A false sense of precision

The last post ended with a value of ₹397.72 a share for Desi Bites. Two decimal places. It looks like a measurement.

It isn’t. It’s the output of a chain of estimates — a growth rate that fades on a schedule someone chose, a margin path someone assumed, a beta borrowed from a sector, an equity risk premium that reasonable people put anywhere in a range several percentage points wide, and a perpetual growth rate for a company that will outlive everyone reading this. Every one of those is arguable. The spreadsheet reports ₹397.72 because spreadsheets always report something.

This post is about the two habits that keep a discounted cash flow (DCF) model honest: showing how much the answer moves when the inputs move, and refusing to act on the answer unless there’s room to be wrong.

Sensitivity: show the range, not the point

A sensitivity table re-runs the model across a grid of assumptions and prints every answer. The two inputs worth gridding are almost always WACC and terminal growth, because those two do the most damage.

Value per share (₹) for Desi Bites, at varying discount rates and perpetual growth rates:

WACC ↓ / g → 3.0% 4.0% 5.0% 6.0% 7.0%
12.91% 384 407 437 474 525
13.41% 369 390 416 449 492
13.91% 357 375 398 426 463
14.41% 345 361 381 406 438
14.91% 334 349 367 389 416

The base case sits in the middle at ₹397.72. But move WACC one percentage point in each direction and terminal growth two points — the edges of the grid, and a range no analyst would call unreasonable — and the answer runs from about ₹334 to ₹525. The high end is 57% above the low end. Even at just one point on each input, it’s ₹349 to ₹474.

That’s the honest output of this model. Not ₹397.72, but “somewhere in the high 300s to low 400s if my central assumptions hold, and plausibly ₹334 to ₹525 across assumptions I can’t rule out.”

A range is less satisfying than a number. It’s also true, which is the better property for something you’re about to risk money on.

Which assumptions actually matter

Not every input deserves equal worry. Rough sense of what moves the Desi Bites valuation, in order:

Assumption Impact Why
Terminal growth rate Very high Sits in the denominator of 74.2% of the value
WACC Very high Compounds through every year, and hits terminal value twice
Revenue growth path High Drives every downstream line in the forecast
EBITDA margin path High A point of margin on ₹5,000 lakh of revenue is real money
Capex intensity Medium Large in the forecast years, fades in importance by the terminal year
Working capital % Low Only the change matters, and it’s small relative to everything else
Tax rate Low Fairly well known, doesn’t move much

The pattern is worth noticing: the assumptions with the most influence are also the hardest to defend. Tax rates are knowable. Perpetual growth rates are not. Precision and importance run in opposite directions here, which is the fundamental awkwardness of the whole exercise.

Margin of safety

If a valuation is a range rather than a number, acting only when the price sits at the bottom of the plausible range gives you room to be wrong. That buffer is the margin of safety — an idea from Benjamin Graham, and probably the single most durable concept in fundamental analysis.

Margin of Safety (%) = (Intrinsic Value − Price) / Intrinsic Value × 100

Run it on Desi Bites at its IPO price. With a DCF value of about ₹397.72 and a price of ₹640, the margin of safety is (₹397.72 − ₹640) / ₹397.72 ≈ −61%. Negative: the price sits well above the model’s value, so there’s no buffer at all. For a 30% margin against this model, the price would need to be around ₹278. That’s arithmetic on a fictional company, not a price anyone should wait for — and the rest of this post is about how little the ₹397.72 itself deserves to be trusted.

The logic is defensive rather than clever. Your model will be wrong; the question is only by how much and in which direction. A price well below your estimated value means you can be substantially mistaken and still not lose money. A price at or above it means every one of your assumptions has to come good just to break even.

How large a buffer? There’s no formula, and anyone offering one is overselling. The sensible principle is that the buffer should scale with your uncertainty:

Situation Buffer typically wanted
Stable, predictable business; long track record Smaller
Cyclical, or a short operating history Larger
Model leans heavily on terminal value Larger
Sensitivity table shows a wide range Larger

Desi Bites hits three of those four. It’s a small-cap with three years of audited accounts, 74.2% of its value sits in the terminal value, and the sensitivity range spans ₹334 to ₹525. That combination argues for a wide buffer, whatever number you’d normally use.

Where margin of safety gets misused

Two failure modes, both common.

The first is treating the buffer as a substitute for understanding the business. A 50% discount to a valuation built on assumptions you can’t defend isn’t safety, it’s a bigger bet on a worse model. Graham’s buffer was meant to absorb ordinary estimation error, not to compensate for not knowing what a company does.

The second is assuming a gap between price and value must be an opportunity. Sometimes the market knows something the model doesn’t — a customer concentration risk, a promoter dispute, a regulatory change working its way through. When the market disagrees with your model, “I’m right and they’re wrong” is one explanation, and it’s rarely the first one to reach for. The reverse DCF from the last post is the better instinct: work out what the price is assuming, then go and check whether that assumption is defensible.

Common mistakes

  • Reporting a DCF as a single number. If you’ve built the model you’ve already got the sensitivity grid — it’s the same code in a loop. Publishing a point estimate while sitting on a range is a choice to look more certain than you are.
  • Tuning assumptions until the model agrees with the price. The most seductive error in valuation, because the result feels like confirmation. It’s the opposite: you’ve fitted the answer to the data. If you catch yourself nudging terminal growth to close a gap, stop.
  • Using a wide margin of safety to justify a business you don’t understand. The buffer covers estimation error, not ignorance.
  • Applying the same buffer to every company. A predictable business with twenty years of stable cash flows and a loss-making three-year-old startup do not warrant the same discount.
  • Forgetting the model can be wrong in the good direction. Sensitivity cuts both ways, and a business that outperforms your fade assumption is worth more than your model says. Margin of safety is about surviving errors, not about assuming the worst case is the true case.
  • Believing precision equals accuracy. ₹397.72 is precise. Whether it’s accurate depends entirely on assumptions no spreadsheet can check.

Takeaway: A DCF produces a number, but what it actually supports is a range — so run the sensitivity grid and quote the range, because that’s the honest output. Then insist on a gap between price and value big enough to absorb the fact that you’ll be wrong about something. The margin of safety isn’t a way of being cleverer than the market; it’s an admission that you won’t be.