How to Build a Tornado Chart for Your DCF Assumptions

Which single assumption in a discounted cash flow model does the most damage if you get it wrong: the discount rate, the terminal growth rate, or something most DCF templates don't even ask you to name explicitly? Ask ten investors before they've built the model and most will guess the discount rate. On a business mid-capex-cycle, they're usually wrong.

That's the practical problem with most DCF work. It produces one number, computed out to the cent, while burying the fact that a handful of underlying assumptions are doing nearly all the work — and one of them is usually doing far more than the rest combined. A tornado chart is the tool that makes this visible. It doesn't replace the model. It ranks the model's own assumptions by how much damage each one can do to the answer, so you know which one actually deserves your scrutiny and which ones are safe to leave as a rough guess.

I'll build one here using Texas Instruments' fiscal 2024 numbers — a useful case precisely because the company's headline free cash flow looks weak right now for reasons that have nothing to do with the durability of the underlying business. That's exactly the setup where getting the ranking wrong costs you the most.

What a Tornado Chart Actually Measures

A tornado chart is a one-way sensitivity analysis dressed up as a picture. Build a base-case DCF first — the number every model has to start from anyway, using this site's own framework or a fair value estimate built the MoatScope way. Then, one assumption at a time, hold everything else fixed and swap in that assumption's low and high plausible values instead of its base case. Record the fair value each swap produces. The gap between the high-value output and the low-value output is that assumption's bar. Do this for every assumption, sort the bars from widest to narrowest, and stack them with the widest bar on top. The resulting shape — wide at the top, tapering toward the bottom — is where the name comes from.

Tornado Chart Construction — for each assumption Aᵢ: hold all other assumptions at base case. Compute FairValue(Aᵢ = low) and FairValue(Aᵢ = high). Bar width = |FairValue(high) − FairValue(low)|. Sort bars widest to narrowest. The assumption with the widest bar is the one actually driving the valuation; everything below the top one or two bars is close to noise by comparison.

Not the model's best guess. The model's most damaging one — that's what the widest bar identifies.

The chart doesn't tell you what any input's "right" value is — that's a separate, harder judgment, and a tornado chart deliberately dodges it. What it tells you is where your judgment matters and where it doesn't. Getting the third-widest bar wrong by 20% might move fair value a point or two. Getting the widest bar wrong by the same 20% can move it by double digits. Most investors spend roughly equal time refining every input in a model. That's backwards, and a tornado chart is the fastest way to see it.

Note what this is not. A tornado chart varies one assumption at a time while holding the rest fixed — it's a one-way sensitivity, not a joint one. A full Monte Carlo simulation, which lets every assumption wobble simultaneously according to its own probability distribution and reports a range of outcomes rather than a set of ranked bars, is a more complete stress test. It's also considerably more work to build correctly, and most individual investors never will. The tornado chart is the practical middle ground: cheap to build in a spreadsheet, and it answers the one question that actually changes how you spend your research time — which input matters most — even though it can't tell you how two inputs moving together might compound each other.

Texas Instruments: A Case Built for This Exercise

TI's fiscal 2024 Form 10-K, filed with the SEC in January 2025 for the year ended December 31, 2024, is a good candidate for this exercise precisely because a naive glance at the headline numbers would lead you astray. Revenue came in at $15.641 billion, down 11% from $17.519 billion the year before, as the semiconductor cycle rolled over. Net income fell to $4.799 billion, down 25% year over year, and diluted EPS dropped to $5.20 from $7.07. On those two numbers alone, TI looks like a company in genuine decline. Same filing, though. A very different story sits two lines down the cash flow statement.

Free cash flow tells a more specific story. Operating cash flow held up reasonably well at $6.318 billion for the year. But capital expenditures ran $4.8 billion, down only slightly from $5.07 billion in 2023, leaving free cash flow at roughly $1.5 billion — a fraction of what the business earns on an owner earnings basis before that spending. TI has been mid-buildout on new 300-millimeter wafer fabs in Sherman, Texas and Lehi, Utah since 2022, a program the company has described as running through the back half of this decade before capital intensity normalizes toward historical levels. A DCF anchored to today's depressed free cash flow and extrapolated forward with a modest growth rate would badly understate what this business is worth once that capex program tapers and the new capacity starts converting into cash instead of consuming it.

That's exactly the setup where the ranking of assumptions matters more than usual. Set up a base case: a discount rate of 9.5%, a terminal growth rate of 2.5% — mindful of how sensitive terminal value assumptions already are on their own — and an input a standard DCF template often skips entirely: an assumption about how many years it takes free cash flow margin to recover from its current depressed level back toward TI's own pre-buildout norm, which ran closer to 30% of revenue for much of the 2010s.

