Valuation Metrics: A Cross-Examination
One of the common questions I get from subscribers is: why can't you make up your mind about how to value companies? The Diff has used free cash flow, option value, inflation-hedge status, the notorious "Adjusted EBITDA," lifetime value of customers, and handwavy approximations as markers of valuation.
In one sense, the question of how to value assets is an incredibly easy one: the net present value of future cash flows, with a few careful exceptions, is, in fact, the One True Way to value any kind of financial asset. I've written before about the "how" of this model—how to convert an investment decision into a series of future cash flows and then measure their value in the present. So this piece will be a bit more about the "why" and the "what else?"
The first claim to get out of the way is that DCF analysis is, fundamentally, what all investors are either doing or are betting that other investors are doing. Since investment means putting aside money today in exchange for a hopefully larger sum in the future, the question of which investment to choose is answered by determining which one's future returns are worth more. This is incredibly easy to apply to, say, a bond; it's trickier with a blue-chip stock; and it feels irrelevant when looking at more esoteric assets. But consider:
Did Facebook perform a DCF on Instagram before acquiring it? Explicitly, probably not. (There might be a spreadsheet saved somewhere that pretended to estimate Instagram's future revenue, but if it exists I wouldn't be surprised if the key decisionmakers hadn't actually opened it.) Implicitly, though, Instagram's value was the present value of future cash flows to Facebook that wouldn't exist if it had a full-scale competitor taking share and, at least initially, operating with a lower ad load.
What about collectibles? They're still worth what you expect to sell them for at some point in the future, and if that point can be predicted with some degree of accuracy then it needs to have a discount rate. For example, if you're betting that art by a particular artist will go up in price when the artist dies, you'd discount back an estimate of future sale value based on the artist's current life expectancy.
What about assets that don't provide a great return on their own, but that offer some hedging value? Gold mines have not been kind to investors, but owning gold mining stocks at times of low real rates driven by high inflation has sometimes been, well, a gold mine. If you're looking at one investment in isolation, a hedge may not be worth it. But when you consider a portfolio, what you're also looking at is the set of future opportunities you'll have to make better absolute-return investments when an external risk comes along and crushes the value of almost everything else. In that case, the hedge pays for itself not because of what you buy upfront but because of the opportunity to make future high-return investments at times when other investors can't.
What about higher-frequency transactions, like daytrading or making a market in an asset rather than owning it for the long haul? On very short timescales, DCF seems not to apply. A high-frequency trader buying 100 shares of AAPL at $145.37 and selling them at $145.38 is not, in fact, benefiting from the gradual upward drift in Apple's value as future cash flows accrue to the present, nor is this trade an attempt to benefit from a variant view that iPhone unit shipments next year will be 90.03m instead of 90m or something. This doesn't mean that DCF doesn't apply here; it does, in two ways. First, market-makers aim to have flat net exposure over time: they'll be long and short each company in roughly equal amounts over a lifetime of trading. But the traders they interact with are net biased towards being long, and are most likely paying for liquidity in order to make a trade based on a more fundamental outlook. On a microstructure scale, valuation exists but it's a rounding error. In the same way that dust in the air is affected by gravity but can bob up and down due to air currents, it's there but not the main force.
But discounted cash flows still come back to haunt us in that case. Because the trader needs to allocate resources, both human and financial, to running a trading strategy that targets a particular opportunity. A prop trading firm that puts a researcher and developer onto the task of improving their liquid single-name HFT signals is implicitly allocating the cost of a certain number of worker-months, plus the capital required to do the trades to buying a future stream of cash flows from the profits of those trades.1
If you want to get even fancier, you can apply discounted cash flow analysis to dumb financial decisions, whether they're on a grand scale (buying an expensive new corporate headquarters at a high point in the economic cycle when running a cyclical business) or on a smaller one (trading meme stocks when you're sure they're fundamentally overpriced). The people who do this kind of thing are generally getting some benefit from it, and it's possible to model their behavior by treating the difference between the cost of what they buy and the net present value of owning it as a form of luxury consumption. This gap can be a very large number, but that's true of plenty of other luxury goods; the bragging rights are in what you pay for, not what you get.
Doing this analysis sensibly means defining two tricky terms. What is free cash flow, and what is the discount rate?
