06.4 · DCF in practice: sensitivity, scenarios, mistakes¶

Why this matters: a DCF that produces one number is a false precision machine. The same model, used properly, produces a distribution — a base case, the range around it, the scenarios that bound it, and a clear statement of which assumptions matter. That is what an investment decision needs. This lesson turns the Kaveri reference DCF into that distribution and catalogues the fifteen ways DCFs usually go wrong.

Learning objectives — after this lesson you can:

Build and read two-way sensitivity tables and know which pairs of inputs to tabulate.
Construct bull/base/bear scenarios that are internally consistent stories, assign probabilities, and compute a probability-weighted value.
Interpret the terminal value's share of EV and the "implied exit multiple" as sanity checks.
Recognise the fifteen most common DCF mistakes in your own and other people's models.
Present a DCF honestly: value as a range with named drivers, not a target price.

Prerequisites: 06.3 DCF step by step · Time: ~90 min

1. Sensitivity tables¶

A sensitivity table varies two inputs across a grid and shows the value at each combination. It answers "how wrong can I be before the conclusion changes?" The classic pair is WACC × terminal growth, because those are the inputs analysts argue about most and the ones with the least evidence behind them.

1.1 WACC × terminal growth (Kaveri, base-case operating assumptions)¶

WACC \ g	4.0%	4.5%	5.0%	5.5%	6.0%
11.2%	₹350	₹359	₹369	₹381	₹395
11.7%	₹324	₹331	₹339	₹348	₹359
12.2%	₹301	₹307	₹313	₹320	₹328
12.7%	₹281	₹285	₹290	₹296	₹302
13.2%	₹263	₹266	₹270	₹274	₹279

from fi.valuation import kaveri_reference
print(kaveri_reference()["sensitivity"].round(0))

Reading it: the entire grid — every combination of a plausible WACC (11–13%) and terminal growth (4–6%) — sits below the ₹390 price. That is a stronger statement than "the base case is ₹320": it says no reasonable discount-rate or terminal-growth choice rescues the price with these operating assumptions. If the price is right, the operating forecast must be wrong. That directs the next table.

1.2 Growth × margin (the operating pair)¶

10-yr revenue CAGR \ terminal EBITDA margin	13.5%	14.5%	15.5%	16.5%	17.5%
8.1%	₹214	₹241	₹268	₹295	₹322
10.8% (base)	₹255	₹287	₹320	₹352	₹385
12.9%	₹293	₹331	₹369	₹407	₹445
15.1%	₹338	₹383	₹427	₹471	₹515

(Grid generated with fi.valuation.sensitivity over dcf_from_drivers; growth rows scale the base path by 0.75×, 1×, 1.2× and 1.4×; margin columns shift the whole margin path.) Here the price is reachable: roughly 13% growth with a 16% margin, or 15% growth with ~14.5%. Now the debate is concrete — is 13–15% ten-year growth with margins at the best peer's level a base case or a bull case?

from fi.valuation import dcf_from_drivers, kaveri_base_drivers, sensitivity
d = kaveri_base_drivers()
def val(cagr_scale, margin):
    g = [x * cagr_scale for x in d["growth"]]
    m = [margin - (0.155 - x) for x in d["ebitda_margin"]]    # shift the whole margin path
    return dcf_from_drivers(**{**d, "growth": g, "ebitda_margin": m})["per_share"]
tbl = sensitivity(val, rows={"cagr_scale": [0.75, 1.0, 1.2, 1.4]}, cols={"margin": [0.135, 0.145, 0.155, 0.165, 0.175]})

Rules for sensitivity tables: pick pairs that are independent (WACC and g; growth and margin) rather than pairs that move together; centre the grid on the base case; extend it far enough to include the market price so the reader can see what it takes; and never show a table without saying which cell you believe.

2. Scenarios¶

Sensitivities vary inputs mechanically. Scenarios vary them together as coherent stories, which is how the world actually moves: if state dues stay stuck, growth is lower and margins are lower and working capital is higher, all at once.

2.1 Kaveri's three scenarios (from the reference valuation)¶

	Bear	Base	Bull
Story	The two state agencies delay further; Kaveri slows solar bidding; provisions; motors ramp stalls; copper stays high	Dues clear over 2 years; solar grows selectively; Hosur ramps to 75–80% utilisation; margins recover toward FY24	Dues clear within a year; a large tender pipeline; industrial motors win share; margins reach the best peer's
Revenue growth path	3% FY27, then 7–8%, fading to 5%	10%, 14%, 14%, 13%, … 7%	15%, 17%, 16%, 15%, … 8%
EBITDA margin	12.0% → 13.0%	13.3% → 15.5%	14.0% → 16.5%
NWC / revenue	29% → 27%	25% → 22%	23% → 19%
WACC, g, RONIC	unchanged	12.19%, 5.5%, 18%	unchanged
Value / share	₹168	₹320	₹416
Probability	25%	50%	25%

\[\text{Probability-weighted value} = 0.25 \times 168 + 0.50 \times 320 + 0.25 \times 416 = \mathbf{₹306}\]

r = kaveri_reference()
print({k: round(v) for k, v in r["scenarios"].items()}, round(r["probability_weighted"]))
# {'Bull': 416, 'Base': 320, 'Bear': 168} 306

