AMD's stock performance during 2017-2018 is a perfect example of Mr. Market in action:
Its stock price fluctuated 1-3x, mostly on fear and speculation. There was very little material news that deviated from AMD management's stated expectations, the drivers for growth and risks were well known and documented (see below), and there were very few (if any) surprises along the way.
Clearly, the value of its business did not increase or decrease by such dramatic multiples. For the patient investor with a reasonable estimate of intrinsic value, there were many opportunities to profit from Mr. Market's mood swings.
The investment story behind AMD is pretty straightforward: Intel (INTC) dominates the x86 CPU space (especially in the datacenter and hyperscale market) and AMD finally has a viable competitor in its new Zen architecture. Datacenter is a growing market as companies migrate their infrastructure to the cloud. Combined with the fact that INTC has been struggling with its manufacturing process, it seems reasonable to assume that AMD will continue to gain market share. This advantage is likely to hold for at least the next few years based on AMD's roadmap with Zen 2.
Skeptics argue that AMD has a history of failing to translate short term technical superiority into long term advantage, a track record of poor profitability, slow product adoption from both data center and OEMs, and that recent earnings have been propped up by crypto-dependent GPU sales.
While the crypto fears are largely overblown, NVDIA (NVDA) is widely regarded as the superior option in the GPU space, INTC has made some key hires and is unlikely to sit still, and increased ARM exposure may finally make it a viable alternative to x86.
Even the most optimistic investor would have to admit that AMD faces many threats from well-funded companies in highly competitive markets, its future success rides primarily on one product line, and it has a narrow time window in which to press its advantage.
AMD was my first in-depth investment analysis, and I'm still working on refining my methodology, so take anything below with a grain of salt. Currently, my process is:
- Understand the story. Make sure that I understand the business and that the prospects seem compelling (see above).
- Project earnings for 1-2 years out using realistic and independent estimates
- Apply a multiple to those earnings to arrive at a valuation
- To correct for what are likely overoptimistic biases, I weight the valuation in (2) with a downside estimate and probability
- Sanity check my valuation using Prof. Aswath Damodaran's equity valuation models
I built a simple revenue and profit model that contained revenue and profit projections for the following segments:
- GPU (Non-Blockchain)
- GPU (Blockchain)
- Enterprise, Embedded & Semi-custom (Non-EPYC)
- Enterprise, Embedded & Semi-custom (EPYC)
These segments are not explicitly spelled out, so some educated guess work was involved. For instance, I estimated AMD's blockchain exposure based on their baseline revenue before the blockchain surge, publicly available estimates of GPU blockchain sales, AMD's presumed crypto market share, and management estimates. These estimates were not perfect, but the numbers clearly dispelled fears that AMD would suddenly collapse along with crypto, even assuming exposure that far exceeded management estimates.
CPU estimates were broken down into desktop and notebook, using readily available market share data and Intel quarterly filings that break out revenue by segment. Discounts were applied to ASP and margins compared to Intel.
EPYC server numbers are harder to come by, but I went with management estimates as a baseline, and looked at historic market shares (such as the Opteron era) to see what type of growth trajectory, peak market share and margins were realistic. Given that the hyperscale market only consists of a few players, it was also possible to monitor various announcements to help validate progress on a high level.
I classify stock options as an expense, and generally ignore non-GAAP earnings (with the exception sometimes of credible one-time charges for things like restructuring, short term changes in fair value on investments, and amortization of intangible assets). A substantial degree of dilution is assumed from the exercise of warrants for 75MM shares to West Coast Hitech L.P. and convertible debt over the next few years.
All estimates erred on the side of being conservative (but also realistic).
Combining all of the above, I arrived at the following EPS estimates:
Figuring out an appropriate PE multiple is challenging, and this is a process that I am still trying to figure out. Some general thoughts and observations, which may be completely naive and incorrect, but seem to make sense:
- PE ratios are inversely related to interest rates. People have competing alternatives for investments: if the risk free interest rate is high, then stocks become relatively less attractive.
- In general, the logical "cap" on PE should be the inverse of the risk free rate. i.e. if the current 10-year treasury rate is 2.71%, then 1/.0027 = 36.9. Otherwise, I could earn more (with less risk) from treasuries.
- Of course, what matters is not what earnings or interest rates are now, but what they will be in the future.
- As anyone who has run a business knows, it is near impossible to predict with much accuracy what earnings will be 1-year out, much less 10-years out.
- Interest rates are even harder to predict.
So where does that leave us? Does history offer a guide?
- The average PE ratio over the last 140 years has been around 15.
