Methods of combining quantitative and fundamental analyses using stock screens on the internet are evaluated. The authors highlight the top 10 data-mining mistakes, recognize that not all quantitative research is created equal, and conclude that a successful model should be designed to fit the needs of the ultimate user.
What’s Inside?
The objective of the authors’ research is to highlight the most effective stock screens used by quantitative analysts and to determine the validity of the factors used in the stock screens chosen using a “quant-amental” approach (i.e., a combination of quantitative and fundamental analyses). They describe the data they use and the research and design of proprietary stock screens. They construct a robust quantitative model designed to be used in a stock screening process. The model factors are based on fundamentals and are commonly used by fundamental analysts and portfolio managers. The final section includes empirical results, and the authors explain how to combine proprietary stock screens with fundamental research to maximize short- and long-term investment portfolio returns.
How Is This Research Useful to Practitioners?
The authors develop PTC (passive turnover constraint) optimization: a systematic security selection process driven by a rigorously tested quantitative model created to support talented fundamental analysts that should lead to strong long-term investment performance.
Also discussed are the top 10 data-mining mistakes to avoid: focusing too much on training or overfitting a model; relying on one technique; asking the wrong or unclear questions; listening only to the data; accepting leaks from the future (i.e., the look-ahead bias); discounting pesky cases (i.e., unexplained results); extrapolating too much within the quantitative calculations; trying to answer every question; sampling without due care; and placing too much reliance on the best models.
How Did the Authors Conduct This Research?
The authors’ research universe is defined as the 1,000 largest publicly traded companies in the United States by market capitalization, excluding American Depositary Receipts (ADRs). To avoid survivorship bias, the authors include not only companies currently trading but also companies that have dropped out of the sample data because of a bankruptcy or merger. As a result, they are confident that their backtesting results do not suffer from upward performance bias.
Furthermore, fundamental data are retrieved from the Compustat non-restated US database as well as the Compustat US fundamental database for the period of January 1987–December 2012. Earnings estimate data are obtained from the I/B/E/S US consensus historical database for the same period. Stock price/return data are from FactSet. Compustat non-restated US data are used with an appropriate data lag to avoid look-ahead bias.
The authors’ research and design are created with the construction approach of a proprietary stock screen. Five model tenets are considered: robust, intuitive, parsimonious, stable, and evolved. A list is compiled of fundamental, macroeconomic, technical, and stock-specific factors based on the latest academic research and practitioner research, including input from fundamental analysts and portfolio managers in keeping with the authors’ “quant-amental” approach.
These factors are grouped into four general categories: (1) market expectations, (2) valuation, (3) profitability, capital deployment, and financing, and (4) earnings quality. For each factor, the authors calculate the following measurement statistics to evaluate the factor selection process during the backtests performed: buy value added, torpedo avoidance value, persistent hit rate, downside persistent hit rate, hit rate, monotonic signal, information coefficient, t-statistic, factor turnover, and Sharpe ratio. From this point, they use fundamental analysis to validate the alpha generated from the quantitative model.
Abstractor’s Viewpoint
This research is useful for security analysts and portfolio managers seeking to generate excess portfolio returns for their clients. Although the research is technical in nature, all the steps and processes are clearly listed and defined. For example, the authors highlight three portfolio approaches to combining a quantitative model’s signal with the research opinions of fundamental analysts—a quantitative model as a screen, a validation source, and an independent source of alpha. Using this knowledge on a standalone basis, practitioners could create their own equity investment portfolio using the steps and sources listed to generate long-term excess market returns.