Risk, Return & the Equity Risk Premium

Phase 3 · Week 27 · ~20 hours

Where it sits. Second module of the valuation phase. The time-value lessons gave you the discounting machine: PV, annuities, perpetuities, the Gordon growth formula. But it left one dial unset, the rate. Every present value you computed there assumed someone handed you "10%." This week is where that number starts to come from. It formalizes the risk-return intuition planted in the literacy bootcamp (the rate stack: waiting + inflation + doubt) into tools you can compute with: expected return and dispersion, diversification arithmetic, systematic versus idiosyncratic risk, CAPM and beta, and the module's center of gravity, the equity risk premium: the market price of a unit of average equity risk, estimated three ways, with Damodaran's implied method and country-risk build taught to full working depth. Next comes the cost-of-capital build, which assembles these parts (plus bottom-up betas and the cost of debt) into a complete WACC. On the five-questions spine (see the master map), this week's work serves question 5 (what is the price implying, and is that beatable?), because "what the price implies" is always measured against a required return, and the required return is what we price here. It also arms question 2: the ~12% hurdle you have been using since Phase 2's ROIC work stops being a rule of thumb and becomes a number you can defend.

Suggested hour budget. Core content: 8h · worked examples: 2h · practice set: 5h · mini-project: 3h · flashcards + mastery check: 2h.

App link. These lessons power the Corporate Finance branch: portfolio-σ and beta computation drills, "which of these risks is priced?" recognition cards, the implied-ERP solver challenge (iterate to the market's IRR), and the country-risk-premium builder. The valuation sandbox's consistency checks (terminal growth ≤ risk-free rate, discount rate consistent with the current ERP card) are seeded by what you learn here.

Learning objectives

By the end you can:

1. Compute expected return, variance, and standard deviation from a scenario table or a return history, and explain in plain words what σ measures, why it is built from squared deviations, and where its "±1σ about two-thirds of the time" reading breaks down.
2. Run two-asset portfolio arithmetic cold (expected return as the weighted average, portfolio σ from the full formula at any correlation, the minimum-variance weights) and state precisely why portfolio risk is less than the average of risks whenever correlation is below +1.
3. Explain systematic vs idiosyncratic risk with the n-stock variance floor, and reproduce the argument for why markets pay a premium only for the risk diversification cannot remove.
4. Sketch the Markowitz efficient frontier from a numeric example (no matrix algebra), add the risk-free asset, and explain how two-fund separation leads to the market portfolio.
5. Derive CAPM's logic intuitively (beta as covariance with the market divided by market variance, Re = Rf + β × ERP), compute a beta by hand from a small return table, and give a fair, two-sided account of CAPM's empirical record (flat SML, Fama-French, unstable betas, Roll's critique) and why practitioners still use it.
6. Estimate the equity risk premium three ways: survey, historical (arithmetic vs geometric, standard error, survivorship, and why India's own history is nearly unusable), and implied (solve the market's IRR from index level and expected cash flows, Damodaran's method). State each estimate's failure modes.
7. Build a country risk premium for India with Damodaran's method (sovereign default spread × relative equity volatility), apply it with revenue-weighting (λ) so exporters and domestic businesses get different premiums, and avoid the double-counting trap with the risk-free rate.
8. Pull every live input from its named source: Damodaran's implied-ERP and country-risk pages, FRED for US Treasury yields, RBI/CCIL for G-sec yields. Maintain a dated "ERP card" in your knowledge system, refreshed every January.

Prerequisites & connections

Builds on. M0.03 gave you the rate stack (waiting + inflation + doubt), the idea that risk premiums emerge from price discounts, and arithmetic versus geometric averages with their compounding drag. All of it gets formalised here. M3.01 gave you PV mechanics, perpetuities and the Gordon growth formula, plus real versus nominal via Fisher, and the implied-ERP method is simply a Gordon or two-stage model run in reverse. M2.03 set ROIC against a cost-of-capital hurdle, and this week you build that hurdle. Phase 1 counts too, because you'll read "dividends + buybacks" off cash flow statements without blinking.

Feeds forward. M3.03 consumes everything here: the ERP and CRP become inputs to the full WACC build, beta gets rebuilt bottom-up from business fundamentals, and the clean-INR risk-free fix previewed here is executed properly there. M3.04–M3.05 discount cash flows at the rates you learn to price this week, and the reverse DCF of M3.05 is the implied-ERP idea applied to a single stock. M7 (macro) explains why the risk-free rate and the ERP move. Phase 9 comes back to diversification, this time as a portfolio decision about concentration versus diversification and hidden correlation, and leaves the pricing argument behind. The mini-project's ERP card becomes a standing input to every valuation you build for the rest of the programme.

In short. Investors hold a diversified portfolio, so the only risk they can charge for is co-movement with everything else. That risk has a market-wide price, the equity risk premium. It's a live number that moves, so you observe it by inverting today's prices instead of copying a textbook constant, and a country like India adds a measurable surcharge on top.

Core teaching content

4.1 The question this module prices

At a 10% discount rate, ₹100 crore arriving in year 10 is worth ₹38.6 crore today. At 13%, it is worth ₹29.5 crore. A quarter of the value, gone from three points on one dial.

Strip valuation to its skeleton: value = expected cash flows, discounted at a rate that compensates for waiting and for doubt. Phase 2 taught you to judge the cash flows, and the time-value lessons taught you to discount. That leaves the doubt dial, and it's worth being clear about what turning it does. Get the rate badly wrong and no amount of careful forecasting saves the valuation.

You met the intuition in the literacy bootcamp as one line: nominal expected return ≈ real riskless rate + expected inflation + risk premium. You can observe the first two terms, because a government bond yield bundles them for you. All the difficulty is in the third term, and the work ahead answers three questions about it, in order:

1. What is being paid for? Not dispersion itself, only the part of dispersion you can't diversify away.
2. How much does one unit of that risk cost? It costs the equity risk premium, the market's asking price for bearing average-stock risk.
3. How does the answer change across countries and companies? Country risk premiums scale it up for India, and beta scales it up or down for each company.

By the end, the phrase "the cost of equity for this Indian business is about 12%" stops being something you quote and becomes something you can construct: from a G-sec yield, a default spread, a volatility ratio, a beta, and a premium you solved from the index yourself.

One guardrail before the mathematics, and it stays up all week. Everything here describes how risk is priced by diversified investors in liquid markets. Whether volatility is what you should fear, as opposed to permanent loss of purchasing power, is a live debate argued in good faith on both sides. Marks and Buffett say it isn't. Phase 6 and Phase 9 give their side a full hearing. Both framings are worth having. You're learning the market's pricing grammar here because valuation is conducted in it.