- The median CAPE ratio has been around 16.
- However, Prof. Jeremy Siegel makes some convincing arguments that the long term sustainable PE ratio should be slightly higher, around 18-20, due to lowered transaction costs (of ~1.5%).
- The median 10 year treasury rate over the last 150 years is 3.85%, which implies a PE of 26. This is inline with the observations above.
So a PE range of 15-20 seems to make historical sense. Where does an individual company fall within the range? Peter Lynch's PEG ratio is a convenient, quick heuristic, but it doesn't take into account quality or sustainability of earnings. Other factors one might want to look at are ROIC, strength of balance sheet, quality of management, sustainability of growth, etc - but this feels like more art than science, and a certain degree of subjectivity seems required.
Ultimately, I decided to go with a 20x multiple on 2020 estimated EPS, using a discount rate of 5%, which resulted in a (present value) price per share of $17.80. The reason I decided to go with 2020 EPS instead of 2019 EPS, despite the increased uncertainty of longer time horizons, is that AMD's long term roadmap is strong, especially compared to Intel, and later years are where I expect most of the value to be created.
Correcting for Overoptimism
It is well documented that analyst target price forecasts are optimistic on average and that most people suffer from optimism bias. To try to correct for this, I ran a similar model with more pessimistic (but still realistic) assumptions. For example, what happens if AMD is only able to take 10% EPYC market share by 2020 instead of 17.5%? What happens if they can't ramp up OEM notebook support?
Using the same calculations as above, I arrived at a $13.60 2020 price target in the pessimistic scenario. I also assigned a 15% chance that this would happen - small, because I think their near term advantages are strong, but not non-existent.
Damodaran Valuation Model
As a final sanity check, I run a valuation using Prof. Aswath Damodaran's fcffginzu all-in-one valuation model spreadsheet. There are instructions and an overview on the linked page, as well as various instructive blog posts of him utilizing the spreadsheet to value some well known companies (GOOGL, FB, AMZN and NFLX).
His models are useful because sometimes the valuations I get are wildly different from my own, and it may cause me to rethink some assumptions (usually, about the quality of earnings growth).
In this case, I arrived at an estimated value of $18.36 per share, which is inline with my estimates above.
Given that all estimates and assumptions are already fairly conservative, I would be fairly comfortable buying AMD below $17. Others may demand additional margin of safety.
Most of my purchases of AMD were during 2017 and early 2018, when the stock price was within the $10-$12 range and I determined that AMD was priced relatively close to my estimate of intrinsic value at that time, using calculations similar to the above.
The greatest challenges during this accumulation phase were psychological. AMD's price fluctuated greatly even before its late 2018 run-up, and at various points my positions were down 10-20%, despite what appeared to be continued positive business prospects. The daily, hyperbolic noise about nuclear war, China tariffs, interest rates and the impending crypto crash also added to the uncertainty. Fortunately, Benjamin Graham's Mr. Market analogy helped put these swings in perspective, and the legwork I had put into my valuations often gave me the confidence to continue adding to my position during any declines (although I won't admit it was easy).
Most of my holdings were sold during June - July 2018, in the $17-$18 range. My reasons for selling were that $18 only seemed justified if AMD executed perfectly during 2019-2020 and the market continued rewarding it with an out-sized (25x+) multiple - both of which might be entirely possible, but which were, in particular, not apparent in AMD's Q3 2018 guidance numbers.
Guidance for a 7% increase in revenue YoY, considering that Zen had been in the market for over a year, seemed like a relatively weak result and implied that EPYC and Ryzen OEM adoption were going more slowly than anticipated, despite Intel's struggles. That the stock continued its remarkable run-up even with these tepid estimates seemed to indicate that speculation, not reality, were at work. It's almost as if people were so relieved that AMD didn't implode from the crypto meltdown that they ignored the more important drivers of its long term growth.
Long term, I still like AMD's business prospects. But I don't like its current price.
More generally, I've tried to be more conservative in my estimates and more diligent about incorporating a larger margin of safety as my investing strategy has evolved, so a price (or multiple) that I might have been comfortable with at the beginning of the year I may not be comfortable with now. A large part of this was the increased market volatility towards the end of the year, which demonstrated clearly how irrational Mr. Market can be in the short term. Although I knew this on an intellectual level, it was another thing entirely to experience it with money on the line. I am less concerned now with missing a rally, and more concerned with getting a discounted price.
As of the date of this post, I no longer hold any stock in AMD, but may re-enter a position if the story remains solid and the price falls to a more acceptable level.