I have read, by my count, somewhere north of two thousand discounted cash flow models. Some of them were excellent. Most of them were elaborate hand-waving exercises with a number at the bottom. The really pernicious thing is that almost none of them were wrong, in the sense of containing an arithmetic error. They were wrong in much more interesting ways.
Here are the three failure modes I see in nine out of ten. None of them are about the math. They're about what the math is forced to do when the analyst doesn't have a thesis.
Failure 1: Terminal value theatre
If more than 65% of your enterprise value comes from the terminal value, you are not modeling the business. You are modeling your assumptions about a steady-state that may never arrive.
This is the most common failure I see, and it's almost always caused by the same thing: an explicit forecast period that's too short for the business to have plausibly normalized. Five years is fine for a mature consumer staples company. It is criminal for an early-stage SaaS company that won't reach steady-state until year nine or ten.
The terminal value is where intellectually lazy DCFs go to hide. - Aswath Damodaran, summarized loosely
The fix is straightforward: extend the explicit period until growth and margins converge. Yes, this means making more assumptions. Yes, this feels worse. It's not. It's just making the assumptions visible instead of compressing them into a single Gordon-growth handwave.
Failure 2: WACC drift
The discount rate is a single number that pretends to capture five things: the risk-free rate, the equity risk premium, the company's beta, its capital structure, and its cost of debt. Every one of those changes over a ten-year forecast. Nearly nobody models the drift.
The most egregious case: an analyst uses today's near-zero risk-free rate to discount a model whose terminal year assumes interest rates have re-normalized. You can't have both. Either rates normalize and the WACC moves with them, or they don't and your terminal-year revenues are a fantasy.
What to actually do
- Use a two-stage WACC: one for the explicit period, one for terminal. They should not match unless you can defend why.
- Tie your terminal risk-free rate to a long-term sovereign average, not today's spot.
- Recalculate beta at steady-state leverage, not current; companies change their capital structure as they mature.
Failure 3: Year-6 fairy tales
Here is a heuristic that has saved me dozens of times. Open any DCF. Find year six. Read the line items.
In most models I review, year six contains growth rates the company has never demonstrated, margins it has never sustained, and a reinvestment rate that requires it to find investment opportunities it has never found. The model has implicitly assumed the company will become a different, better company, without any explanation of how.
The fix is what I call thesis equivalence. Every line item in year six should map back to a sentence you could say out loud in committee. If you can't say it, delete it.
What I run instead
I still build DCFs. But I run them with a scenario triplet: base, sober, and (briefly) reverse-engineered. I require the model to survive all three. The reverse-engineered one is the most useful: solve for what the market is pricing in, and ask yourself whether you believe it.
None of this is novel. Most of it is in Damodaran's textbook. But it's amazing how rarely the basic checks are run, and how many of those two thousand models would have benefited from the equivalent of "read it back to me".
- Rafael, from a hotel desk in Rio de Janeiro, 4:30am.