4.3 The only free lunch: two assets and a correlation

Now the theorem that reorganised finance. Portfolios average returns. But whenever assets don't move in lockstep, they carry sub-average risk. Harry Markowitz got a Nobel for working out the arithmetic, and the arithmetic itself needs nothing beyond what you just learned.

Start with correlation, as an idea. The correlation coefficient ρ ("rho") runs from +1 to −1 and answers one question: when A has an above-average period, what does B tend to do? At ρ = +1, B is above average in lockstep, like two ice-cream stalls on the same beach. At ρ = 0, B's result tells you nothing, as with an ice-cream stall and a pharmacy. At ρ = −1, B is reliably below average when A is above, like an airline and an oil producer in the crude-price dimension. Fuel spikes crush one and enrich the other. Formally, ρ is the covariance, the average product of the two assets' deviations from their means, rescaled by both σs to land in [−1, +1]. You'll compute a covariance by hand in Worked Example 2 and see that it's nothing exotic. Within one equity market, typical large-cap pairs run ρ ≈ +0.2 to +0.6. They share an economy, a rate cycle, a currency, but not customers, plants or scandals.

Here are the two-asset formulas, with weight w on A and (1−w) on B:

Expected return:  E(Rp) = w·E(RA) + (1−w)·E(RB)                       (a plain weighted average)
Variance:         σp² = w²σA² + (1−w)²σB² + 2·w·(1−w)·ρ·σA·σB        (NOT a weighted average)

The return line holds at any correlation, because returns simply blend. All the action is in the third term of the variance line, the co-movement term, which is the only place ρ appears. Lower ρ means a smaller term and smaller portfolio risk, at no cost in expected return.

Put numbers on it. Let A be Steady Staples, with E(R) of 10% and σ of 18%, and B be Cyclical Motors, with E(R) of 14% and σ of 28%. Hold them 50/50. The expected return is 12% whatever ρ is. Risk is another matter, as the table shows:

Check the ρ = +0.3 row yourself, because this computation has to become automatic: σp² = 0.25(324) + 0.25(784) + 2(0.25)(0.3)(18)(28) = 81 + 196 + 75.6 = 352.6 → σp = 18.8%.

Read the table carefully. At ρ = +1 there's no free lunch. Risk is the weighted average of risks, and portfolios are just blends. At every ρ below +1, portfolio risk falls below the weighted average while the expected return stays put. That wedge is the free lunch, and it's the only one in finance. You gave up nothing and you carry less risk. Perfectly negative correlation is a textbook limit. You'll never find it in equities, and hedges built from derivatives approximate it only at a price. Real portfolios are in the realistic ρ ≈ 0.3 row, and four free points of σ is a large gift.

One more step while the machinery is warm. What mix of these two assets minimises risk? Calculus, or a spreadsheet's trial column, gives the minimum-variance weights:

w(A, min-var) = (σB² − ρσAσB) / (σA² + σB² − 2ρσAσB)

Here: (784 − 151.2)/(324 + 784 − 302.4) = 632.8/805.6 ≈ 0.786. A 78.6/21.4 mix has σ ≈ 16.9% and E(R) ≈ 10.9%. That's less risk than holding the safe asset alone, and more return. Adding the riskier asset made the cautious holder safer and better paid at the same time. The 100%-Staples investor is doing something strictly worse, and would be even if Cyclical Motors were, in isolation, "too risky for them." Judge an asset by what it does to the portfolio, because its solo statistics don't settle the question.

That single sentence is Markowitz's revolution.

There's a caveat, though. Correlations are estimates from history, and they aren't constants of nature. Their nastiest habit is rising toward +1 just when you need them low. In October 2008 and March 2020, stocks that normally co-move at 0.3 fell together as if at 0.9, so within one asset class, diversification is a fair-weather friend. It still works across enough names and enough sectors, and especially across asset classes and geographies. But the σ you compute from calm-market data understates the joint tail. Phase 9 treats this as a first-order portfolio problem.

For now it gives you one design rule: never treat a measured ρ as a promise.

4.11 The live-numbers ritual: your ERP card

Every number above decays. The method is permanent, but the values are dated. So the teaching ends with a routine, because a table would go stale. Give it thirty minutes each January, since Damodaran's big annual data refresh arrives in the first days of the year, and do a quick re-pull after any major market move (±15% on the index, a ratings action, a rate shock):

1. US implied ERP. Damodaran's site, "Damodaran Online" at NYU Stern, headlines the implied ERP on its home page and updates it monthly. For the series history, go to Current Data → Implied Equity Risk Premiums for the United States (the histimpl spreadsheet). His January "Data Update" posts on the Musings on Markets blog walk through that year's numbers with full workings. Read the ERP instalment each year, because it's a free annual masterclass.
2. India ERP, CRP, default spread, relative-vol multiplier. Go to Current Data → Country Default Spreads and Risk Premiums. On the ctryprem page and spreadsheet you'll find rating-based and CDS-based spreads, the multiplier, assembled ERPs for every country.
3. Historical US table. You want it for context and for Form-B answers to "what did equity risk pay?": Current Data → Historical Returns on Stocks, Bonds and Bills — United States (the histretSP spreadsheet).
4. US risk-free rate. FRED (St. Louis Fed), series DGS10, the 10-year Treasury constant maturity. One search box, one number, dated.
5. India risk-free raw material. You want the 10-year G-sec yield, from RBI's Database on the Indian Economy (DBIE) government-securities tables or the Weekly Statistical Supplement. CCIL (ccilindia.com) publishes traded G-sec yields daily, and financial portals work as a cross-check. Then clean it, as the country-risk build showed.
6. Log the card. Write one dated note in your knowledge system, in the format you set up with your toolkit: each value, its source page, its date, and the one-line derivation of your India ERP. Every valuation you build until next January cites this card. One place to look, one place to update, no silent staleness.

The mini-project below runs this ritual for real, today, and the card it produces becomes a standing input to the full cost-of-capital build and everything after it.

Common mistakes & how experts think differently

1. Pricing with total σ where beta belongs. "This stock is twice as volatile, so it should return twice as much" charges the market for risk a diversified holder cancels for free. Total σ matters to the undiversified, a promoter or a founder, and that's total-beta territory, which comes up when you build the full WACC. Required returns in liquid markets run on co-movement. When an expert is handed any risk, the reflex is to ask first who is holding it and what else they hold.

