30 Years Of Accounting Lessons: Penman, Piotroski, Beneish And More – Part III by Steven De Klerck – Also follow him on Twitter here
5. Beneish, Lee and Tarpley, 2001
Name of investment strategy: /
Number of variables used: 21
Use of statistical techniques: YES
Before starting with a brief discussion of the Beneish, Lee and Tarpley (2011) paper, it is important to mention that the paper does not report realizable portfolio returns; this fact is noted in a single line on p. 179 of the paper:
Michael Mauboussin: Here’s what active managers can do
The debate over active versus passive management continues as trends show the ongoing shift from active into passive funds. Q2 2020 hedge fund letters, conferences and more At the Morningstar Investment Conference, Michael Mauboussin of Counterpoint Global argued that the rise of index funds has made it more difficult to be an active manager. Drawing Read More
“The disadvantage of this approach is that, strictly speaking, we cannot implement a trading strategy based on this model.” (Beneish, Lee and Tarpley, 2001, p. 179)
Beneish, Lee and Tarpley (2001) start from the observation that US stocks in the top two percent realize an average +69% size-adjusted return per quarter over the 1976-1997 period; stocks in the bottom two percent realize an average -28% per quarter. Starting from this observation it would be quite lucrative when we as investors should be capable to forecast ex ante the stocks with extreme (positive) returns. To that end Beneish, Lee and Tarpley (2001) apply a two-stage approach, first using market-based signals to identify likely extreme performers (both winners and losers), and then applying fundamental signals to differentiate between extreme winners and extreme losers among the firms identified as likely extreme performers in the first stage. Beneish, Lee and Tarpley (2001) consider their two-stage procedure an application of a contextual financial statement analysis.
They find that extreme performers, i.e. stocks with extreme positive and negative quarterly returns, have the profile of young growth firms engaged in more speculative ventures as reflected in the variables AGE, SIZE, AVGVOL, SGI, STDRET, R&D and S/P. All the variables used by Beneish, Lee and Tarpley (2001) are shown in the following table.
Within the group of extreme performers Beneish, Lee and Tarpley (2001) find that extreme losers are confronted with lower sales growth (SGI), decreasing margins (GMG), lower R&D expenditures (R&D), more negative earnings surprises (CHGEPS), negative momentum (FRTN6), more aggressive accruals (ACCRUAL) and higher capital expenditures (CAPX). They conclude that extreme losers have the profile of over-extended growth stocks. As a consequence, these variables might be useful to growth investors who want to avoid growth “torpedoes” in their portfolio.
The paper is discussed by Sloan (2001). Sloan (2001) argues that the selection of the context (i.e. extreme performers) is characterized by a poor business-economic rationale on the one hand and that the market and accounting variables are mechanically selected from the past literature on the other (i.e. the variables are not tailored to the specific context).
Overall the in-sample methodology used by Beneish, Lee and Tarpley (2001) and the corresponding absence of realizable portfolio returns lead to a disappointing reading.
6. Mohanram, 2005
Name of investment strategy: G-SCORE
Number of variables used: 8
Use of statistical techniques: NO
Mohanram (2005) follows in the footsteps of Piotroski (2000). Mohanram (2005) applies a financial statement analysis to a broad sample of low book-to-market or glamour firms. The financial statement analysis consists of eight signals: signals related to earnings and cash flow profitability, signals related to naïve extrapolation and signals related to accounting conservatism. The following table provides an overview of the three categories.
Accounting signals used to compute G-SCORE (Mohanram, 2005)
The first three signals of G-SCORE are part of the “Earnings and Cash Flow Profitability” category. It is assumed that relatively highly profitable firms are more likely to maintain their relative high profitability. Consequently companies with net income to total assets and cash flow from operations to total assets larger (smaller) than the industry median are rewarded with a score of +1 (0). The third signal in this category incorporates the findings by Sloan (1996).
As part of the “Naïve Extrapolation” category two accounting signals – relative earnings variability and relative sales growth variability – are added to the list. Mohanram (2005) reasons that the odds that current strong performance is sustainable, is significantly lower for firms with relatively high variability in earnings and sales growth. Consequently firms with below (above) industry median variability in earnings and sales growth get a score of +1 (0).
The three signals under the heading of “Accounting Conservatism” reward companies with above median levels of investments in R&D, capital expenditures and advertising. These investments depress current earnings and book values – the two drivers underlying the residual earnings model – but may boost future earnings growth rates when the investments start to pay off. R&D and advertising expenditures are often expensed based on traditional accounting rules. Mohanram (2005) fails to explain why the Capital Expenditure to Total Assets ratio belongs to the “Accounting Conservatism” category.
The eight signals are again assembled in an index: G-SCORE. Mohanram (2005) shows that G-SCORE is able to identify future winners and future losers among low book-to-market firms. In line with Piotroski (2000) Mohanram (2005) finds that firms with the strongest business fundamentals – as measured by G-SCORE – realize the highest returns. Finally, Mohanram (2005) shows that the G-SCORE strategy is less successful in high book-to-market firms compared to low book-to-market firms. Analogously, the F-SCORE strategy introduced by Piotroski (2000) is less effective for growth stocks compared to value stocks.
Piotroski (2005) provides a critical discussion of the G-SCORE strategy in general and gives an in-depth comparison between the F-SCORE and G-SCORE strategies in particular. Piotroski (2005) remarks that the implementation costs associated with G-SCORE are significantly higher compared to F-SCORE. F-SCORE uses an absolute performance benchmark and does not require an investor to gather any accounting data outside of the most recent annual report. G-SCORE, on the other hand, is implemented using industry benchmarks. The use of industry benchmarks results in at least two additional costs: (a) investors must collect the annual reports of all firms in the industry and (b) if all firms in the industry do not have the same fiscal year end, the investor must synchronize the industry data through the use of quarterly accounting statements.
In regard to the documented results, Piotroski (2005) notes that the real strength of G-SCORE resides on the short-side. Low G-SCORE firms earn annual size-adjusted returns of -17.5 percent whereas high G-SCORE firms earn average annual size-adjusted returns of only 3.1 percent. This is shown in the following table.
F-SCORE, on the other hand, is able to identify both successful long and short positions. High F-SCORE firms generate a mean annual market-adjusted return of 13.4 percent; for low F-SCORE firms Piotroski (2000) documents a mean annual market-adjusted return of -9.6 percent. Piotroski (2005) also notes that – in contrast to F-SCORE – the effectiveness of G-SCORE increases as the information environment and/or information dissemination mechanisms improve.
I believe that the Beneish, Lee and Tarpley (2001) and Mohanram (2005) variables can be used mainly by growth investors trying to avoid so-called “torpedo” growth stocks, i.e. growth firms with the highest probability of negative changes in their development curve. As a consequence and also noted by Piotroski (2005) in his discussion on G-SCORE, these two investment strategies are closer to informed speculation than informed investment as outlined in Graham and Dodd (1934) and Graham (1949).
In Part IV, I focus on the structured fundamental analytical approach introduced by Nissim and Penman (2001) and applied by Penman and Zhang (2006) in their working paper on measuring earnings/profitability sustainability. The empirical results by Penman and Zhang (2006) are used by Wahlen and Wieland (2010) in the development of PEI-SCORE, which stands for Predicted Earnings Increase-SCORE or PEI-SCORE.