Discover more from Mostly metrics
The Death of the DCF Model
How discounted cash flow models work (and don't work)
With the recent market correction, DCF models have been flying off the shelves over at the valuation store.
As a primer to the majority of readers who are not valuation dorks, a DCF (discounted cash flow) model is basically a way to estimate what a company is worth by projecting the cash it’s expected to spit out to shareholders over the course of its lifetime, and then discounting it back to a smaller present day value.
To most high growth investors in the 2019 to 2021 timeframe, DCF models were useless, nonsensical. This is largely because, well, you can’t discount back a loss.
What I’d like to do in this post is:
Review the components of a DCF model
Discuss why DCF models are inherently flawed
DCF Models are flawed because:
There’s a disconnect between DCF timeframes and investor timeframes
WACC is not a single number in reality, but changes as the company matures
Invoicing for multi year contracts can distort cash flow trends
Share based comp can cause major dilution and result in less FCF than expected
Modelers often think of Exit Multiples and WACCs as offsetting, instead of independent variables
There’s just too much value locked up in the Terminal Value for comfort
Timeframe: Typically 10 years
Revenue Forecast: Usually assumes base case top line achievability, plus a bit more conservatism if there’s no historical free cash flow to date
Medium Term Growth Rate: What you believe the company will grow at over the next 3 to 5 years
Long Term Growth Rate: The estimated growth rate towards the end of the ten year timeframe, and what you think it should grow at in perpetuity. This should generally match what the sector as a whole is forecasted to grow at
EBITDA Forecast: Based on a target operating model where OPEX plateaus and profits creep up towards ~20% to 30% (or more)
Weighted Average Cost of Capital: Think of the WACC as two things: 1) A hurdle rate investors expect to exceed in return for investing their money 2) A way to discount (or shrink) the free cash flow forecast back to present day values based on risk. Here are the building blocks:
Market Beta: A measure of how the company moves with the larger equity market. As an example, Gitlab, a high growth risky-ish tech stock, uses a Market Beta of more than 2. High risk, high return.
Risk Free Rate: Usually baselined to the price of US treasuries (aka “that free free”)
Equity Risk Premium: The excess return an investor gets when they invest in the stock market, compared to the risk-free rate. Just take the average market return (say 8%) less the Risk Free Rate (say 3%) to get your Equity Risk Premium (in this example, 5%). That’s what you get for taking on additional risk.
Cost of Debt: Think of this as how much the company would pay in interest for a loan if they went to a big bank. Larger, more stable companies get better rates. A small SMB HR Tech Co. pays a lot more for a revolving debt facility than Apple.
Exit Multiple: Multiple of Free Cash Flow. MSFT trades in the low 20x’s right now based on growth, Snowflake in the mid 30x’s, and very high growth companies up to 50x. The last year in the forecast gets hit with this multiple.
Terminal Value: This is a product of the Exit Multiple and the final year’s free cash flow. The terminal value is typically a large portion (>50%) of the total assessed value and is therefore a huge component of the overall analysis.
2. Why DCFs Are Flawed
The investors using 10 year time horizons are probably buying everything and pretty passive. Maybe this works for Blackrock, but most investors aren’t Blackrock.
“On the venture buy side we don’t really hang our hats on DCFs”
-VC friend I asked about DCFs
For example, private equity and venture capital firms are only looking to hold for 5 to 7 years. Some for only 3 or 4 years. What would they use a WACC for if they’re never making it to the years to the left of the Terminal Value label?
The lesson: Know what type of investor you are before building out a DCF. It may not apply to your holding period.
WACC is Whacky
Did you predict the cost of debt would need to be scraped off the floor in early 2022 like a chewed piece of gum? The WACC formula seems easier to accurately calculate than it really is. Because certain elements of the formula, such as the cost of equity, are not consistent values, it’s not uncommon for different parties to have different inputs.
In a ten year timeframe, we’re almost certain there will be a “riskier” 18 to 24 month period of time. History says there’s a market correction or inflation spike buried in there somewhere. And when it occurs, this is where the bulk of the valuation hit would occur. We just don’t know when the shock will happen, so we equally burden all periods with a blended rate somewhere in the middle. We peanut butter spread the hit.
Nirvana would be if we could use a different WACC for different timeframes. But then we’d need a crystal ball. In the absence of that, we’re left using a blended WACC that’s equally imperfect for all forecasted years - overly burdensome for the good times (like a year ago) and not burdensome enough for the bad times (like right now).
The lesson: WACC is not a single number in reality, but changes as the company matures.
Stock Based Comp without Dilution
Stock based comp, what employees are paid through firm ownership, impacts free cash flow. The more stock based comp that’s rewarded, or the more shares that are given out, the more existing shareholder’s are diluted. Increasing the denominator means shareholders are entitled to a smaller chunk of whatever FCF is left over.