2. Reading beta as a danger or quality grade. A smallcap with stale prices and a fraud brewing can print β = 0.6. Thin trading dampens measured co-movement, and the fraud is idiosyncratic until it detonates. Likewise, Hindustan Unilever's β ≈ 0.5 doesn't mean an undiversified holder of it is "half as risky as the market." Beta measures one thing: systematic co-movement. Business risk, balance-sheet risk, fraud risk, permanent-loss risk are other instruments on other pages of your dashboard, mostly the ones Phase 2 built. An expert says "low beta," never "low risk," and means only the covariance.

3. Quoting the historical ERP without its error bars, or its survivorship. "The ERP is 6.94%" sounds like measurement. With SE ≈ 2 points, it's a coin resting on its edge. And the US series is the winner's history. The DMS world data run about a point lower, and the czarist-Russia investor isn't in anyone's average. Experts hold historical premia as a sanity band (≈ 4.5–5.5% mature-market), never a decimal.

4. Arithmetic/geometric confusion. Mixing an arithmetic premium (≈ 6.5–7% US) into long-horizon compounding logic, or quoting geometric (≈ 5%) as "the expected return next year," moves valuations by ~2 points of discount rate without anyone noticing. Know which your source reports (Damodaran's tables show both), know which your use case needs, and never mix vintages mid-model.

5. The frozen ERP. A 5.5% scribbled from a 2019 textbook, still discounting DCFs in 2026, has missed a pandemic, an inflation cycle, and several full repricings of risk. The ERP is a price. It moved from ≈ 2% at the 2000 peak to ≈ 6.4% at the 2009 trough, within a decade. Experts date-stamp every premium ("4.3%, Damodaran implied, 1 July 2026") and refresh it on a calendar instead of from memory.

6. Double-charging India. Use the raw G-sec yield as Rf plus the full CRP in the premium and you bill sovereign default risk twice. It costs ≈ 2 extra points of discount rate, or ≈ a fifth off a terminal value. The clean sequence is mechanical. Strip the default spread out of the "risk-free" rate, then add country risk back once through the ERP, with λ. The mirror error exists too. A USD Rf plus a mature ERP for an INR forecast is a currency mismatch, and the WACC build will arm you against it.

7. λ blindness. Full India CRP on an IT exporter whose cash flows are 70% American is as wrong as zero CRP on a London-listed consumer giant earning half its profit in emerging markets. Country risk follows the cash flows, not the ticker's postcode. Experts read the segment note before they pick a premium, which Phase 1 trained you to do.

8. Trusting calm-market correlations. The ρ = 0.3 you measured makes diversification look magical. In the crash it lurches toward 0.9, and the magic thins just when you need it. Experts size diversification benefits on stressed correlations and treat measured ρ as weather, not climate.

9. Implied-ERP mechanics slips. Three slips recur. A dividends-only cash yield in a buyback market understates the yield. That understates the premium too, and it can misread the US by a full point. Terminal growth above the risk-free rate breaks the perpetuity's economics, as the terminal-value work formalises. And solving to the wrong Rf vintage pits a July index level against January's bond yield. The method holds up. The bookkeeping has to be exact.

10. Spending the hour where the decimals live instead of where the value lives. This is the deepest difference. Amateurs burn days tuning the discount rate's second decimal and minutes on whether the moat holds, yet a point of misjudged margin durability swamps 30bp of ERP refinement. Experts set a policy for rates (stated ERP source, stated λ, stated Rf cleaning) and apply it mechanically in every memo. Their scarce attention goes on cash flows, reinvestment, and competitive position, where variant perception actually pays. They also know when the rate matters most. Long-duration, high-growth stories, with value far in the future, are exquisitely rate-sensitive. Near-cash deep value barely flinches.

Worked examples

Worked example 1: Two-asset portfolio arithmetic, end to end

Given. Steady Staples: E(R) = 10%, σ = 18%. Cyclical Motors: E(R) = 14%, σ = 28%. ρ = +0.3. All numbers are annual, and weights are 50/50 unless stated.

(a) Expected return. 0.5×10 + 0.5×14 = 12.0% at any ρ, because returns blend linearly.

(b) Portfolio σ at ρ = +0.3. σp² = (0.5)²(18)² + (0.5)²(28)² + 2(0.5)(0.5)(0.3)(18)(28) = 0.25×324 + 0.25×784 + 0.15×504 = 81 + 196 + 75.6 = 352.6 σp = √352.6 = 18.8%, versus the 23.0% weighted average of the two σs. That's 4.2 points of risk removed by imperfect correlation, for free.

(c) Re-run at ρ = +1 and ρ = 0. ρ = +1: 81 + 196 + 2(0.25)(1)(504)… = 81 + 196 + 252 = 529 → σ = 23.0%, the weighted average itself, so there's no lunch. ρ = 0: the co-movement term vanishes, so 81 + 196 = 277 → σ = 16.6%. That's below the safer asset's 18%, with 2 points more return.

(d) Minimum-variance mix. w(Staples) = (784 − 0.3×18×28) / (324 + 784 − 2×0.3×18×28) = (784 − 151.2) / (1,108 − 302.4) = 632.8 / 805.6 = 0.786 At 78.6/21.4: σp² = (0.786)²(324) + (0.214)²(784) + 2(0.786)(0.214)(0.3)(504), which gives σ ≈ 16.9% and E(R) = 0.786×10 + 0.214×14 ≈ 10.9%. Compare holding Staples alone, at 18.0% risk and 10.0% return. The mixed portfolio wins on both axes. Dominance, computed.

(e) The sentence to retain. Adding a riskier asset reduced total risk because portfolio risk is priced off covariance, not off the newcomer's solo σ.

Worked example 2: A beta by hand (and what the scatter really looks like)

Given. Six quarters of returns (%) for market M and stock S, a capital-goods maker (synthetic):

Step 1: means. Market: (8−4+12+2−6+6)/6 = 18/6 = 3.0. Stock: (16−6+10−8−12+12)/6 = 12/6 = 2.0.

Step 2: deviations from means. Market: +5, −7, +9, −1, −9, +3 Stock: +14, −8, +8, −10, −14, +10

Step 3: products of paired deviations. This is covariance being born. Same-direction quarters give positive products and opposite-direction quarters give negative ones: 5×14=70 · (−7)(−8)=56 · 9×8=72 · (−1)(−10)=10 · (−9)(−14)=126 · 3×10=30, which sum to 364.

