I was recently invited to visit the headquarter office of an investment research platform, whose research product I’ve been using for quite a while. I had a meeting with the product team, during which I walked through my research process and shared some feedback of their product.
During the Q&A section, one of the product manager asked this question: “How has your research process changed over time and how can we make sure that our product can create value for you even with the change in your process?”
It was a great question. I knew the way I conduct fundamental research has improved dramatically since I embarked upon the value investing journey, but I haven’t put it in writings. Today’s article is a reflection of the evolution I went through in my research process.
Phase One – Cigar Butts Hunter
It was back in 2011 that I officially began my value investing journey. I had read “The Intelligent Investor” and “One of on Wall Street”’ and the like. I was enthusiastic, confident and ready to make money right away. It seemed like the fastest way to pick the winners was Ben Graham’s net-net strategy and the low P/E low P/B strategy. Peter Lynch’s way of explaining the P/E ratio stuck with me – you can roughly think of the P/E ratio as how many years it would take you to recoup your money. Of course if you think that way, the lower the P/E the better. It didn’t take long for me to find a “perfect stock” – a Chinese reverse merger trading at approximately three times P/E and way below book value. I bought it with great joy, only to see the stock sink more than 20% rapidly. I panicked and sold.
That failure didn’t stop me from picking up cigar butts one at a time. For the first 2-3 years, I would screen for cheap stocks based on P/E and P/B and research them one by one. It was a relatively speedy process – I read the annual report to understand the business a little bit and run the numbers to see whether it was cheap or not. If it’s cheap, say below book value and 10x P/E, I’d have the urge to buy it.
Phase Two – High Quality Hunter
While I was having fun searching for cigar butts, I also started to read more and more about Warren Buffett and Charlie Munger. Sometimes I made money on cigar butts and sometimes I had significant paper losses on seemingly cheap stocks. It didn’t take too long before I found that the question I often asked myself was, "How do I know this piece of crap is not a value trap?” So it’s natural that I gravitated towards the high quality camp.
With the high quality companies, I would spend more time learning about the fundamentals of the business. But I was still looking at one company at a time. I studied Walmart (WMT, Financial), Coca-Cola (KO, Financial) and other well-known high quality companies. Depending on the company, but in general, a few days is more than enough to come up with a general moat analysis of a company. Again, I was analyzing one company at a time, but with more focus on quality than quantity.
Phase Three – Putting Literacy and Numeracy Together
It was in late 2014 or 2015 that my research process went through another dramatic change. I was recommending DaVita (DVA, Financial) and my boss asked for my model. Unfortunately, my model was very simple and didn’t have any of the key drivers to show how earnings and cash flow would change under different scenarios. Fortunately, one of my colleagues happened to build a model for DaVita himself and in his model, he drilled down to the business drivers such as patient growth, average reimbursement rate per dialysis (both commercial and government), commercial mix and so forth. And he tied all those assumptions to the cash flow and calculated the intrinsic value of the business using a very impressively detailed DCF model. I thought his model was magical because I could see that if reimbursement increased by X%, it would increase DaVita’s intrinsic value by Y. Every number in the model had some math logic in it. It was really something.
After the DaVita incident, I became a huge fan of the super-detailed DCF model. For every company I research, in addition to qualitative analysis, I would try to drill down to the bottom math layer and connect every cell in my Excel model to the DCF calculation.
The detailed DCF model has its merit – it shows exactly how value is created mathematically, assuming the assumptions you make are all reasonable. It also adds a lot of comfort because all the numbers in the model have some sort of mathematical logic to them.
But the problem with the detailed DCF model became very obvious soon after I got on board with it (believe me, I loved the model so much that sometimes I dreamed about the models I made).
There are a few issues, but the biggest I realized is that you have to make so many assumptions and sometimes an assumption is the derivative of another assumption. You can see how one bad assumption can compound the error in the calculation of the perceived “intrinsic value.” To use an analogy: If we were to project the GDP growth of the U.S., we could do it in the detailed way – that is, to break down the GDP components to the most fundamental level possible and then add them all up. But doing so would require us making thousands of assumptions, most of which require substantial judgment. I bet when you come up with the number, it would seem obviously wrong and then you’ll go back and tweak the assumptions until the GDP growth comes to between 2-4%. Then you say, OK, now it seems reasonable.
The second issue with detailed DCF is that doing it dramatically increases our commitment and consistency bias, the anchoring bias, as well as the reason-seeking bias. It’s a lot of work to get the model done, and once you put in all the work, you want to think it’s good. And the model gives you a precise number to as many decimal places as you want. You compare the DCF intrinsic value and the stock price and it’s an easy decision. I’ll use DaVita again as an example. When we did the DCF analysis, we came up with a value of more than $120 per share, and the stock was at the low $70s, call it $74. If we believed $120 was the precise intrinsic value, which we did, we came up with a substantial margin of safety.
What happened with DaVita? Well, we got the reimbursement assumptions all wrong, which was the biggest driver of the value. We also could catch the amount of patients who went through the American Kidney Fund to get commercial coverage in the model. We also got the assumptions on Healthcare Partners wrong. I wrote an article last year detailing our fiasco. I still get chills reading what I wrote.
The evolution from Phase 1 to Phase 2 was natural and it was easy. I owe a great amount of debt to my good friend and colleague for making the jump from Phase 2 to Phase 3 happen. It was transformational.
I’ll pause in my reflection now. In my next article I’ll write about the other few phases I went through in the last three years.
