We’ve launched a blog, which you can access here: blog.portfolio123.com. (In the near future there will be a link to the blog at the top of the page.)
I will be posting at least one new article a week (all the articles currently on the blog have already been published on Seeking Alpha, but a new one will come out this week). Occasionally these articles will be revisions of articles I wrote a little while ago, but I hope to keep the content fresh by publishing mostly new pieces. There will also be articles written by others. If you would like to contribute an article to the Portfolio123 blog, please e-mail me at yuval@portfolio123.com.
The blog articles are designed for a general audience and are meant to interest outsiders in our website. They are not for intricate discussions of Portfolio123 features. Those discussions belong in the forum.
If you have any suggestions for subjects to cover, I’d welcome them.
Yuval,
Your latest blog “The Magic of Combination: How Mixing Strategies Can Improve Results” refers.
Can you please put the 16 factor system up as a DM so we can follow out of sample performance. As we now know, backtest performance does not indicate future performance as proven by the poor historic 5-year performance of the 75 DMs which are at least 5 years old. And it’s not just a function of members being unfamiliar with the P123 platform either, the 9 DMs from the P123 Team members have performed no better.
Your conclusions in the blog are based on backtest results and may not be valid out of sample. One will have to watch this for at least 5 years to validate the inference of your blog.
The 16-factor system I proposed was one I pretty much made up off the top of my head to show how combining seemingly incompatible strategies improves results. It is not a system that I am going to use for investing, nor one that I would suggest that others use for investing. It is therefore not appropriate for a designer model. Designer models should ideally be all-weather, well-thought-out, tried-and-true models that the designer really believes in and that the designer is willing to put his or her own money behind. The system I proposed was proposed for the sake of argument. It was not backtested beforehand and for all I know the system might be better without some of the sixteen factors; it certainly could be improved by adding additional factors.
The piece was written in the hope that people would try taking their OWN seemingly incompatible strategies and combine them by averaging the weights of their factors in one more comprehensive system. I believe that people will get better results by doing so because of the way our ranking systems work.
I do plan to offer a new designer model at some point in the next few months, but it’s going to take me a while to develop it.
Dear Yuval
I love your work, and I was looking forward to finding time to test “The Magic of Combination” system in the Seeking Alpha article. I now have, as best I can decipher it. I get totally different results from you.
I have 4 factors more or less matching yours in each of four main nodes in P123. I used the last 10 years to the date of the article (9 Dec 2019) (with 0 slippage, so I should get better results). I used the Russell 3000 stocks. I chose the best 25 stocks. I get 12.6% annualised return with 29% max drawdown.
I have used P123 for ages but haven’t trawled through forums to learn much from others - just tried to self teach. I am sure you are busy but can you please review my ranking system to see where I went so wrong? https://www.portfolio123.com/app/ranking-system/366775
I think you’re on the right track. Here are a few suggestions for improvement.
With value factors it might be better to use industry rather than universe comparisons. You also might want to use forward earnings yield rather than PE, CurFYEPSMean/Price or NextFYEPSMean/Price.
With dividends, you probably want to add something about the payout ratio so that you’re not buying stocks whose dividends are greater than their income or cash flow.
You wouldn’t want to use R&D on its own because you’re favoring large rather than small companies. Instead compare it with market cap, sales, assets, etc.
The Industry=Health or Tech node should be Boolean.
Sales3YCGr% is not a good quality measure. The market tends to punish companies with super-high sales growth. Neither is the debt-to-equity ratio. Free cash flow per share is not a quality measure either because it’s sensitive to the number of shares outstanding. I would suggest looking at accruals, debt to EBITDA or debt to cash flow, free cash flow margin or free cash flow to assets, and asset turnover.
The screen shouldn’t exclude all companies in all those industries. Instead you can put those industries into your R&D node with an eval command and make them N/A, then use N/As neutral in your ranking system. Or use a conditional node to give those companies NA values for R&D. The idea is not to find companies that are doing well in ALL these ways but instead to have a broad mix of companies. If you exclude those industries you’ll exclude most of the dividend growth stocks.
Also, I think you should try to make each of the four individual strategies work on their own before combining them. Each of them should be written to produce somewhat strong returns.
Could you clarify what you mean by forward earnings yield though? Sorry for being a bit dim but what’s that? My understanding is that EarnYield is just the inverse of PE and the only ways that have occurred to me to calculate future earnings ratios are with CurFYEPSMean/Price and NextFYEPSMean/Price.
With regards to accruals as quality factors, are they supposed to be high or low? Are there any particular characteristics one should look for with them? Turnover? Stability?
Sterling, it’s a case of a misplaced comma–my apologies. What I meant was that instead of using PE, use forward earnings yield, which is indeed CurFYEPSMean/Price or NextFYEPSMean/Price. As for accruals, they’re supposed to be low. There are two very different ways of calculating accruals. One is to subtract operating cash flow from net income, the other is to look at the increase in net operating assets. Both of them should, ideally, be negative. But accruals isn’t a stand-alone factor. First, you have to have a ratio–I divide accruals by total assets, but some folks divide by net operating assets, by market cap, or by earnings. Second, it should only be used in combination with other factors; it’ll give you poor results if you just make it a ranking system on its own.