Step 4: covariance and market variance. We use the population convention, dividing by 6, which is fine for teaching. Software divides by n−1, and with real 60-observation samples the difference is trivial. Cov = 364/6 = 60.67 Var(M) = (25+49+81+1+81+9)/6 = 246/6 = 41.0

Step 5: beta. β = 60.67 / 41.0 = **1.48**. When the market moves 1%, this stock has tended to move ~1.5%. It's an amplified, cyclical profile, just what its capital-goods economics predicts.

Step 6: how tight is the relationship? Stock σ = √(720/6) = 10.95, and market σ = √41 = 6.40. ρ = 60.67/(6.40×10.95) = 0.87, so R² ≈ 0.75, and three-quarters of this stock's variance is market-linked. Real single stocks are much noisier. An R² of 0.2–0.4 is typical, which means most of a real stock's turbulence is idiosyncratic, the kind that washes out in a portfolio, and the fitted slope wobbles with the sample. That's why the full cost-of-capital build averages betas bottom-up across a whole sector instead of trusting one stock's six-point scatter.

Worked example 3: The implied ERP, solved by hand (Damodaran's method in miniature)

Given (synthetic, US-flavoured). An index stands at 5,000. Last year it delivered 200 in dividends + buybacks, a 4.0% cash yield. Aggregate analyst expectations have cash flows growing 5% a year for five years, after which the market grows with the economy at the risk-free rate. The 10-year Treasury yields 4.2%. Solve for the return the market is pricing, and the implied ERP.

Step 1: lay out expected cash flows. CF(t) = 200 × 1.05^t:

Terminal value at end of year 5 (Gordon, g = Rf = 4.2%): TV = CF₅ × 1.042 / (r − 0.042) = 265.98 / (r − 0.042).

Step 2: the equation. Find r such that Σ CF(t)/(1+r)^t + TV/(1+r)^5 = 5,000.

Step 3: iterate. Start at 8%. The discounted cash flows for years 1–5 sum to 920, and the terminal term gives 265.98/0.038 = 6,999.5, worth 6,999.5/1.08^5 = 4,763 today, for a total of about 5,683. That value is too high, so the market must be pricing a higher return. Try 9%. The CFs come to about 896 and TV = 265.98/0.048 = 5,541 with a PV of about 3,601, so the total is about 4,497. Too low. Try 8.5%. The CF PVs are 193.5 + 187.3 + 181.3 + 175.4 + 169.8 = 907.3, TV = 265.98/0.043 = 6,185.6 and its PV is 6,185.6/1.085^5 = 4,113.6, which makes the total about 5,021. A whisker high.

Interpolating between 8.5% (5,021) and 9.0% (4,497): r ≈ 8.5% + 0.5% × (5,021−5,000)/(5,021−4,497) = 8.5% + 0.02% ≈ **8.52%**.

Step 4: the premium. Implied ERP = 8.52% − 4.20% = **4.3%**. That single number is the market's current asking price for average equity risk, and you pulled it out of today's prices instead of your memory.

Step 5: watch it behave like a price. Suppose the index crashes 20% to 4,000 tomorrow while expected cash flows hold. Re-solve and you get r ≈ 9.6% and an implied ERP of ≈ 5.4%. The crash raised the compensation on offer by over a point. The implied premium reads fear directly off the tape, and a historical average can never do that. When you meet reverse DCF later, asking "what does this stock's price imply?", you'll recognise it. It's this same inversion, pointed at one company.

Worked example 4: The India stack: from G-sec to two costs of equity

Given (all inputs are illustrative and undated, so check each against the live-numbers ritual's sources before real use). The 10-year G-sec yields 6.5%. Moody's rates India Baa3, which maps to a default spread of 2.0%. The EM relative-volatility multiplier is 1.4, the mature-market (US implied) ERP is 4.3%, and the US 10-year yields 4.2%.

(a) Build India's ERP. CRP = 2.0% × 1.4 = 2.8%, so ERP(India) = 4.3% + 2.8% = 7.1%

(b) Clean the rupee risk-free rate. The G-sec is a Baa3 credit's paper and isn't default-free, so Rf(INR, clean) = 6.5% − 2.0% = 4.5%.

(c) A domestic business. Model it on a mid-size private-sector bank, β = 1.0, λ = 1 (all-India cash flows): Re = 4.5% + 1.0 × 7.1% = **11.6%** (INR) Done naively, with the raw G-sec plus the full CRP, you'd get 6.5% + 7.1% = 13.6%. This is the 2-point double-count from the country-risk build. At a Gordon terminal (g = 4.5%) it's the difference between valuing next-year cash flow at 14.1× and at 11.0×, a fifth of the terminal value lost to bookkeeping.

(d) An exporter. Model it on a large IT-services firm of TCS's scale, with ~70% developed-market revenue, so λ ≈ 0.3, and β = 0.9: Firm ERP = 4.3% + 0.3 × 2.8% = 5.14%, so Re = 4.5% + 0.9 × 5.14% = **9.1%** (INR) Two Indian-listed companies, 2.5 points apart, because their cash flows come from different economies.

(e) A US comparison. Model it on a mega-cap with β = 1.15: Re = 4.2% + 1.15 × 4.3% = **9.1%** (USD) It looks identical to (d), but it isn't comparable. One is a rupee rate and one is a dollar rate, and the currencies carry different inflation, which is Fisher again. Never rank required returns across currencies. The WACC build gives you the conversion method. The India example runs ~4.5% inflation expectations inside its Rf-and-premium stack, against ~2.3% for the US. Most of any INR-vs-USD rate gap is inflation, not opportunity.

Worked example 5: A bottom-up (levered) beta, and the INR cost of equity it builds (India)

Given (synthetic, and a preview of the full cost-of-capital machine). Bharat Components (synthetic) is an Indian auto-parts maker with a target debt-to-equity ratio D/E = 0.50 and a 25% tax rate. Its own regression beta is uselessly unstable. It's 1.05 on five years of monthly data and 1.55 on two years of weekly, which is CAPM's third empirical failure in action. That swing by itself moves the cost of equity from 4.5% + 1.05×7.1% = 11.96% to 4.5% + 1.55×7.1% = 15.5%. So build beta from the sector instead. Three listed peers carry published levered betas and their own leverage, and you have clean INR Rf = 4.5% and India ERP = 7.1% from WE4's stack.

