Factors fluctuate between periods of under and out performance. In addressing the question of what worked in the trailing 12-months, our results indicate that factors based on analyst revisions, which investors tend to believe do not work well anymore, performed solidly. Factors based on earnings quality and valuation also produced positive returns.
A common question investors ask these days is what factors have performed well over the course of the past year?
In this note, we examine the trailing 12-month (T12M) performance of our StarMine models. We focus on the performance from July 2011 to June 2012, using a universe of the largest 3,000 securities in the U.S., the largest 2,000 securities in Developed Europe, 1,000 securities in Developed Asia ex-Japan, and 1,500 securities in Emerging Markets. Portfolios are rebalanced monthly; we ignore the impact of transaction costs in calculating returns. The results presented below are purely out-of-sample as these models were all been released between 2006 and early 2011 and models are not modified once released in production.
Although StarMine offers ten distinct stock selection factors, we concentrated on factors that represent a broad range of styles. As investors often comment that data tied to analyst revisions no longer correlates with future returns, we decided to include the StarMine Analyst Revisions model (ARM) which predicts future changes in analyst sentiment. We also employed the StarMine Relative Valuation model (RV), a stock ranking model that sorts companies by intelligently combining information from six powerful valuation ratios into a single comprehensive measure of relative valuation. Investors tend to take advantage of the complementary information in value and momentum signals, so we therefore include the StarMine Value-Momentum model (Val-Mo). That model combines StarMine’s two valuation models (Intrinsic Valuation and Relative Valuation) with StarMine’s two momentum models (StarMine Analyst Revisions Model and StarMine Price Momentum), into one exceptionally strong stock-ranking model. Also important to factor in is the StarMine Earnings Quality (EQ) signal, which is based on accruals, cash flow, and operating efficiency. When investors become more risk averse, they tend to put greater emphasis on these qualities. Finally, we look at StarMine Smart Holdings (SH), which ranks stocks based on the predicted future change in institutional ownership.
The first step we took was to examine a simple long-only strategy that creates portfolios from those stocks that rank in the top 10% of our universe, by each StarMine model. Figure 1 looks at the T12M return calculated as the difference between the top decile return and the equal-weighted average return (i.e., the market return). With the exception of RV in the US and Developed Europe, all the StarMine models performed solidly as part of a long-only strategy. A strategy based on reacting to signals coming from analyst revisions (ARM) and quality (as reflected in high EQ scores) outperformed the market in the trailing 12 months in all regions.
For the benefit of investors who employ quant and equity long/short strategies, we also compared the StarMine models with the appropriate commonly used quantitative factors that most closely approximate a simple alternative factor. The benchmark model for StarMine ARM is a basic revisions model, defined as the percentage change in the current fiscal year (FY1) consensus earnings estimate over the last 30 days. We compare the results of RV to a commonly-used valuation metric — earnings yield — based on the consensus forecast for next year’s (FY2) earnings per share. (When that isn’t available, we replace it with the consensus estimates for the current fiscal year, FY1.) For StarMine Val-Mo, we used an equal-weighted blend of the rank of the benchmark factor associated with each StarMine Val-Mo component (earnings yield, analyst revisions and trailing 12-month total return). We then compared the results of StarMine EQ to those of a basic earnings quality model based on total accruals (where total accruals is defined as the year-over-year change in net operating assets, scaled by total assets). The benchmark model for Smart Holdings is defined as the percentage change in number of owners from the previous quarter.
In Figure 2, below, you can see the results of portfolios formed from the StarMine models as well as the benchmark models, with the T12M decile spread plotted. The StarMine alpha factors have performed solidly over the past 12 months, outperforming their benchmark models in all regions. That outperformance was particularly strong in Developed Europe for the EQ model. Indeed, the EQ model typically performs very well during periods of high market volatility, as investors flock to the relative safety offered by stocks with good earnings quality. Given the impact of the current economic crisis in Europe on financial markets, it is hardly surprising that the EQ model has fared so well. The underperformance of the basic accruals model indicates that the StarMine EQ model’s more comprehensive view of earnings quality, incorporating Cash Flow and Operating Efficiency as well as Accruals, is of considerable value. Val-Mo also has performed solidly globally, but has demonstrated particular strength when it comes to generating outperformance in Developed Asia ex-Japan, where the model has produced a decile spread of 48% in the trailing 12 months.
It is also interesting to note that despite the often-heard comments that analyst revisions models just don’t work as well as they used to, we find signals tied to analyst revisions have done a good job of contributing to outperformance over the last year. An investor using either the StarMine Analyst Revisions Model or building a basic revisions signal from the I/B/E/S estimates would have generated positive returns in the trailing 12 months in all regions. As quant factors contribute to underperformance or outperformance over time, and fluctuate in their ability to generate alpha, analyst revisions seem to have reverted to more ‘normal’ performance after the difficulty of 2008 and 2009. StarMine Smart Holdings, which provides a unique perspective on utilizing ownership data, has also performed solidly in all regions.
The results depicted above also dispel another commonly held notion that models, once they become commercially available, are rapidly exploited by the broad market and become less useful tools. Two of our best performing signals, ARM and EQ, are among those that have been available for longest (since 2007 and 2005, respectively). No factor works all the time, but it is clear that StarMine quant signals offer significant advantages over the commonly used alternatives in all global markets. Investors would be well advised to consider StarMine signals for their portfolio construction process.