None of this is a bet that the capex program itself is misguided. TI spent $3.8 billion on research and development in fiscal 2024 alongside the $4.8 billion of capex, and the company has been explicit that the Sherman and Lehi fabs are built for a multi-decade demand horizon in analog and embedded processing, not a single product cycle. The question a tornado chart forces you to confront isn't whether that spending is justified — it's how much of your valuation's accuracy actually depends on guessing correctly when the spending tapers, versus how much depends on the discount rate you'd otherwise agonize over in a spreadsheet cell.

Building the Bars, One Assumption at a Time

Vary each of three assumptions across a plausible range, one at a time, holding the other two at base case, and the resulting fair-value swings rank in an order most investors wouldn't guess in advance. Running that pass — illustrative, not a published fair value estimate — produces a ranking like this:

• FCF margin recovery pace (eight years versus four years to return to a 30% margin): roughly a 35–40% swing in fair value, the widest bar by a wide margin.
• Terminal growth rate (1.5% versus 3.5%): roughly an 18–22% swing.
• Discount rate (8.5% versus 10.5%): roughly a 15–18% swing.

The order is the point, not the exact width of the third decimal place. On a business mid-capex-cycle, the assumption doing the most damage isn't the one most DCF critiques spend their energy attacking.

The mechanism behind that ordering is worth naming, because it isn't specific to TI. Discount rate and terminal growth both operate on the whole projection at once, at a fixed percentage — their effect is real but proportionally bounded by how the math discounts a stream of cash flows. Margin recovery pace works differently: it changes the actual dollar amount of cash the model generates in every one of the next several years, compounding forward before a single discount factor gets applied to it. A one-year delay doesn't shave a percentage off the answer. It reroutes several years of free cash flow from arriving soon, worth more today, to arriving later, worth less today — and that reshuffling shows up directly in the explicit forecast period, before terminal value even enters the calculation. For any business running a large, temporary capital program — a fab buildout, a data center campus, a fleet expansion — expect this kind of near-term operating assumption to outrank the two inputs most DCF critiques treat as the whole conversation.

I'd treat the precise widths above as illustrative rather than load-bearing — small changes to the base case shift each bar's exact size, and reasonable people will draw the ranges differently. What doesn't move much, in my experience building these across other capital-intensive names, is which assumption sits on top. For a business running an unusually heavy, temporary capex program, the pace of margin recovery tends to dominate the discount rate almost every time. And that's the opposite of where most DCF critiques focus their energy.

What the Shape Actually Tells You to Do Next

Once the ranking exists, the instruction is simple: spend your research time on the top bar, not the bottom one. If margin-recovery pace is the widest bar, the highest-value question isn't "what discount rate should I use" — it's "when does TI's own capex, as a share of revenue, start declining, and does management's own guidance support a four-year recovery or an eight-year one." That's a question you can actually research: capex guidance, fab utilization commentary on earnings calls, semiconductor capital-intensity history across past cycles. The discount rate, by comparison, is mostly a house view you set once and rarely need to revisit line by line.

This is also the honest answer to why a fair value estimate should never be presented as a single number. A DCF's output looks precise because it's computed to the cent. It isn't precise. It's exactly as precise as its least-examined widest-bar assumption, and that's usually far less precise than the output format implies. Treat any DCF, including this one, as a compass rather than a GPS coordinate — it points you in a direction and gives you a rough magnitude, not an exact destination.

That's not a knock on the tool. It's a reminder to use it honestly.

Key Takeaways

• A tornado chart varies one DCF assumption at a time across a plausible range, holding the rest at base case, then ranks the resulting fair-value swings from widest to narrowest — the ranking, not any single input's exact value, is the useful output.
• Texas Instruments' fiscal 2024 numbers (revenue down 11% to $15.641 billion, free cash flow near $1.5 billion against $4.8 billion of capex) show why the discount rate isn't always the widest bar: for a business mid-capex-cycle, the pace of margin recovery can swing fair value more than the discount rate and terminal growth rate combined.
• Spend research time on the widest bar, not the assumption a template happens to list first — most DCF critiques target the discount rate by habit, not because it's always the input doing the most damage.
• A DCF's apparent precision is a function of its least-examined, widest-bar assumption. Treat the output as a compass, not a GPS coordinate, and let the tornado chart tell you which direction to actually double-check.