(a) Strip each peer's leverage out, giving the asset (unlevered) beta. Debt amplifies equity beta, so remove it with βU = βL / [1 + (1−tax)×(D/E)], the Hamada relation, which the full cost-of-capital build derives: Peer 1: 1.20 / (1 + 0.75×0.35) = 0.951 · Peer 2: 1.45 / (1 + 0.75×0.65) = 0.975 · Peer 3: 0.95 / (1 + 0.75×0.10) = 0.884 Three noisy levered numbers collapse to a tight band of pure business risk, with an average asset β ≈ 0.936.

(b) Re-lever at Bharat's own capital structure. βL = 0.936 × [1 + (1−0.25)×0.50] = 0.936 × 1.375 = **1.287**.

(c) Price it through the India stack. Re = 4.5% + 1.287 × 7.1% = **13.64%** (INR).

(d) The point. The bottom-up 13.64% lies sensibly inside the 11.96–15.5% range Bharat's own regressions spat out. But it was triangulated from the sector's economics and Bharat's actual leverage, not one company's six-quarter scatter. When one regression is noise and the sector is signal, build up from the sector. Building the full WACC turns this into routine.

Worked example 6: Decomposing a drawdown: cash-flow news vs risk-premium news (US)

Given (synthetic, US$, the same index as WE3). The index stands at 5,000. Expected cash flows (dividends + buybacks) were 200 last year, growing 5% for five years and then at the risk-free rate, and the 10-year Treasury yields 4.2%. WE3 solved this market's implied return at ≈8.52% and its implied ERP at ≈4.32%. Now, in one week, the index falls to 4,300 (−14%) and analysts cut expected cash flows 6% on a weakening economy. Two different things moved. Separate them.

(a) The base (from WE3). Market IRR r₀ ≈ 8.52% and implied ERP ≈ 4.32% (trials: 5,683 at 8.0%, 5,021 at 8.5%, 4,497 at 9.0%, interpolated to 5,000).

(b) Isolate the cash-flow news. Freeze the required return at r₀ and cut every cash flow 6%. The valuation is linear in the cash flows, so the fair level falls in exact proportion: 5,000 × 0.94 = **4,700**. That 300-point fall, from 5,000 to 4,700, is cash-flow news. It's where the index would be if only the earnings view had changed.

(c) Whatever remains is risk-premium news. The index is actually at 4,300, below the 4,700 that cash-flow news alone justifies. Re-solve for the rate that discounts the cut cash flows to 4,300. The trials give 4,720 at 8.5% and 4,227 at 9.0%, interpolation gives 8.5% + 0.5% × (4,720−4,300)/(4,720−4,227) ≈ 8.93%, and an exact solve gives r₁ ≈ 8.92%. New implied ERP = 8.92% − 4.2% ≈ **4.72%**.

(d) The decomposition. Of the 700-point drop, 300 is cash-flow news and 400 is risk-premium news. The risk-premium part shows up as the implied ERP climbing 4.32% → 4.72% (+0.4 points). The market both marked down what companies will pay and raised its price for equity risk, and only the implied method tells them apart. Run this on a real selloff and you'll own the institutional habit the implied-premium discussion named. You'll meet the same inversion again, pointed at one stock, as the reverse DCF.

Practice set

Work every problem before reading its solution. The set runs from guided (P1–P4) to independent (P5–P9), then integrated (P10–P11) and timed (P12).

P1 (guided). A stock has three scenarios: +28% with probability 0.25, +12% with probability 0.50, −10% with probability 0.25. Compute E(R), variance, and σ.

Solution. E(R) = 0.25(28) + 0.5(12) + 0.25(−10) = 7 + 6 − 2.5 = 10.5%. Deviations: +17.5, +1.5, −20.5. Squared, weighted: 0.25(306.25) + 0.5(2.25) + 0.25(420.25) = 76.56 + 1.13 + 105.06 = 182.75. σ = √182.75 ≈ 13.5%.

P2 (guided). Asset X: E(R) 9%, σ 15. Asset Y: E(R) 13%, σ 25. Weights 60/40, ρ = +0.2. Compute portfolio E(R) and σ. Then recompute σ at ρ = +1 and state the diversification gain in points of σ.

Solution. E(Rp) = 0.6(9) + 0.4(13) = 10.6%. σp² = 0.36(225) + 0.16(625) + 2(0.6)(0.4)(0.2)(15)(25) = 81 + 100 + 36 = 217 → σp = 14.7%. At ρ = +1: σ = 0.6(15) + 0.4(25) = 19.0% (weighted average). Gain: 19.0 − 14.7 = 4.3 points of σ, at identical expected return.

P3 (guided). For P2's assets (ρ = +0.2), find the minimum-variance weight in X and the portfolio σ there.

Solution. w(X) = (625 − 0.2×15×25)/(225 + 625 − 2×0.2×15×25) = (625 − 75)/(850 − 150) = 550/700 = 0.786. σp² = (0.786)²(225) + (0.214)²(625) + 2(0.786)(0.214)(75) = 139.0 + 28.7 + 25.2 ≈ 192.9 → σp ≈ 13.9%. That's below X's own 15%, with E(R) ≈ 0.786(9)+0.214(13) = 9.9% > 9%, so the mix dominates holding X alone.

P4 (guided). Classify each as systematic or idiosyncratic for a diversified Indian equity investor, one clause of reasoning each: (a) RBI hikes the repo rate 50bp, (b) a USFDA import alert on one pharma company's Goa plant, (c) monsoon fails nationally, (d) a promoter is arrested for fraud, (e) Brent crude jumps 40%, (f) a two-week strike at one auto plant.

Solution. (a) Systematic: repricing hits every rupee cash flow's discount rate and credit conditions economy-wide. (b) Idiosyncratic: one firm's plant, so it washes out across a portfolio (sector-wide only if peers share the issue). (c) Systematic (within India): rural demand, food inflation, rate policy all move. (d) Idiosyncratic: firm-specific governance detonation, and a portfolio of 40 names shrugs. (e) Systematic for India: an oil importer's currency, inflation, current account, and margins move together. (f) Idiosyncratic: one plant, one quarter. Note the pattern. Systematic events transmit through prices everyone faces (rates, fuel, currency), while idiosyncratic events stay inside one company's walls.

P5. From this data, compute beta: Cov(stock, market) = 54 (%²), Var(market) = 45 (%²). Then interpret β for a reader in one sentence.

Solution. β = 54/45 = 1.2. Interpretation: when the market moves 1%, this stock has tended to move about 1.2% in the same direction. It's modestly amplified systematic exposure, so its required return premium should be 1.2× the market's.

