Quantitative Investing – Utilities Strategy

Today’s post will cover the second half of the Combined Consumer Staples/Utilities Strategy presented in the book What Works On Wall Street. The first part was covered in my last post. This strategy has one of the best risk adjusted returns of all the strategies tested and also has the lowest downside risk. The utilities portion of this strategy offers compound returns of 16%, a sharpe ratio of 0.76, and a drawdown of 33%. With strong compound returns, lower risk, and reasonably high base rates this strategy should be at the top for more conservative investors interested in adding quant strategies to their portfolio. Lets jump right in.

The Utilities Strategy takes a little more work to implement than the consumer staples strategy. It is a pure value strategy like the consumer staples strategy with the difference being that it uses a composite of value factors as its screening criteria. One of the biggest insights from What Works On Wall Street and other quant researchers is that choosing stocks based on a composite of value factors outperforms portfolios based on solely one factor over time. For example, in the previous edition of What Works On Wall Street Price to Sales was the top performing value factor. In the latest edition EV/EBITDA was the top performing value factor. Just like strategies different value factors outperform in different time periods. Also, it is not optimal to rank companies with different capital structures based on the same value metric. For example, P/E ratios do not work well for REITs and insurance companies. The Utilities Strategy uses a combination of value factors called Value Composite 2. It sums up the ranking of companies based on P/E, P/B, P/FCF (free cash flow), EV/EBITDA, and SHY (shareholder yield). It then chooses the top 25 companies and invests in an equal weighted portfolio for one year and then re-balances. Its important to learn how to calculate these composite rankings so lets look at that in some more detail.

To calculate the composite value factor you start by ranking the companies based on each individual factor. Normally, you need fancy financial software to do this but excel works well enough. You just need a screener that allows excel downloads. Both screeners I have discussed, FINVIZ and Stock Investor Pro, have this feature. For each factor you can use the PERCENTRANK function in excel. Starting with P/E ratio you apply the PERCENTRANK function which ranks the stocks by P/E from 0 to 100. Low numbers are better in this case. You do the same for P/B, P/FCF, and EV/EBITDA. For SHY you need to do 1-PERCENTRANK since high numbers are better. After you have all the ranks you sum them up into a total rank and sort from lowest to highest. Done. Now, there is some devil in the details. Not all stocks have data in all fields. For missing data, you assign it a neutral rank of 50 so as not to unfairly bias the results against the stock. If you are using FINVIZ there are no fields for SHY or EV/EBITDA. You would ignore EV/EBITDA and substitute dividend yield for SHY. I don’t know how much this would change the return results but the underlying principle is the same. You’re sorting for the cheapest stocks based on a combination of traditional value factors. Below are the results of the Utilities screen I ran using Stock Investor Pro with data as of June 10.

Utilities Screen June 10 2013

Utilities Screen June 10 2013

Note the the 25th stock on the list NVE received a buyout offer from Berkshire Hathaway and is trading within 90% of the buyout price so we exclude it and add the next stock on the list VVC. Also, in the table I listed each value factor used in the screen, the total rank for each stock, and the dividend yield. Besides being conservative strategies with great returns the consumer staples and utilities sectors also offer investors nice yields which makes sticking with these strategies even easier. The utilities portfolio here has a dividend yield of 3.7% and the consumer staples portfolio I discussed last time has a yield of 2.6% (3.9% if dividend yield vs shareholder yield is used). A combined portfolio would offer the investor a yield of 3.8%. The combined strategy is a powerful combination. The table below summarizes the stats. Note that the benchmark used, All Stocks, is pretty darned good and has outpeformed the S&P500 (which is a large cap strategy).

Combine consumer staples utilites strategy stats

There you have it. After an initial learning curve you can now implement a quantitative value strategy that has been show to crush the market over time with superior risk adjusted returns, high base rates, and reasonable drawdowns for literally a few hours work per year.

Full Disclaimer - Nothing on this site should ever be considered advice, research or the invitation to buy or sell securities. These are my personal opinions only.

28 thoughts on “Quantitative Investing – Utilities Strategy

  1. I’ve been wanting to learn more about quant investing, great post, thanks! I have a few questions though.

    1) These returns are only based on fitting to historical data; will you keep us up to date on how this portfolio performs? I made it on FINVIZ, but not sure if it includes reinvested dividends.

    2) Since the S&P average return is within one standard deviation of this strategy, how can we say this gives better returns?

    3) What method do you use to see if any of the 6 rules from page 56 are violated (e.g., company takeover offer >95% share price)?