Free cash flow has an accounting definition, but investment analysts wouldn't work such notoriously long hours if most of the work was copying and pasting a few lines from the income and cash flow statements and then doing some arithmetic. What free cash flow is really getting at is: how much cash, in a typical year, flows to the owner of the business while maintaining whatever growth rate has been modeled in the investment case. Lots of tech companies like to cite a free cash flow number that omits the impact of stock-based compensation, but if you're a 100% owner of a company, you either have to pay that in cash or have to accept that you're a 97% owner—wait, 94%—wait, the CEO you hired just hit a free cash flow hurdle that allows them to exercise another tranche of options so it's 93%—and so on. Skimping on capital expenditures is another source of short-term free cash flow, though defining the difference between "skimping" and "effectively economizing" can be the full-time job of many people working at the company. And there are still other tricks; one common element of the private equity playbook is to start requiring longer terms for accounts payable and shorter ones for accounts receivable—collecting money faster than usual and disbursing it more slowly. Suppose these are both due in 60 days originally, but the company stretches payables to 90 days and collects receivables in 30. In year 1, they collect sixty days worth of cash but probably haven't materially affected their taxable income. In years two through ten, they can't repeat the magic trick but may have more annoyed customers and suppliers. A careful analysis that starts in year 1 will assume that this cash benefit doesn't increase, and perhaps that there's a long-term hit to revenue growth or the cost of goods sold. A sloppier analysis may miss it.2
Determining discount rates is also its own art. There are some notorious cases where analysts have made a mistake in their valuation of a company, had the error pointed out to them, fixed the model, and adjusted their discount rate to get the same answer. Here is a brief piece on a one example. But a discount rate is just asking the question: given a set of assumptions about a business's fundamentals, what's the expected return? So lowering the discount rate without changing the valuation is, in fact, a way to say that the business isn't as compelling as it looked before: an investor paying the same amount should expect a lower future rate of return than they would with more optimistic assumptions.
The two basic ways to use discount rates are, essentially, semantic differences covering the same approach:
- Try to use a market discount rate based on the current risk-free rate and the historical risk premium for the asset in question, and use that to get a valuation for the company. If there's a big gap between the implied valuation and the current price, that's a trading opportunity.
- Hold the price constant and back into an implied discount rate based on the valuation. This is useful in both directions; a long thesis is basically saying "investors can expect a 15% annualized return from buying and holding this asset, under the following conditions," while a short thesis ends up being phrased as "Even if this company reaches all of its wildly optimistic targets, your expected return from owning it amounts to what you'd get from owning risk-free treasurys."
These are really performing the same operation, but they imply a difference in attitude. The first is a bet that the market will quickly recognize the pricing gap and rectify it. The second is a more serene approach, basically saying that an investor would be happy to hold because they know how much wealth they're accruing.3 It's possible to spend entirely too much time thinking about discount rates, but all they're really doing is offering a way to compare different streams of income from different assets. So as long as you're using a consistent approach—market-based discount rate and rank-ordering by price/value discrepancy or imputed discount rate and rank-ordering by expected return—you'll make the same investing decisions.
Edit: reader Benjamin Clark emailed to point out that this is not true, and for an important reason. His email is worth quoting in full:
This is only true if duration is equivalent among the investments being ranked and not otherwise. For sake of example, consider the following toy case. Investment A priced at $100 with a cash flow of $140 in 2 years. Investment B priced at $100 with a cash flow of $220 in 5 years. The appropriate market-based discount rate for each is 12%. Then, the implied present value of A is $111.61 (140/1.12^2) and the implied present value of B is $124.83 (220/1.12^5). So under that approach, investment B is preferable. However, the implied discount rate in order to make the net present value equal the current price is 18.3% (1.4^.5-1) for investment A and 17.1% (2.2^.2-1) for investment B. So under that approach, investment A is preferable.
Getting beyond the math and to why this actually matters as an investor, this is fundamentally a question of reinvestment opportunity/risk and time horizon and there are differences of approach among investors on this. You address that with "The first is a bet that the market will quickly recognize the pricing gap and rectify it. The second is a more serene approach, basically saying that an investor would be happy to hold because they know how much wealth they're accruing.", but I think it's important to recognize that the most exciting opportunities to investors with each of those approaches won't actually be the same.
The rank-ordering will be similar, but whether you optimize for highest expected return or highest valuation gap will give you different results, and work better with different strategies.
