The Usefulness of Accounting Fundamentals at the Industry Level: Which Performance Metric Matters? Jack Strauss Philipp Schaberl University of Denver Beijing Jiaotong University
Overview Can we form industry portfolio using accounting variables to outperform a buy-and-hold out-of-sample? What variable(s) are most useful in selecting industry portfolios? Why are these variables useful? What type of forecasting framework makes sense?
Background Traditional Predictability Model regresses: R t = a + b 1 E t-1 /P t-1, Where R t is the market return Predictability measured by OOS R 2 00S < 5% Motivated by Timing Varying Risk or Gradual Diffusion of Information (Hong, Lim and Stein (JF, 1999); Hong, Torous and Valkanov (JFE, 2007; Rapach, Strauss and Zhou (JF, 2013). Pesaran and Timmermann (1995) report `An alternative approach to evaluating the economic significance of stock market predictability would be to see if the evidence could have been exploited successfully in investment strategies. This can be done' by evaluating portfolio allocation in "real time," and see if these portfolios systematically generate excess returns of forecasting performance, such as the directional accuracy.
Industry Level Very Little Academic Literature on Industry Predictability or Portfolio Allocation Rapach, Strauss and Zhou (JPM, 2013) show size and book to market can both predict R it = a + b i1 SZ/BM i,t-1 Size/Value Weighted sorts Portfolio Allocation: Rotate into long positions for industries with high returns and short industries with
Value Relevance Approach Ball and Brown (1968); Beaver (1979) identify the top and bottom decile of firm earnings and then plot the cumulative returns of these firms. A variable is value relevant if there is a large gap between the two deciles as the firms with high earnings should have increasing payoffs and the firms with low earnings should have declining payoffs. Earnings is more value relevant than cash-flow if the gap is larger as it is more linked to returns. Predictable and Relevant for Returns and CF or Earnings (Ohlson, RAS, 1999) So Transitory earnings should not affect returns – they are not predictable, and should also not be related to permanent stream of cash-flow/earnings
Our approach Use forecasted EA it = a + b i1 EA it-1+ + b i2 EA it-1 Then Sort Typically, choose High and Low R it To allow for lags in data release, we choose Highest and Lowest Decile R it+2M Use to forecast returns in
Data We choose Compustat data from Net Income/Assets Cash-Flow/Assets Operating Profits (Ball et al, 2014) Gross Profits (Novy-Marx, JFE 2013) Earnings/Price Returns are from the Fama-French 38 Industry Database. We use 31 industries
Predictability of Variables
Industry Portfolio Results – using Actuals
Industry Portfolio Results – using Forecasts
Portfolio Allocation using 4 month Lead
Consistent Over time
Granger Causality
Portfolio not Riskier
Portfolio not Driven by Size
Portfolio not determined by size