Using models and/or screens to find potential investments is an efficient method, especially if those models have been backtested and are based on solid logic. However it is important to highlight that there are pitfalls to using black boxes. The biggest issue in my opinion is incorrect data. This article will discuss some of the advantages and disadvantages of using screens, as well as provide several examples where incorrect data was found by the author. While this may sound obvious, I don't believe enough investors take note of this issue. As a result, these problems could lead an investor to a wrong conclusion and investment.
There are several advantages to using screens or models such as O'Shaughnessy's Value Composite (see my article here). These include:
- The model selection on a stand-alone basis has historically outperformed the market. So basically, if you had blindly followed the screens recommendations you would have made $$$.
- The models provide a framework and focus to the qualitative analysis. There is a clear definition of what is value and this definition does not shift based on manager preferences. For example, a manager may define value as a low P/E in one instance and in another a low PEG ratio. This in reality is a conflicting. A model or screen is strict and thus is very disciplined.
Personally, I am against blindly following any model or screen. I believe that by doing the qualitative analysis (aka bottoms-up fundamental analysis) an investor can generate superior returns with lower risk compared to both the market and the model. This is achieved by understanding the 'story' (i.e. knowing what has caused the drop in stock price and what factors have and will affect it). In my experience of using models, my fundamental analysis has helped produce better results (i.e. in 2014 my main value model generated 4% on a stand-alone basis vs my 30% return).
Many may disagree with my view but where there can be no disagreement are problems models can have. Some that I have encountered are:
For example, when I exported my data from Bloomberg and looked at my holdings I noticed that Anthem (NYSE:ANTM) had a lot more shares than previously. Specifically, ~297m shares vs ~270m that I believed existed. So I looked at the filings and then inquired about this with Bloomberg. Their response:
Thank you for your question regarding [ANTM] and the Shares Outstanding value for the most current period.
...we have realized that there was an error in our data and the correct number has been reflected on FA. We are very sorry... In this case for [ANTM] the data was not updated from the correct source...
...We will do our best to ensure this error does not happen again. Thank you for your understanding and patience, and thank you for using Bloomberg Help!
So while Bloomberg did have the correct market cap, the FA screen and export field had an incorrect number of shares. There appears to be some glitch in their system. This was also the case with Green Plains (NASDAQ:GPRE).
Another example, is trailing 12M EBITDA for foreign companies. This was something I noticed in the past but was also confirmed recently by a fund manager friend of mine who invests globally. If you use the EBITDA figure in USD as translated by Bloomberg then you will get the wrong result. This is because Bloomberg translates the EBITDA for each quarter based on the FX at the end of the quarter. So the work around is to export in foreign currency and then translate with current FX to get the correct USD amount.
Share class per share problems:
If you export per share data via a Bloomberg screen on Berkshire Hathaway, Class B (NYSE:BRK.B), you get the per share data for Class A. To correct for this I have to multiple the share price by 1500. If I blindly followed my models, I would have been loading up on BRK.B which had a P/E of 0 due to the large denominator.
Non-operating/Non re-occurring income:
Domtar (NYSE:UFS), EPS before extraordinaries is stated as over $6 per share so it looks like the stock has a very low P/E ratio. The truth is that half of that EPS is due to a deferred tax benefit. Take that out of the equation and P/E is double what you thought.
Corporate action implications:
A company spins-off or sells a large chunk of its operations however the data still reflects the entire company. This may not be the fault of the provider. There are several examples such as these. One such positions that I experienced Cash America (NYSE:CSH). Even after CSH spinned off Enova, which was about 1/2 its revenues, the data providers showed CSH data as if the spin-off never occurred.
A well known recent example is Radioshack (NYSE:RSH). I rejected at $4 after fundamental analysis determined it was a trap. Other more obvious examples are:
You may be able to better time your entry by putting your potential investment on a watch list. When Domtar Corp came to my attention it was in free fall. By watching it, a better entry was achieved in the thirties. This gave me a margin of safety and an easier profitable exit.
High concentrations in a sector:
If someone today blindly followed a value screen then they could very likely end up with a ton of oil related stocks. A high concentration would be very risky. The positive side of this is that if the model gives many stocks in the same sector/industry then there could be an opportunity. This is thus brought to the investors attention for investigation. For example, in the past, health care stocks and technology distributors were constant opportunities given by my models. Once doing my own fundamental research I found out that they were indeed cheap and profited from this. But at the same time limiting their % in the portfolio to a certain maximum level.
These are some of the issues I have faced and it is why I prefer to have all the raw input data in my spreadsheet. If you uses a product or tool to screen for stocks without checking up on the numbers then you could be at risk. It is important that whatever service you use, that the input data can be fed to you to check, otherwise you are at its mercy. One way to help reduce the problem is to use two tools but each must use a different data source or process the data differently.
A source of mine who works in pricing at one of the large market platforms has told me they have all platforms (i.e. Bloomberg, CapitalIQ, Reuters, etc) running side by side so that they can make sure their pricing isn't too far off. If it's far off then its an alarm for them to check out for errors.
I hope this article is food for thought and helps you avoid some costly mistakes.
Disclosure: The author is long ANTM.
The author wrote this article themselves, and it expresses their own opinions. The author is not receiving compensation for it (other than from Seeking Alpha). The author has no business relationship with any company whose stock is mentioned in this article.