Is That All There Is?
All this is a long-winded way to say that, in the end, DCF is all there is to valuation. But "the end" is a short chunk of time! In most of the financial models I've seen, the DCF portion is a) the most important part of the model, and b) the simplest, mostly drawing on other metrics.
It's not the only way to value things. For example, many investors are fans of using comparable multiples—pick a metric, find a basket of similar businesses, and just buy the cheapest. Easy!
But it's a bit like a classic joke format. An analyst has a problem: valuing a company that doesn't produce free cash flow, and won't for several years. "I know! I'll use a comparables analysis." Now, the analyst has two problems.
Specifically, the first problem is choosing a set of comps that's actually comparable. And the second is ensuring that these comps are not all mispriced for the same reason. A comps analysis of SaaS companies would produce a 40% lower valuation today than it did at the beginning of the year, for example. But the even trickier problem is figuring out which companies are similar, and how similar they are. There are somewhat trivial examples of this; sometimes, one company in an industry will use slightly different accounting standards than the others, so comps need to reflect this. But also, the weirdest companies—which tend to be the companies whose current financials are the least useful guide to their long-term prospects—are the hardest to compare to anything else. If you're looking at a business category with high margins and strong network effects, for example, your comps table might have the market leader trading at 15x revenue, the #2 player at 2x, and the #3 player at 0.5x (with that multiple reflecting the potential value of acquiring patents and employees, not the underlying business).
Ironically, the best use case for comps is when the businesses are wildly incomparable in their current fundamentals because they're completely different approaches to solving the same problem. It was useful a decade ago to think of a media conglomerate like Viacom as a comparable for both Netflix and Facebook, since Viacom could answer the question "How much money per attention-hour does a company make when it's a mature business providing fairly passive entertainment that millions of people default to without thinking hard about it." This didn't offer a useful valuation multiple for either company (though—fun fact!—you could buy Viacom and Netflix at the same P/E in mid-2007 and the first half of 2010). What it does instead is offer a sort of valuation framework, telling you roughly how big the business can get. For DCFs, modeling growth companies is partly a question of measuring growth and partly one of measuring how durable growth is; if a business can consistently grow at higher than the discount rate being applied to it, then investments can work out very well for two reasons:
- Keeping up with the growth story means consistently looking at how current operations reflect on a very long-term outlook. If part of the thesis is network effects, that means ensuring that the company is still adding users faster than competitors; if the thesis is management's execution skills, then it means keeping an eye out for avoidable mistakes, even minor ones; if the thesis is that the business is persistently willing to allocate capital to high-risk/high-reward opportunities, it means keeping an eye out for the absence of mistakes—the best way to achieve a 100% success rate on new product launches is to never launch anything daring.
- If the expected period of high growth rounds up to "a really long time, basically forever," then it's possible for the company to trade at a consistent multiple of rapidly-growing earnings. Multiples compress as growth expectations decline, but they increase as investor confidence in growth goes up. Google traded at a higher P/E ratio in 2018 than it did in 2010, even though it was 3x bigger and growing a bit slower, in part because everyone who had ever bet that Google's growth would materially slow down had turned out to be wrong—the stability premium offset the slower-growth discount.
And this is why analysts spend so much time figuring out the guts of financial models. They're building estimates of drivers that will spit out a discounted cash flow analysis, because those drivers need to be accurate in the short term and realistic in the long term.4
Those metrics matter, and they're often a useful shorthand. For example, for a company that's still reinvesting in its business, current free cash flow is not an especially useful hypothetical, but other measures of growth can work. I typically use EBITDA, since it's a measure of cash earnings that ignores financial structure, and then back into what actual cash is available afterwards. For earlier-stage companies, revenue is fine as a metric.5 But its use is still to track a business’s progression towards some mature state where it can be valued based on something else. A company that grows for a long time and never turns into a business that returns cash is not a great business. It might be an interesting or socially useful one, but any returns to investors will be contingent on other investors mistaking those traits for good economics.