P6. Cost of equity, two companies (inputs given, all annual): (a) Indian consumer firm: clean INR Rf 4.5%, β 1.1, India ERP 7.1%. (b) US industrial: Rf 4.2%, β 0.9, US ERP 4.3%. Compute both. Then explain in two sentences why the 4-point gap does NOT mean the Indian firm is the worse investment.

Solution. (a) 4.5 + 1.1×7.1 = 12.31% ≈ 12.3% (INR). (b) 4.2 + 0.9×4.3 = 8.07% ≈ 8.1% (USD). The gap is a hurdle, not a verdict. The Indian firm's cash flows are in rupees (higher expected inflation) and carry country risk, so its 12.3% is the fair asking price for those conditions, just as 8.1% is for the US firm's. Value depends on whether each firm's cash flows and growth clear its own hurdle. An India business compounding revenue at 15% against a 12.3% bar can be a far better business than a US firm growing 3% against 8.1%.

P7. A market's excess returns over bonds for five years: +35%, −10%, +22%, +8%, −5%. Compute the arithmetic and geometric mean premium. State which you'd pair with a long-horizon DCF and why, then reconcile the gap with the σ²/2 rule qualitatively.

Solution. Arithmetic: (35 − 10 + 22 + 8 − 5)/5 = 50/5 = 10.0%. Geometric: (1.35 × 0.90 × 1.22 × 1.08 × 0.95)^(1/5) − 1 = (1.5208)^(0.2) − 1 ≈ 8.75%. For discounting multi-decade cash flows, the geometric mean (or a long-window blend leaning geometric) matches the compounding nature of the exercise. Arithmetic answers "one independent year ahead." The 1.25-point gap is the dispersion drag. These returns are volatile, and arithmetic ≈ geometric + σ²/2 predicts a gap of roughly this size at σ ≈ 17%.

P8. A stable, mature market index trades at 1,000. Next year's dividends + buybacks are expected at 35, growing 4% forever. The 10-year government bond yields 4.0%. Implied expected return and implied ERP?

Solution. Gordon inverted: r = CF₁/P + g = 35/1,000 + 0.04 = 7.5%. Implied ERP = 7.5 − 4.0 = 3.5%. That's the eyeball version, one line you can run on any market in ten seconds.

P9. An index stands at 1,200. Expected cash flows: 48 next year, growing 6% in years 2–3 (i.e., CF₂ = 50.88, CF₃ = 53.93), then 4% forever. Solve the implied return by trying 8% and 9%, then interpolating. The bond yield is 4.0%. What's the implied ERP?

Solution. TV at end year 3 = 53.93 × 1.04/(r − 0.04) = 56.09/(r − 0.04). At r = 8%, the CF PVs are 44.44 + 43.62 + 42.81 = 130.88. TV = 56.09/0.04 = 1,402.3, with PV = 1,402.3/1.2597 = 1,113.2, so the total is about 1,244. It's above 1,200, so r must be higher. At r = 9%, the CF PVs are 44.04 + 42.83 + 41.65 = 128.51. TV = 56.09/0.05 = 1,121.8, with PV = 866.2, so the total is about 995, which is below. Interpolate: 8% + 1% × (1,244 − 1,200)/(1,244 − 995) = 8% + 0.18% ≈ 8.2% (exact: 8.15%). Implied ERP ≈ 8.2 − 4.0 ≈ 4.2%.

P10. Build a cost of equity: India sovereign default spread 2.0%, EM relative volatility 1.4, mature ERP 4.3%, clean INR Rf 4.5%. The company: an Indian specialty chemicals maker, β = 1.2, with 60% of revenue exported to developed markets (λ = 0.4). Compute CRP, India ERP, the firm's ERP, and Re. Then: what would a careless analyst using full CRP get, and how many points too high?

Solution. CRP = 2.0 × 1.4 = 2.8%. ERP(India) = 4.3 + 2.8 = 7.1%. Firm ERP = 4.3 + 0.4 × 2.8 = 5.42%. Re = 4.5 + 1.2 × 5.42 = 11.0% (INR). The careless version (λ = 1) gives Re = 4.5 + 1.2 × 7.1 = 13.02% ≈ 13.0%, which is 2 points too high. It materially over-discounts cash flows that mostly arise in developed markets.

P11 (short answer). Name three empirical failures of CAPM (with the associated study or anomaly name) and then give the two strongest reasons practitioners retain it for cost-of-capital work. Six sentences maximum.

Solution. Failures (any three) include the flat SML, where low-beta stocks out-earn the model and high-beta ones under-earn (Black-Jensen-Scholes 1972, monetised as Frazzini-Pedersen's Betting Against Beta). Size and value dominate beta in explaining average returns (Fama-French 1992, and their 2004 review calls the applied record poor). Momentum returns go unexplained by beta (Jegadeesh-Titman 1993). Beta instability shows up across windows/frequencies/indices. Roll's critique says the true market portfolio is unobservable, so tests are of proxies. And the Mehra-Prescott equity-premium puzzle finds the realised premium too large for the theory's risk aversion. Retention: (1) valuation needs a transparent, consistent ex-ante price of risk, not a return-forecasting machine, and CAPM's three auditable inputs give you just that. (2) The alternatives swap one noisy input for several (unstable factor premia and loadings) while the valuation's real error budget is in cash flows, so practitioners fix CAPM's inputs (bottom-up beta, implied ERP) instead of replacing the frame.

P12 (timed, 15 minutes, no notes). (a) Given: clean INR Rf 4.6%, India default spread 2.2%, relative volatility 1.35, mature ERP 4.4%, β 0.85, λ = 1. Compute the firm's INR cost of equity. (b) A 70/30 portfolio: σ₁ = 16, σ₂ = 26, ρ = 0.25. Compute portfolio σ. (c) One sentence: your implied-ERP series jumps from 4.5% to 6.0% in a quarter while earnings expectations are roughly flat: what happened, mechanically?

Solution. (a) CRP = 2.2 × 1.35 = 2.97%, so ERP = 4.4 + 2.97 = 7.37% and Re = 4.6 + 0.85 × 7.37 = 10.9%. (b) σp² = 0.49(256) + 0.09(676) + 2(0.7)(0.3)(0.25)(16)(26) = 125.44 + 60.84 + 43.68 = 229.96 → σp ≈ 15.2%. (c) Prices fell hard while expected cash flows didn't. The index level, which is the PV, dropped, so the discount rate that equates the two rose. The market repriced risk and now demands ~1.5 points more per unit of equity risk. That's what capitulation looks like in this instrument.

