • E.F. Fama et al.

    Size and the book-to-market factors in earnings and returns

    Journal of Finance

    (1995)
  • E.F. Fama et al.

    Dissecting anomalies

    Journal of Finance

    (2008)
  • E.F. Fama et al.

    Risk, return, and equilibrium: empirical tests

    Journal of Political Economy

    (1973)
  • R. Grinold et al.

    Active portfolio management

    (1999)
  • B. Graham et al.

    Security analysis: principles and technique

    (1934)
  • B. Graham et al.

    Security analysis: principles and technique

    (1962)
  • J.B. Guerard

    Is there a cost to be socially responsible in investing?

    Journal of Forecasting

    (1997)
  • J.B. Guerard

    Global earnings forecast efficiency

    Research in Finance

    (2012)
  • B. Guerard

    Introduction to financial forecasting in investment analysis

    (2013)
  • Cited by (59)

    • Harry Markowitz: An appreciation

      2023, International Journal of Forecasting
    • A novel two-stage method for well-diversified portfolio construction based on stock return prediction using machine learning

      2022, North American Journal of Economics and Finance
      Citation Excerpt :

      Investment in the stock market helps to generate more returns than other financial instruments but has the threat of market risk that may lead to a high loss. Since investors naturally like returns and hate risks, they hope to spread funds reasonably to high-quality assets, so as to achieve the purpose of maximizing returns and minimizing risks (Guerard, Markowitz, & Xu, 2015). As we all know, stock quality is significantly contingent upon its future performance.

    • Proposing a new function for evaluation of the financial risk of construction projects using Monte Carlo method: Application on Iranian construction industry

      2021, Journal of Building Engineering
      Citation Excerpt :

      These studies presented the forecasting models to estimate both time and cost of the projects. The bases of these forecasting models are Monte Carlo simulation, regression and correlation analyses, earned value management, genetic algorithm and fuzzy methodology approaches [1,5,10–20]. These researches are mostly performed based on the simulation methods by different subjects such as, budgeting, asset pricing, investing, financial risk estimation, investment decision-making model, risk mitigation, cost prediction and etc [1,5,13,20–23].

    • A new approach to estimating earnings forecasting models: Robust regression MM-estimation

      2021, International Journal of Forecasting
      Citation Excerpt :

      Estimating the So model using MM-estimation helps evaluate the differential impact of outliers on OLS regression and has the potential to help assess analysts’ forecast errors and value stocks. Analyst forecasts are frequently used in investment decisions, such as Guerard Jr., Markowitz, and Xu (2015). However, the analyst’s forecasts are found to contain analysts’ biases due to different incentives.

    • Portfolio formation with preselection using deep learning from long-term financial data

      2020, Expert Systems with Applications
      Citation Excerpt :

      According to Brooks et al. (Brooks, Rew, & Ritson, 2001), brokerage costs for purchasing and selling the stocks of FTSE 100 index is from 0.00 bps to 0.30 bps. Referring to the parameters of several empirical research (Almahdi and Yang, 2016; (Guerard, Markowitz, & Xu, 2015, Paiva, Cardoso, Hanaoka, & Duarte, 2019)), we decide to simulate transaction costs as 0.10 bps, 0.05 bps to present the results finally. Tables 6 to 8 provide insights into the financial performance of the LSTM+MV, compared to the baselines, without transaction cost, including transaction cost (0.1 bps, 0.05 bps) separately.

    View all citing articles on Scopus

    John B. Guerard is Director of Quantitative Research at McKinley Capital Management, in Anchorage, Alaska. He earned his AB in Economics from Duke University, his MA in Economics from the University of Virginia, MSIM from the Georgia Institute of Technology, and Ph.D. in Finance from the University of Texas, Austin. Mr. Guerard has published several monographs, including Quantitative corporate finance (Springer, 2007, with Eli Schwartz) and Introduction to financial forecasting in investment analysis (Springer, 2013), and edited The handbook of portfolio construction: contemporary applications of Markowitz techniques (Springer, 2010). John serves an Associate Editor of the Journal of Investing and the International Journal of Forecasting.

    Harry Markowitz  is the Chief Architect at Guidedchoice.com in San Diego, CA, and a quantitative research scientific advisor at McKinley Capital Management LLC in Anchorage, AK. He is a Nobel Laureate and the father of modern portfolio theory, and was named “Man of the Century” by  Pensions and Investments  magazine. He has received the prestigious John von Neumann Theory Prize for his work in portfolio theory, sparse matrix techniques, and the SIMSCRIP programming language. Dr. Markowitz earned his Ph.D. from the University of Chicago.

    GanLin Xu is the Chief Technology Officer of Guidedchoice.com, where he leads a team of finance, economics, and mathematics experts to develop Guidedchoice’s financial methodology, software, and systems. Dr. Xu earned his Ph.D. from Carnegie Mellon University. His research on portfolio theory has been published in various journals. He is a quantitative research scientific advisor at McKinley Capital Management LLC in Anchorage, AK.

    View full text