And there are some special cases where the right route to a DCF calculation is somewhat circuitous. Financials get their own special criteria. They’re hard to analyze on a cash flow basis, because cash is constantly moving around; a lender that’s able to add loans instead of growing its cash balance is, hopefully, getting more capital-efficient over time and increasing the rate at which it compounds. Net income can be wobbly, too. The high-level criterion I look at for financials is changes in their tangible book value per share. Financial companies are deploying balance sheets, and over time their ability to grow the equity portion of that balance sheet is the criterion by which they should be judged. But they’re also returning capital; per-share metrics are a good way to give them credit for that. (You’ll also want to add in dividend returns.) At maturity, the right model for a financial company is that it's scaled up to the point that it roughly earns its cost of capital; the compounding of tangible book value per share is basically the growth rate of what will look, at maturity, like a portfolio of bonds.
But all of these are really heuristics and workarounds. Early on, a DCF is mostly hypothetical, but is at least worth thinking about because an early-stage, money-losing company is a call option on a mature business that throws off cash.6 For founders, that might be a depressing thought, but there's an upside. In the long run, DCF analysis puts the highest value on mature businesses, especially quality ones, because it reflects how much they can return to investors, how confident those investors need to be, and the timing of their returns. So the full valuation of a mature company is basically compensation for losing a more interesting immature one. It's yet another example of the lack of free lunches in economics: sticking with a business until it's valued like a bond is the compensation you get for starting with something fun and slowly turning it into a bond.
A Word From Our Sponsors
Event contracts are here. Bloomberg announces a new market type.
Last week, Bloomberg covered the story of the first CFTC regulated prediction market, Kalshi. For the first time, traders now have the ability to buy the asset class: regulated Event Contracts.
We couldn’t believe it when we saw that these markets have been live for almost a year. With inflation at 8.3% and interest rates climbing 50 bips each month traders are looking to get their feet wet in new markets.
While the S&P 500 is down 14% on the year, the top 10% of traders on Kalshi’s exchange are enjoying an average 59% ROI.
Instead of trading noise in the stock market, predict outcomes of events. Or if you want to stick with what you know, trade options on stock indices with as little as $1.
We’ve partnered with them to get readers of the Diff exclusive access.
What's Still Fundable (Con't.)
A recurring theme in this newsletter recently has been funding rounds for companies that resist the current tech sector cyclical reversal. There are funding rounds that wouldn't look that out of place in 2021, albeit being done at lower valuations and with longer deal cycles. But since VC funding lags VC fundraising, there's still room to raise large and high-valued rounds if they back companies that seem safe in the current economic environment. For example, career education provider Guild Education just raised at a $4.4bn valuation. The general flow of macro data has been that inflation is partly driven by food and energy, but also strongly influenced by wages, so a company that increases the supply of workers is more of a growth business than it was a few months ago.
ESG and Measurement
Axios highlights the trend towards more self-driving EVs for the construction and mining industries. This might be a function of where the EV opportunities are, but it's also caused by where they're measurable. Usually tenants in a building don't think of themselves as holding responsibility for a pro-rata share of the emissions required by the building's construction. Construction companies, by contrast, will get that cost attributed to them. In cases where there's alignment between the emission-creating decision and the penalty for emissions, we'll probably see more carbon-conscious investment, while in other areas, there's a mismatch between who cares about environmental impact and who can affect it.
A topic that goes in and out of favor in investing and marketing is tracking social media sentiment as a predictive factor. Two factors make this challenging: first, people on social media tend to riff on what other people are talking about, even if it doesn't affect their real-world behavior all that much. A 10x increase in social media attention might translate to a 10% increase in revenue for a brand. The other issue is that sarcasm is hard enough for humans to detect, and can be even more difficult for automated systems (especially when there's such a weak connection between online and offline behavior—you don't even have to know if you're kidding to crack a topical joke). A great case study in this is the recent re-release of Morbius, a weak box office performer that was the topic of some popular memes in the last few weeks.
For companies, it would be nice to live in a world where they can measure eddies in demand in real time, and consistently give customers exactly what they want. The problem is that the closest they can get is to give customers what they're asking for, and as it turns out nobody was seriously asking for any more Morbius than they'd already gotten.
Commodities giant Rio Tinto says that price volatility is likely to stay high in commodities ($, Nikkei) both due to geopolitics and the high cost of a transition to green energy. One way to look at the mining industry right now is that we've implicitly applied a very high discount rate to emissions: the net present carbon impact of digging up a ton of nickel, cobalt, copper, etc. could be negative if it enables substitution from higher-emissions products, but the impact on a company's stated carbon emissions if it opens up a new mine will be higher.