    1. Hi Jeff,

      1. The tracker on FINVIZ does not include dividends. But that’s easy. I just keep track of the beginning dividend yield and add that to the price returns at the end of the one year period. I will publish the results after the one year period for the strategies I’ve discussed.
      2. This is a large topic addressed in detail in the book. O’Shaughnessy goes to quite some effort to make sure these are not results due to chance. The key stat for each strategy to look at is the t-stat. A t-stat tells you the likelihood that the results are due to random chance. A t-stat above 2 says the results are statistically significant, at a value of 2 the results are not random with a 95% confidence interval. More in the book.
      3. I use the tracker at FINVIZ. Before I invest in the portfolio I go through the recent news from the company, after I invest I just check in every so often, maybe once a month.

      Paul

  2. Great post Paul. Would be very interested to learn about your entry point/price in the market to implement this outstanding strategy. I’d really like to follow your lead. Thank-you sir!

    1. Hi Derrick,

      Basically, doesn’t matter over the long term. The strategies don’t look at market valuation to determine when to buy.

      Paul

  3. Paul,

    Thank you for the well done explanation of the strategies outlined in this book. Have you (or do you intend to) started implementing any of these strategies?

    Hope you enjoy your time here in Oregon.

    Steve

    1. Steve, yes I am implementing 2 of the strategies, the Combined Consumer Staples/Utilities strategy and the Trending Value strategy.

      Paul

  4. Paul.
    The high returns generated from these conservative quantitative strategies are impressive, to say the least. What percentage of your overall portfolio do you recommend investing in these? Given the negative real returns on cash, should one go “all-in” for the long term?

    1. Hi Stan, I would never go all in on any strategy. As I did, I would recommend starting out buy learning to run the screens and tracking the results for a year before committing any real money. Then I would start with a small allocation to quant strategies, say 10%, and see how it goes. Then increase from there if you are satisfied with the results. Personally, I’m up to about 20% allocated to quant strategies and will increase it starting next year. Investing is a marathon, not a sprint. No need to rush into it.

      If you really wanted to get started quick I’d recommend a less than 5% allocation to one quant strategy after you learn the screening process.

      Paul

  5. Paul

    About 1/2 way into WWOWS. Was looking at FINVIZ, is there method to get the 6 month performance as a column on the Custom screen or do you do 2 exports?

    Like your west coast swing too, makes the Texas heat seem cooler!

    1. Scott, yes there is a 6 month performance column in the FINVIZ screener. Its titled, “Performance (Half Year)”. Just add it to your custom view and then you just need to do one export.

      The weather out here on the OR coast is great.

      Paul

  6. Paul,

    For the screener, aren’t values usually missing because they are zero or negative? Wouldn’t putting in a neutral rank of 50 bias the results?

    Thanks,
    Jeff

    1. Jeff, No I don’t think so. Not putting a netural rank would bias the results. Think about throwing out stocks that have a zero yield. The other screen parameters are all ratios based on price. The only way they can be zero is if the price of the stock is zero. The only way they are undefined is if the denominator is negative – at least that’s the way screeners tend to handle negative ratios. But there are good reasons why companies can have negative earnings, or free cash flow, but still be considered for the screener. For examples, companies like REITs or others that have high non cash charges may have a great EV/EBITDA ratio but have a negative or undefined PE. Leaving them out of the screen for an undefined PE would not be the right thing to do.

      Also, in general, a neutral rank of 50 for one parameter is not enough to drive a company into the top 25 in a given screen. It takes strong metrics in several ratios to make the list. Early on I was throwing out companies due to undefined ratios and it led to worse results so I think the O’Shaugnessy method is right here. You can try your screens both ways, if you’re up for it, and see the difference in the results.

      Paul

      1. I mostly had the negative denominators in mind, but good points about companies like REITs and the other factors needing to drive the company into the top 25.

        Another subtlety is that if I am not mistaken, O’Shaughnessy’s VC2 uses price-cash flow and not P/FCF. I can only use FCF with Finviz, but you might want to use cash flow since Stock Investor Pro has it.

        Cheers,
        Jeff

        1. Jeff, yeah P/CF or P/FCF. I get somewhat conflicting messages from the book. I’ve tried both and it doesnt seem to make to much difference. I’ve switched to P/CF in Stock Investor. Fundamentally, I prefer screening on P/CF since the the CAPEX I really care to measure is maintenance CAPEX and not growth CAPEX. P/FCF could penalize companies with large growth CAPEX. P/CF is closer to the famous Buffett measure of ‘Owner Earnings’.

          Paul

      2. Sorry to come back to this, but shouldn’t undefined ratios due to a negative denominator be ranked last for that ratio and not neutral? With P/E for example, a company with earnings of $1 will have a very high ratio, but its P/E rank should still be above a company with negative earnings. Then the company will need the other metrics to be exceptional to bring it up from the bottom rather than the middle. O’Shaughnessy does say to rank them neutrally though…

        1. Jeff, for all ratios (except shareholder yield and div yield) I rank negative ratios neutrally. Most of the time you don’t have a choice. If the denominator is negative, most screeners will have no value (blank) for that ratio. Div can’t be negative but shareholder yield can and should be ranked with negative numbers I think.

