Models and Methodologies

** Updated August 20, 2013 **

This post is Part 1 of a series on Valuation Metrics Technology and the mathematics behind it.

Valuation Metrics uses a set of finely tuned and backtested proprietary algorithms to determine the strategies employed by individual funds, and to identify companies that fit best with those strategies. The technology helps companies target the funds that are most likely to buy their stock, while avoiding unproductive meetings with funds that are unlikely to be interested in them. It also helps funds identify new stocks that may fit well with their particular strategy, as well as currently held positions that may be candidates for sale.

Model Descriptions

Each fund’s overall strategy is assessed by measuring the metrics of the fund’s holdings relative to the entire universe of stocks with respect to 29 descriptive, fundamental and technical models, defined as follows:

Model Components Funds Using Strategy(%)
Market Cap Market Capitalization 99.8%
Institutional Coverage Market Cap & Analyst Coverage 96.5%
Liquidity Ratio 10day Average Volume to Daily%∆Price Ratio 93.4%
Trailing P/E Ratio Trailing 12mth Price/Earnings Ratio 91.3%
P/CF Ratio Trailing 12mth Price/Cash Flow Ratio 90.3%
Share Turnover Share Turnover 89.2%
Forward P/E Ratio Forward 12mth Price/Earnings Ratio 87.6%
P/B Ratio Current Price/Book Ratio 86.4%
Historical Alpha Trailing 5yr Alpha 71.5%
Yield Current Dividend Yield 70.4%
P/S Ratio Trailing 12mth Price/Sales Ratio 65.8%
PEG Ratio Forward 12mth P/E / Forward 5yr EPS Growth Rate 65.3%
Long-Term Growth Estimate 5yr Forward Estimated EPS Growth Rate 63.0%
EV/EBITDA Enterprise Value/Earnings B4 Int/Tax/Depr/Amort 62.8%
Cash Plowback Trailing 12mth Retained Earnings 61.9%
Relative Strength Model 12mth Price Appreciation 59.2%
Short-Term Revenue Growth Average of Recent and Estimated Revenue Growth 55.0%
Debt/Equity Ratio Current Total Debt/Total Shareholder Equity Ratio 54.6%
Industry-Relative Momentum 12mth Price Appreciation Relative to Industry 49.1%
Historical Revenue Growth Trailing 5yr Revenue Compound Annual Growth Rate 47.9%
Historical EPS Growth Trailing 5yr EPS Compound Annual Growth Rate 42.0%
Dividend Discount Model Present Value of Future Cash Flows 40.4%
Short-Term EPS Growth Average of Recent and Estimated EPS Growth 39.9%
Industry Momentum 12mth Industry Price Appreciation 37.2%
Estimate Revisions Model Change in Number of Analyst Upgrades/Downgrades 30.8%
Revenue Momentum Average of Recent and Estimated Revenue Momentum 18.4%
Margin Growth Trailing 5yr Profit Margin Compound Annual Growth Rate 16.0%
Earnings Momentum Average of Recent and Estimated EPS Momentum 11.6%
Estimate Changes Change in Annual Earnings Estimates 7.4%

The data for each model is standardized to facilitate ease of comparison between multiple variables and models, and truncated using multi-pass Winsorization techniques to reduce the adverse effect of outliers due to data anomalies.

For each fund, an “importance” rank is then calculated for each of the 29 models according to how likely it is that the fund is using that particular criterion as part of its overall strategy. The calculation of the ranking is done using a term from inferential statistics called a t-stat. In mathematical terms, a t-stat is defined as the difference between the sample mean and the population mean, divided by the standard deviation of the sampling distribution of the sample mean (aka the standard error of the mean). It sounds daunting, but the t-stat is really just a measure of the difference between two averages, expressed in terms of standard deviations.

When looking at samples of data, we never really know if the average of a sample is definitely different from the true average of the entire population – we have to infer it based on the data we have available. This is why we calculate such an arcane measure like a t-stat. The t-stat tells us how likely it is that the sample average is really different from the true population average. The higher the t-stat, the more likely it is that the sample average is indeed different. Inferential statistics are used to tell us just how likely, or how confident we are, that the averages are in fact different. We are 95% confident that the sample average is different from the true population average when the t-stat is above 2.0, which is the standard threshold used by statisticians to determine whether or not a result is statistically significant. The higher the t-stat, the greater the degree of confidence we have that the sample average is in fact different from the true population average.

Model Importance Ranks (t-stats)

As mentioned above, each of the Valuation Metrics models is measured using standardized data, which simply means that all measurements are divided by the standard deviation of the entire universe of stocks for each particular metric. By definition, the population average for each metric is always zero and the population standard deviation is one. The sample, or portfolio, average for each metric is then measured in terms of its respective population standard deviation.

Generally, the farther the portfolio average of a particular metric is from the overall universe average, and the lower the standard deviation of the portfolio, the higher the t-stat for that model will be. The number of holdings in the portfolio also has an effect on the t-stat- the more positions in a portfolio, the higher the t-stat will be. This is true because a higher number of holdings in a sample portfolio is more likely to give a more accurate estimate of the true average; hence it increases the confidence that a particular sample average is different from the true universe average.

The models are presented in the Fit With My Company chart from top to bottom according to their t-stats. The models at the top have the highest t-stats and are most likely to be used as a part of the fund’s overall investment strategy. It should be kept in mind that the model “importance” ranks (t-stats) do not actually measure how important the model is to a particular fund’s strategy, but rather the degree of confidence we have that the fund is pursuing a strategy with respect to that model, based on the confidence we have that the fund’s average for that model is different from the overall universe average.

By examining the t-stats for all the models, we can get a pretty good sense of the strategies that a particular fund is employing in their portfolio with respect to size, growth/value, quality, safety, liquidity, etc. Using this information, we can then look at the same model with respect to individual companies to determine which companies fit best with a particular fund and vice versa.

Match Scores

In a similar manner to how the importance ranks for each model are calculated, Valuation Metrics uses a proprietary ranking algorithm to determine how well a company fits with the overall strategy of a fund. This algorithm incorporates the input of all 29 models described above, weighted by the t-stats related to each of those models, as well as other factors related to things like industry classification. A Match Score ranking on a scale of 1 to 99 is computed for each company with respect to each fund. Based on that score, it is then placed in one of five Match category quintiles ranging from Very High to Outlier.

Extensive backtesting, as described in greater detail in Part 2 of the series on Valuation Metrics Technology, has shown that companies with Match Scores falling in the Very High category were 30 times more likely to be purchased by a fund than those that fell in the Outlier category. Companies with Match Scores in the High or Very High categories accounted for 84% of all new purchases made by funds. Those in the highest quintile (Very High category) accounted for 62% of all new purchases, with those in the highest percentile (99) accounting for 7% of all new purchases.

These results illustrate the true power of the Valuation Metrics technology to identify and bring funds together with companies that fit their particular strategy, in much the same way as dating services bring together couples with similar interests, and as physicians run diagnostic screening tests to identify patients with particular illnesses.