A few months back, we published a note tilted, “MACRO MEETS MICRO: ARE YOU HUNTING WHERE THE FISH ARE IN THE US EQUITY MARKET?” in which we introduced our US equity market screening application.
The model is designed to triangulate sell-side, buy-side and insider sentiment on a particular stock, recording deviations in those metrics (from the sample) in order to systematically produce a list of names that may be under or over-owned; we consider that to be a good starting point to dig into the long or short side of a particular ticker(s) from a bottom-up perspective.
In the months since publishing that note, we’ve greatly expanded upon the model, sourcing valuable feedback from our institutional client base. Noteworthy changes include:
- Expanding the sample of tickers from the S&P 500 to the Russell 3000 Index, with the latter representing ~98% of publically-traded equity market cap;
- Tracking MoM deltas to see which tickers moved the most with respect to the scores assigned to it from the model, as big moves in sentiment over a reasonably short time frame may signal a meaningful change in the underlying fundamental trends; and
- Adding in-depth summary statistics to the backtest results.
Please CLICK HERE to download the Excel file containing the latest refresh of the model and its summary results. For more information on how the model works, the associated backtest statistics or strategies for how best to apply it to your specific investment process or mandate, please review the following three sections.
***Email us if this is something you’d like to see more often or to have the sample customized for a specific coverage universe.***
1. HOW THE MODEL WORKS
The model is setup as a quantitative screen that assigns +/-1 point for a specific ticker's reading being in excess of [1x] standard deviations from the sample mean, for each of the following three metrics:
- Sell-side Sentiment: Bloomberg Consensus Ratings (1-5 point scale): -1pt if the figure is in excess of +1x standard deviations relative to the mean of the broader sample; +1pt if the figure is less than -1x standard deviations relative to the mean of the broader sample; 0pts if within [1x] standard deviations of the mean.
- Buy-side Sentiment: Short Interest as a % of Float: +1pt if the figure is in excess of +1x standard deviations relative to the mean of the broader sample; -1pt if the figure is less than -1x standard deviations relative to the mean of the broader sample; 0pts if within [1x] standard deviations of the mean. We cap the readings at 15% to dampen the variance of the sample in order to produce a more robust collection of deviations.
- Insider Sentiment: 6M % Change of Insider Ownership: +1pt if the figure is in excess of +1x standard deviations relative to the mean of the broader sample; -1pt if the figure is less than -1x standard deviations relative to the mean of the broader sample; 0pts if within [1x] standard deviations of the mean. We cap the readings at +100% to dampen the variance of the sample in order to produce a more robust collection of deviations.
The model produces scores for each ticker ranging from -3 to +3 on an integer scale. For example, a reading of +3 signals extreme negative sentiment amongst both the sell-side (+1) and buy-side (+1) amid extreme insider buying (+1), while a reading of -3 would signal extreme positive sentiment amongst both the sell-side (-1) and buy-side (-1) amid extreme insider selling (-1).
2. BACKTEST RESULTS
Interestingly enough, the model was initially setup as a contrarian screening tool and the -3 to +3 scale was setup with the assumption that names the model signaled as under or over-owned would subsequently mean revert from a performance perspective.
What we learned in the backtesting phase, however, is that names which “the market” loved the most were typically rewarded with subsequent outperformance of considerable magnitude over the intermediate-to-long term (as defined by 1Y forward returns). Moreover, names that the market disliked the most were typically punished with subsequent substantial underperformance. Furthermore, the hit rate of positive absolute performance is substantially higher for names which "the market" loved the most that it is for names in which "the market" greatly disliked.
Additionally, names that the market “learned to love” in a short time (as defined by its score dropping -2pts over the span of one month) generally outperformed those that the marked “learned to hate” (as defined by its score growing +2 over the span of one month).
Using these results, we have decided to assign any ticker that scores a -3 or -2 as a compelling LONG idea and any ticker that scores as a +2 as a compelling SHORT idea going forward.
In summary, there appears to be some element of efficient markets present in these backtesting results. While this is certainly not the proper forum to debate the merits of the Efficient Market Hypothesis, we can reasonably conclude that “the market” itself can be a reliable tool for identifying good opportunities on both the long and short side.
It also lends support to our belief that momentum exists in the market place over the immediate term, intermediate term and long term durations, which is precisely why we use a three-factor quantitative risk management model to identify key levels of breakout/breakdown risk in specific securities and/or asset classes.
Lastly, the only potential draw-back to interpreting these results at face value is the lack of historical data beyond MAR ’10 (that’s as far as Bloomberg’s insider ownership change data goes back). While the mini-crises of summer 2010 and 2011 are certainly included in the sample, we’d ideally like to see how the model would perform across multiple market cycles (i.e. 2007-09 would be a good place to start).
All that being said, using end-of-month observations for each ticker in the Russell 3000 Index, we were able to produce 86,976 subsequent 1Y forward return data points to analyze spanning MAR ’10 to OCT ‘12. We’d argue that’s plenty robust enough of a sample to work with.
3. USING THE TOOL EFFECTIVELY
Offhand, we think there are two types of primary users for this tool:
- Fundamental Analysts: Sift through our list of long and short ideas to find names that you think are either cheap or expensive on your preferred metric(s), or those that have some industry-specific tailwinds or headwinds that you think will either benefit or hurt a particular name(s). We’re giving you names we think will work anyway, so marrying your own bottom-up analysis with our top-down analysis may yield superior performance results.
- Quants: Use our list of long and short ideas to run your own proprietary factoring on; again, we’re giving you names we think will work anyway, so layering your own proprietary factoring on top of our analysis may yield superior performance results. Moreover, we’re including our US Equity Market Style Factor Divergence Monitor to the extent you see a particular factor outperforming on a trending basis and you (or we) feel it is appropriate to chase or fade it. In the Excel download above, we’ve included all of the relevant style factor characteristics for each of the individual tickers; please note that the data is presented on a percentile basis to make it easier to determine how much a specific ticker is weighted to a given factor.
Lastly, all interested parties should feel free to ping us for our proprietary risk management levels on any given ticker(s). As you know, we are hyper-focused on timing, so allowing us to flag names that are breaking out or breaking down from an immediate-term TRADE or intermediate-term TREND perspective should only enhance the probability of generating alpha from this process.
Associate: Macro Team