(BTW - here’s a primer on how dilution works.)
There’s been a recent shift where more and more public investors are thinking about share based comp as a real charge to run the biz. Call it a revelation of sorts!
If we were all being academically honest, we’d assume 2% to 3% dilution annually and fully burden our models for share based comp.
But everyone chose to ignore SBC for a long time. And it’s hard to put the toothpaste back in the tube.
“We’re at the point where if we started to include SBC we couldn’t invest in anything”
-Hedge Fund friend I asked about DCFs
You may counter that share based comp and free cash flow dynamics cancel out in the long term by way of share buy backs. That’s true. In a perfect world, the company becomes MSFT or Oracle and can use all those fat stacks of cash to actually buy shares. Doing so reduces the denominator of shares outstanding, and entitles each shareholder to a bigger slice of the FCF pie. It’s kinda like reverse dilution.
But in my opinion, this is purely philosophical thinking. Not every company becomes MSFT or Oracle.
The lesson here: Investors should pay a lower multiple on FCF than on Earnings, since EBITDA, and as a result FCF, excludes SBC. Don’t get fooled again.
Cash Flow without COGS
Most software companies sell multi year contracts. This is amazing from a cash conversion cycle standpoint. Cash today is better than cash tomorrow.
But it can drive huge deviations between your FCF and Earnings. Crowdstrike and Palo Alto are prime examples, with divergences of 30% to 40% between the two. They benefit from multi year contracts with sweet, sweet up front collections.
The issue is you’re getting credit for deferred revenue (which isn’t in your earnings yet) without the burden of your cost to serve (the hosting, infrastructure and customer support costs you’ll incur down the road, which belong in your gross margin).
The lesson here: Getting credit for def rev w/o COGS is a big mismatch. You have to discount FCF back as a result.
Exit Multiples vs WACCs
Ah, the classic see-saw.
Companies are riskier when they have higher growth rates they’re expected to hit over an extended timeframe. As a result, they get discounted more heavily in their WACC. However, they get a rich exit multiple because they’re growing so fast.
Companies with lower revenue forecasts are less risky, and inherently get lower WACCs. So they aren’t discounted as heavily. However, they get crappy exit multiples because, well, they aren’t growing as fast.
My observation is many people running valuation models talk themselves into just counterbalancing a higher WACC with a higher Exit multiple. They treat the two as counterweights rather than two independent variables in the model.
The lesson here: Don’t just put a big kid on one end of the see saw to balance out another big kid. Treat each variable as its own.
DCFs are highly dependent on the quality of assumptions. No where is this more true than the Terminal Value.
This is my biggest gripe with DCFs. You have this massive chunk of value sitting in a lump sum, which is largely dependent on assumptions you’re making about a company a decade away. It’s hard to predict the operating model for a company in a market condition that’s yet unknown. It’s even harder to predict if the company is super high growth at the moment - when do they come back down to earth? Realistically budgeting for anything that’s growing over 100% Y/Y is really hard, and gets even harder the further out you look.
Finally, it’s nearly impossible to get the Terminal Value right if there’s no history of free cash flows yet. You may as well throw darts at a wall, since companies at this stage update their operating models every 6 months.
The lesson here: If you have more than 75% of the company’s total value wrapped up in the terminal value lump sum, you’re skating on some thin ice and may want to rethink your assumptions. How good do you feel about your model?
At the end of the day, valuation isn’t just one metric or one model - it’s a triangulation of multiple views. Valuation doesn’t work when we over index on one methodology. It also doesn’t work when we denote a level of false precision. And it’s dishonest if we don’t call assumptions what they are - merely assumptions on what we think could happen.
You may call me dense, and say everything I’ve outlined is really just user error.
Sure, some of my considerations should be relatively simple to catch and fix for.
Or maybe you counter that the real problem is people trying to apply DCF models to stuff they shouldn’t be applied to (i.e., cash burning companies). The classic square peg, round hole predicament.
Both are fair arguments. But in aggregate, I think there’s a whole whack of assumptions we place too much confidence in which can lead to false signals.
“The only thing I know for certain is in eight years my model will be wrong and your model will be wrong”
What I’ve Been Reading
Thinking in frameworks is something that’s ingrained into my very being. I used frameworks to organize my baseball cards faster in second grade, and use mental models now to cut through the noise when I’m dropped into a data room.
If you’re interested in improving your general thinking skills, Mental models, Concepts, and Frameworks by Alex Brogan is a weekly newsletter you should subscribe to.
Quote I’ve Been Pondering
"You have your whole life to write your first [album/book/movie] and only 18 months to write your second."