          Paul

      3. Paul,

        In regards to:
        > If the denominator is negative, most screeners will have no value (blank) for that ratio.

        Part of this script functions with the ability to grab each data point independently, i.e. it will grab price and earnings data, rather than P/E. To avoid any errors that would occur when you encounter a zero earnings value, it’s simpler to rank the E/P of the asset. Now you’ll have a simple scale with no blank data and the highest number representing the “best” E/P out there.

        This process is easily repeated for the rest of the metrics.

        Regards,

        Justin

        1. Got it. That’s a great way to do it. I can do that in Stock Investor Pro as well with custom fields. The way I’ve been dealing with it is to use what I call the ‘mother ratio’, EV/EBITDA, to sort out the negative values. I can make a case for a company with negative earnings but positive EBITDA to be included in the screens without being penalized but a company with negative EBITDA is truly weak independent of corporate structure. So what I do is leave all ratios as-is, ranking blank fields neutral, then calculate EV/EBITDA in a custom field directly from the financial data thus eliminating any blank fields. Then the companies with negative EV/EBITDA ratios simply get deleted from the screen results.

          Paul

  7. I agree div can’t be negative and because of BBY, I allow SHY to be negative. I have been modifying the matlab code from reddit introduced by nicolamr’s comment. For P/E, P/B, P/CF, and EV/EBITDA, the code currently assigns missing values to 100000, so they get ranked last. I was going to change this to rank them neutrally because of O’Shaughnessy, but then I began to question it. I think the missing P/E, P/B, P/CF, or EV/EBITDA values are because of a negative denominator and should be ranked below a high ratio from a slightly positive denominator rather than in the middle. The fact that you also rank them in the middle makes me feel like I’m missing something because you and O’Shaughnessy are more knowledgeable than me.

    1. Hi Jeff. Author of the matlab code here. I struggled with the “neutral vs. zero” debate for a while and may again switch it up. I wish I had Compustat and CRSP databases to backtest against, but alas I don’t have that kind of cash. Those with missing data points do often make the top 25 even with the artificial 100000 value assigned, so moving it to a middling number would only further put them into the top tier. By being pessimistic with this mystery number, it allows us to hold our choice stocks to an even higher standard. A value of 50 artificially assigned is almost like cheating; right now a 90th percentile stock will have a score in the 410-420 range (out of 600). A freebie 50 accounts for a lot of that.

      The justification for neutral in the book wasn’t convincing, so I’m not sure that either method is “wrong;” we still end up with a solid set of financially sound companies that the market is rallying behind.

      1. Hey Justin,

        You bring up a couple of great points. We’re all going to struggle with database issues, in particular since we’re not using the CSRP and Compustat databases used in the book. This is just part of the challenge. But as you point out, as long as we stay true to the original intent of the screen, for trending value that’s cheap stocks on the mend, then the overall results should be similar.

        This also allows to make the screens our own, taking into account our personal biases and preferences. For example, I’ve been testing versions of the trending value screen which filter out any stocks with dividends less than zero. This gives me a dividend weighted screen of cheap companies with momentum.

        As far as blanks, I can see O’Shaughnessy’s argument for a neutral ranking if the blank is truly causes by missing data but not so much if its because of negative data.

        Paul

      2. Paul,
        I totally agree about not losing sight of the screen’s purpose, but I think these subtleties can make a big difference over time. It hasn’t been a very long sample period, but I screened and saved a portfolio on FINVIZ the same day as your June 14 screen and mine has consistently under-performed yours by 2-4%.

        Justin,
        I really appreciate you posting this code since I don’t have enough capital to justify buying Stock Investor Pro and I didn’t know how to use MATLAB to pull data from the internet. Could you upload the script for individiually calculating the metrics so there are no blanks? If not, I can implement it in my version. I started with the script from:
        http://www.mathworks.com/matlabcentral/fileexchange/41940
        Then I changed it from +Small (over $300mln) to +Micro (over $50mln) and delete anything under $200mln. I also changed it from P/FCF to P/CF.

        Thanks to both of you for the added insight.

  8. Is there a big difference in doing the full screen for the utilities / consumer staples and
    then buy individual stocks that meet the screen and just buy the ETF XLU and XLP ?

    GREAT WORK

    1. John, there is huge difference. That is the point of the whole strategy – to use a value screen to improve the performance of the sector only results. For example, as documented in What Works on Wall Street, the Utilities value screen returned 16% a year from 1967-2009 vs 11.25% a year for the XLU sector.

      Paul

Comments are closed.