P13 (guided). A diversified Indian equity investor holds equal weights in n NIFTY-universe stocks. The average stock has σ = 32% and the average pairwise correlation is 0.20. (a) Compute the average variance and average covariance (in %²). (b) Compute portfolio σ at n = 1, 5, 20, and in the limit n → ∞. (c) What single quantity is the floor, and what does it represent?

Solution. (a) Average variance = 32² = 1,024, and average covariance = ρ × variance = 0.20 × 1,024 = 204.8 (%²). (b) Using σp² = (1/n)(1,024) + (1 − 1/n)(204.8): n = 1 → 1,024 → 32.0%, n = 5 → 368.64 → 19.2%, n = 20 → 245.76 → 15.7%, and n → ∞ → 204.8 → 14.3%. (c) The floor is √(average covariance) = √204.8 = **14.3%**, the systematic co-movement every added stock carries in with it. Own-variance is divided away by n. Co-movement isn't, and ~20 spread names already do most of the washout.

P14 (guided). The tangency (market) portfolio has E(R) = 11% and σ = 16%; the risk-free rate is 4%. (a) Compute its Sharpe ratio. (b) A cautious investor wants portfolio σ = 10%: what weight on the tangency portfolio (rest in the risk-free asset), and what expected return? (c) An aggressive investor wants σ = 20%: the same two answers, and what must she physically do to get there?

Solution. (a) Sharpe = (11 − 4)/16 = 0.4375. It measures extra return per unit of total risk, the slope of the capital allocation line. (b) Mixing with the risk-free asset scales σ linearly, so w = 10/16 = 0.625 in tangency (0.375 in Rf) and E(R) = 4 + 0.625 × (11 − 4) = **8.375%**. (c) w = 20/16 = 1.25, so E(R) = 4 + 1.25 × 7 = **12.75%**. A w > 1 means borrowing at Rf and levering the tangency portfolio. This is two-fund separation in action. Everyone holds the same risky bundle, and only the dose changes.

P15. US inputs: Rf = 4.2%, ERP = 4.3%. An analyst pitches two stocks. Stock J: β = 0.7, expected return 8.5%. Stock K: β = 1.5, expected return 10.0%. (a) Compute each stock's CAPM-required return (its point on the SML). (b) Which one lies above the SML with positive alpha, and by how much? (c) What documented pattern from CAPM's scorecard does this pair echo?

Solution. (a) J: 4.2 + 0.7×4.3 = 7.21%. K: 4.2 + 1.5×4.3 = 10.65%. (b) J's 8.5% expected return exceeds its 7.21% required, giving +1.29% alpha, above the SML (underpriced). K's 10.0% falls short of its 10.65% required, giving −0.65%, below the SML (overpriced). (c) It's the flat-SML anomaly. The low-beta name offers more than CAPM asks and the high-beta name less, which is the Black-Jensen-Scholes and "Betting Against Beta" result that low-beta portfolios have historically out-earned the model.

P16. An Indian domestic firm (β = 1.0, λ = 1). Inputs: 10-year G-sec yield 6.5%, EM relative-volatility multiplier 1.4, mature-market ERP 4.3%. The default spread is contested: rating-based 2.0%, CDS-based 0.8%. Build the INR cost of equity both ways, cleaning the risk-free rate each time. (a) Rating route. (b) CDS route. (c) Why are the two answers so close, and what would a careless analyst who forgets to clean the G-sec get?

Solution. Clean the Rf with the same spread you use for the CRP. (a) Rating: clean Rf = 6.5 − 2.0 = 4.5%, CRP = 2.0 × 1.4 = 2.8%, ERP = 4.3 + 2.8 = 7.1%, and Re = 4.5 + 1.0×7.1 = **11.6%**. (b) CDS: clean Rf = 6.5 − 0.8 = 5.7%, CRP = 0.8 × 1.4 = 1.12%, ERP = 4.3 + 1.12 = 5.42%, and Re = 5.7 + 1.0×5.42 = **11.1%**. (c) Once you clean, Re = G-sec + mature ERP + spread×(multiplier − 1) = 6.5 + 4.3 + spread×0.4, so the spread's only net effect is the small ×0.4 term and the rating-vs-CDS choice moves Re by ~0.5 point. The real error is failing to clean. Raw 6.5% G-sec + full 7.1% ERP = 13.6%, ~2 points too high, with sovereign default charged twice (Step 5 of the country-risk build).

P17. A stable, high-payout market index trades at 1,000. Next year's dividends + buybacks are expected at 40 (a 4% cash yield), growing 4% forever, and the government bond yields 4.0%. Then the index falls to 850 while, the same week, next-year cash-flow expectations are cut to 36. (a) Implied return and ERP before the shock. (b) After the shock. (c) Decompose the 150-point fall into cash-flow news and risk-premium news.

Solution. Gordon inverted, r = CF₁/P + g. (a) r₀ = 40/1,000 + 0.04 = 8.0%, so implied ERP = 8.0 − 4.0 = 4.0%. (b) r = 36/850 + 0.04 = 8.24%, so implied ERP = 8.24 − 4.0 = 4.24%. (c) At the old required return (8.0%), the cut cash flow alone justifies 36/(0.08 − 0.04) = 900. So 100 points, from 1,000 to 900, is cash-flow news. The remaining 50 points, from 900 to 850, is risk-premium news. You can see it in the implied ERP rising from 4.00% to 4.24%. The market both lowered its earnings view and raised its price of risk.

P18 (timed, 15 minutes, no notes). (a) A US sector peer has a levered β of 1.5 at D/E = 0.8 and a 25% tax rate. Unlever it, re-lever for a target firm at D/E = 0.3, and compute the firm's cost of equity with Rf = 4.2% and ERP = 4.3%. (b) A 50/50 portfolio: σ₁ = 20%, σ₂ = 35%, ρ = 0.35. Compute portfolio σ. (c) One sentence: your implied-ERP series jumps from 4.3% to 5.1% in a month while analysts' aggregate earnings forecasts are unchanged: what happened, mechanically?

Solution. (a) Unlever: βU = 1.5 / (1 + 0.75×0.8) = 1.5/1.6 = 0.9375. Re-lever: βL = 0.9375 × (1 + 0.75×0.3) = 0.9375 × 1.225 = 1.148. Then Re = 4.2 + 1.148×4.3 = **9.14%**. (b) σp² = 0.25(400) + 0.25(1,225) + 2(0.5)(0.5)(0.35)(20)(35) = 100 + 306.25 + 122.5 = 528.75 → σp = **23.0%**. (c) Prices fell relative to unchanged expected cash flows, so the IRR that equates them rose. The market repriced risk (≈0.8 point more per unit of equity risk), and earnings didn't move.