Three design rules:

Scenarios are stories, not haircuts. "Bear = base × 0.8" is not a scenario. Each one should name the specific things that happen and derive the drivers from them.
Keep WACC constant across scenarios. Risk is being handled in the cash flows; changing the discount rate too double-counts (06.2).
Probabilities are a judgement — write down why. 25/50/25 is a default; Kaveri's say-do record (05.5) and governance score (05.6) might justify 35/45/20, which gives 0.35 × 168 + 0.45 × 320 + 0.20 × 416 = ₹286. The point is that the weights are visible and arguable.

2.2 What the scenarios tell you that the base case cannot¶

The downside (₹168, −57% from ₹390) is far larger than the upside (₹416, +7%). At ₹390 the payoff is asymmetric against you: a 25% chance of losing half against a 25% chance of a single-digit gain. This is the margin-of-safety argument of 06.8 in numbers.
The base case alone (₹320) would be a "watch" verdict; the distribution makes it an "avoid at this price".
The bear case identifies the monitoring variables: state payment news, receivable days, provisioning, Hosur utilisation. If those move, the probabilities move, and so does the value.

3. Sanity checks on any DCF¶

Check	Kaveri reference	Reading
Terminal value as % of EV	55%	Normal for a 10-year growth fade; > 75% = the explicit forecast is decorative
Implied exit EV/EBITDA	6.4x FY36 EBITDA	Reasonable for a 5.5%-growth industrial; if it came out at 15x, the terminal inputs are too generous
Implied terminal P/E	TV / (NOPAT₃₆ × (1+g)) ≈ 3,639 / 350 ≈ 10.4x forward NOPAT	Consistent with 1/(WACC − g) × (1 − g/RONIC)
Reinvestment consistency	Terminal reinvestment 30.6% of NOPAT for 5.5% growth at 18% RONIC	If the model reinvested 10% and grew 5.5%, it would be claiming a 55% RONIC
Year-10 ROIC	NOPAT₃₆ / IC₃₆ — check it is not absurd (> 40%) or below WACC	Kaveri's FY36 NOPAT of ₹332 Cr on NWC of ₹807 Cr plus a fixed-asset base growing with capex ≈ D&A implies a ROIC in the high teens — near the FY24 peak, consistent with the 18% terminal RONIC
FCFF vs history	FY27 FCFF ₹109 Cr vs FY26 actual FCF ₹13 Cr	Flag it: the model needs the working-capital story to be true
Growth vs industry	10.8% 10-year CAGR vs industry mid-single digits ex-solar	Requires share gain or solar; say so

4. The fifteen most common DCF mistakes¶

#	Mistake	Why it is wrong	Fix
1	Terminal g ≥ WACC, or g > long-run nominal GDP	Infinite or implausible value	g ≤ 5–6% in ₹; below WACC by a margin
2	Growth without reinvestment	Free growth; violates g = RR × ROIC	Use the value-driver terminal; check capex/NWC against growth
3	Exit multiple as the terminal	Imports today's market pricing; usually a growth multiple applied to a mature year	Value-driver formula; use the multiple only as a check
4	Explicit period too short (5 years for a 15% grower)	Fade never happens; terminal carries 80%+	10 years, with a visible fade
5	Hockey-stick margins	Margins jump above every peer with no mechanism	Cap at the best peer; require a driver for every step
6	Double-counting risk	High WACC and bear scenarios and a big haircut	Systematic risk in WACC; specific risk in scenarios; estimation error in margin of safety
7	Inconsistent inflation	Real growth with nominal WACC, or USD rates with ₹ flows	Same currency and inflation basis throughout
8	Ignoring dilution	ESOPs, warrants, convertibles left out	Diluted share count via treasury-stock method; warrant cash in the bridge
9	Book debt used blindly	Missing leases, guarantees, disputed taxes; or subtracting gross debt without adding cash	Full bridge (06.3 §7)
10	Circular WACC	Weights depend on the equity value being computed	Target weights, fixed
11	Other income in FCFF and cash in the bridge	Double-counts the treasury	Exclude other income from EBIT; add cash once
12	Mid-year convention misapplied	Applied to TV, or forgotten for FCFF	FCFF at t − 0.5; TV at N
13	Valuation date ≠ price date	Comparing a March value with a September price	Roll forward at \(k_e\), or restate the bridge to the latest balance sheet
14	Working capital as a % of revenue growth rather than revenue	Understates the cash absorbed	Model NWC as a level (days or % of revenue); ΔNWC falls out
15	Reverse-engineering inputs to the price	The model becomes a justification	Set inputs first, from evidence; run the reverse DCF separately and openly (06.6)

Number 15 is the one that matters. Every other mistake is a technical error; this one is a discipline failure, and it is the default behaviour of anyone who already owns the stock.