Diff Jobs is our service matching readers to job opportunities in the Diff network. We work with a range of companies, mostly venture-funded, with an emphasis on software and fintech but with breadth beyond that.
If you're interested in pursuing a role, please reach out—if there's a potential match, we start with an introductory call to see if we have a good fit, and then a more in-depth discussion of what you've worked on. (Depending on the role, this can focus on work or side projects.) Diff Jobs is free for job applicants.
- A company which helps local restaurants expand to new markets through micro-foodhalls (written up in the Diff here) is looking for its first market launcher. (various locations)
- A company building a marketplace for private equity secondaries is looking for a co-founder to lead deal sourcing and research efforts. (US, remote)
- A company in the education space which offers advice to people from high school age through to early career professionals is looking for GM/mini-CEO candidates to own a variety of company functions. It's likely the right candidates will have a product background, but not necessary. (US, remote)
- A company adding a software layer to the VC fund closing process is looking for their first Product Manager and Product Designer. (NY, Philadelphia, Remote)
- A company offering working capital loans to small, traditionally-underbanked businesses is looking for engineers across the stack; infra, backend, and web. (SF)
Stay tuned for a future post on this business model. The proprietary trading business is fascinating, not just because of what the companies do—provide constant liquidity while determining the efficient frontier between automation and human judgment—but because of how their model lets them think about opportunities. They may be the world's most systematic and effective allocators of research time, so in a meta sense it's very good news that some of them are getting quite rich and donating their money to research on things other than systematic trading signals. ↩
An analysis in year zero, i.e. someone who is planning to buy a company and perform this maneuver, may find that it's highly profitable for them because it allows them to pull cash out of the acquisition right away, producing a fancy-looking internal rate of return. Since the cash is expected to come out of the business with high confidence, that portion of the acquisition will probably be funded by debt rather than equity, so any additional improvements to the acquisition will look that much better on a levered basis. For the buyer, there's yet another layer of DCF: what is the net present value of fees raised on a larger fund in a year or two, conditional on cosmetically good numbers from the current fund today? On that basis, the transaction probably looks very good. ↩
Note that the discount rate in that second scenario is not necessarily the expected return. If the analysis is right, it's an understatement; if you buy at an implied discount rate of 12% and immediately sell at an implied discount rate of 8%, you've made 50%. Whether you do sell at this point depends on whether or not you see other opportunities whose return, after accounting for capital gains taxes and other transaction costs, is higher than the implied 8% you're getting at the new price. ↩
Technically some funds are only looking at the accuracy of those short-term drivers, but in effect what those funds are doing is mentally modeling a long-term investor who will adjust long-term views based on short-term results. You can go to the extra effort of estimating what a 2% revenue miss this quarter does to 2028's probably free cash flow, but in practice if you're good at estimating those near-term deviations, the binary beat/miss outcome is more tractable than getting the magnitude perfect. ↩
Even here, there are complexities depending on margins and business structure. Some companies are in an industry where they’re buying something, marking it up, and selling basically the same thing to someone else. That’s broadly true for many businesses, like retail, but it’s usefully true for companies in adtech or any business that analogizes well to wholesale distribution. For these companies, the real topline metric is not revenue, but the markup. For example, an adtech company might have some lines of business where it’s taking a gigantic markup, but can’t easily scale, and others where the markup is minimal but the scope of business is enormous. The way this gets expressed in their financials is that they cite revenue and revenue less traffic acquisition cost; the latter number (RexTAC if you want to be fancy) is the key metric. Since the company is benchmarking its fixed costs to some measure of incremental profitability, it makes sense to let all of these markup differences wash out. Google, for example, pays a healthy amount to intermediaries for some of its ads, but runs other ads on owned-and-operated properties where there’s no traffic acquisition cost. ↩
Option-based analysis can also be very useful for cases where a company produces some natural resource at a predictable cost; Saudi Aramco is long an option on crude oil at a strike price of about $3/barrel, but is short call options at higher prices reflecting the royalties it owes the government. High-cost producers that aren’t actively producing—owners of mothballed mines or undeveloped land, for example—are long an out-of-the-money option. And this thread is a great primer on thinking of a short position as an option. Selling volatility can be a tough business, especially if you don’t know that’s the business you’re in. ↩