5. Presenting a DCF¶

A DCF output for a memo (11.3) is one table and three sentences:

	Bear	Base	Bull	Prob.-weighted	Price
Value / share	₹168	₹320	₹416	₹306	₹390
Probability	25%	50%	25%
Key assumption	State dues stuck; margin 12–13%	Dues clear in 2 yrs; margin → 15.5%	Dues clear in 1 yr; margin → 16.5%

Value is most sensitive to the terminal margin and return on new capital (±₹40/share each), then WACC (±₹27 per 50 bps); the working-capital normalisation is worth ₹23/share and is the assumption the next two quarters will test. The market price of ₹390 requires the bull case's operating outcome at the base case's discount rate, or a ~14% ten-year growth rate. At the current price the downside (−57%) exceeds the upside (+7%) by a wide margin.

That is the whole output. A single target price with two decimals is a worse product than this, however much work went into it.

Trader's lens

Scenario weighting is expected-value pricing, and the asymmetry table is the payoff diagram. A trader would never take a position with a 25% chance of −57% and a 25% chance of +7% unless paid a large premium to do so — and here there is no premium; you are paying ₹390 for an expected ₹306. When the market's price sits above your probability-weighted value, you are short a put you did not sell. Reverse DCF tells you the strike.

Common mistakes

Presenting the base case as "the" value.
Sensitivity tables that vary only WACC and g — the operating inputs usually matter more.
Scenarios built as percentage haircuts rather than stories.
Changing WACC across scenarios.
Probabilities that are secretly chosen to make the weighted value land near the price.

Key terms¶

Term	Meaning
Sensitivity table	A grid of values across two inputs, holding everything else constant
Scenario	A coherent set of assumptions describing one way the future could unfold
Probability-weighted value	Σ (probability × scenario value)
Payoff asymmetry	The ratio of upside to downside from the current price across scenarios
Terminal value share	PV of terminal value ÷ enterprise value
Implied exit multiple	Terminal value ÷ final-year EBITDA (or EBIT); a check on terminal assumptions
Hockey stick	A forecast in which margins or growth jump implausibly after a flat history
Double-counting risk	Penalising the same risk in the discount rate, the cash flows and the margin of safety
Roll-forward	Moving a valuation from its date to today at the cost of equity
Tornado chart	A ranked bar chart of value sensitivity to each input (see 10.4)

Check your understanding¶

From the WACC × g table, what is the value at 11.7% WACC and 5.0% g, and what does it mean that the whole table is below ₹390?

Answer

₹339. It means no plausible discount-rate/terminal-growth combination justifies the price under the base operating forecast; the price can only be justified by better operating outcomes (faster growth, higher margins, quicker working-capital recovery).

Recompute the probability-weighted value with 35/45/20 weights. What single fact would justify moving the weights that way?

Answer

0.35 × 168 + 0.45 × 320 + 0.20 × 416 = 58.8 + 144.0 + 83.2 = ₹286. A fourth consecutive quarter of rising >6-month overdues, or a state agency formally disputing a claim, would justify raising the bear probability.

Why should WACC be held constant across the bear, base and bull scenarios?

Answer

The scenarios already capture the company-specific risk in the cash flows; raising WACC in the bear case would count the same risk twice. WACC reflects systematic risk, which does not change with the company's receivables outcome.

A model shows terminal value at 82% of EV and an implied exit multiple of 16x EBITDA for a company growing 6% in the terminal year. What has gone wrong?

Answer

Probably a short explicit period (the fade never happened), a terminal growth or RONIC too high, or an exit multiple pulled from today's market. A 6%-growth mature company should not command 16x EBITDA in perpetuity; the terminal assumptions need to be rebuilt with the value-driver formula.

Kaveri's FY27 FCFF in the model is ₹109 Cr. Write the sentence you would put in the memo about this number.

Answer

"The model's FY27 free cash flow of ~₹109 Cr compares with ₹13 Cr in FY26 and depends almost entirely on receivable days falling from 96 toward 80; if Q2–Q3 FY27 results do not show this, the base case should be re-weighted toward the bear case."

Go deeper¶

McKinsey, Valuation, chapter on "Using Multiples" and the scenario-analysis material — the source of the sanity-check list.
Aswath Damodaran, "The Dark Side of Valuation" lectures and his blog posts on scenario vs simulation approaches.
10.4 Scenarios, sensitivities & model QA — the spreadsheet/Python mechanics (data tables, tornado charts, stress tests).
06.8 — from scenarios to a distribution and a position size.

← Previous: 06.3 DCF step by step · Module index · Next: 06.5 Relative valuation & multiples →