P19. Two investors face the same market: Rf = 4.2%, E(Rm) = 8.52%, σm = 16%. Their risk-aversion coefficients are A = 2 and A = 6, and utility is U = E(R) − 0.5 × A × σ² with returns as decimals. (a) Compute each investor's optimal weight in the market portfolio, and the expected return, standard deviation and certainty equivalent of the resulting mix. (b) Show that at each optimum the slope of the investor's indifference curve equals the capital market line's Sharpe ratio. (c) The cautious investor's adviser proposes putting 100% in the market instead. By how much does that reduce their certainty equivalent, and what does the size of the drop tell you?

Solution. (a) w* = [E(Rm) − Rf]/(A σm²). Sharpe = (0.0852 − 0.042)/0.16 = 0.270. For A = 2: w* = 0.0432/(2 × 0.0256) = 0.844, so E(R) = 4.2 + 0.844 × 4.32 = **7.845%**, σ = 0.844 × 16 = **13.50%**, U = 0.07845 − 0.5 × 2 × 0.135² = **6.02%**. For A = 6: w* = 0.0432/(6 × 0.0256) = 0.281, E(R) = 5.415%, σ = 4.50%, U = 4.81%. (b) The indifference curve is E(R) = U + 0.5 A σ², so its slope is A σ. At the first optimum it's 2 × 0.135 = 0.270, and at the second 6 × 0.045 = 0.270. Both equal the line's slope, which is what tangency means. The investor stops buying risk at the point where their own price of risk equals the market's. (c) At 100% market the cautious investor's utility is 0.0852 − 0.5 × 6 × 0.16² = **0.84%**, against 4.81% at the optimum, a drop of 3.97 points. That size is the lesson. For the A = 2 investor the same comparison costs only 6.02 − 5.96 = 0.06 points, because their optimum was already near 100%. The cost of a wrong dose rises steeply with aversion. So "just buy the index" is nearly free advice for one client and expensive advice for another.

P20. You hold a portfolio with an expected return of 9.5% and a standard deviation of 22%. The market portfolio offers 8.52% at a standard deviation of 16%, and Rf is 4.2%. (a) Compute both Sharpe ratios and say which capital allocation line you are standing on. (b) At a 13.5% standard deviation, what does each line deliver? (c) Your portfolio has the higher expected return, so a colleague argues it is the better holding. Answer them in two sentences, and name which of the two lines in the efficient-frontier comparison table each of you is implicitly using.

Solution. (a) Yours: (9.5 − 4.2)/22 = **0.241**. The market's: (8.52 − 4.2)/16 = **0.270**. You are on a capital allocation line that is flatter than the capital market line, which is the definition of an inefficient risky holding. (b) At σ = 13.5%, your line gives 4.2 + 0.241 × 13.5 = **7.45%** and the market's gives 4.2 + 0.270 × 13.5 = **7.85%**. (c) The higher expected return is bought with more than proportionally more risk, so at any risk level you care to name, a mix of bills and the market beats your portfolio. Your colleague is comparing raw returns, which is a security-market-line habit applied where it doesn't belong. The capital market line is the relevant test here, because the question is about a whole portfolio's total risk. One asset's contribution to a diversified portfolio is the security market line's question.

Applied mini-project

The live ERP card: pull it, build it, apply it to two real companies. Budget ~3 hours. This is the live-numbers ritual executed for real, today. The card you produce becomes a standing input to the full cost-of-capital build and every valuation you build afterward. The numbers themselves will decay. What you're really delivering is proof that you can regenerate them in thirty minutes next January.

Step 1: pull the US legs (~20 min).

• US implied ERP: Damodaran Online (pages.stern.nyu.edu/~adamodar), where the home page headlines the current month's implied ERP. Record the value and its month. For the history, Current Data → Implied Equity Risk Premiums for the United States (the histimpl spreadsheet).
• US risk-free rate: FRED (fred.stlouisfed.org), series DGS10. Record the yield and its date.

Step 2: pull the India legs (~30 min). From Damodaran's Country Default Spreads and Risk Premiums page (ctryprem), pull India's Moody's rating. Note S&P's separately, since the agencies currently disagree by a notch, and verify both. Then record the rating-based default spread and the CDS-based spread if it's listed. Add the EM relative-volatility multiplier and his assembled India ERP. Finally, get the 10-year G-sec yield from RBI's DBIE (dbie.rbi.org.in, government-securities tables) or CCIL (ccilindia.com), and cross-check it against a financial portal. Record every value with its date.

Step 3: build the card (~30 min). Compute these, showing each line: CRP = rating-based spread × multiplier, your India ERP = US implied (mature) ERP + CRP and clean INR Rf = G-sec yield − rating-based spread. Reconcile your India ERP against Damodaran's assembled figure. They should agree within rounding. If they don't, diagnose why. The usual culprit is that his country page uses a start-of-year mature ERP while the home page shows the latest monthly. Note the rating-vs-CDS gap and write one sentence on which spread you adopt and why.

Step 4: the NIFTY cross-check, and why it misbehaves (~30 min). From niftyindices.com (Reports → daily index data, or the NIFTY 50 factsheet), pull the NIFTY 50 dividend yield, recently on the order of 1–1.5%. Run P8's Gordon one-liner with the terminal convention g = clean INR Rf: r = dividend yield + g, then implied ERP = r − clean Rf. That collapses to the dividend yield itself, a number visibly far below your ~7% structural build. Diagnose the failure in four to six sentences. You'll need these ingredients. India's payout is low because reinvestment is high (Phase 2: g = reinvestment rate × ROIC), so near-term cash-flow growth must run far above the terminal rate for years, and a one-stage model has no room for that. Dividends alone also miss buybacks and misstate total cash returned. The conclusion to reach is that the one-liner reads stable, high-payout markets adequately (P8) and structurally misreads young, reinvesting ones. So India's premium needs either the full multi-stage inversion of WE3 or the structural build you just did.

Step 5: price three costs of equity (~40 min). Work through them in this order:

• HDFC Bank (domestic lender, λ = 1): pull a published beta from the NSE symbol page or Yahoo Finance's statistics tab. Note the estimation window, since 5-year monthly is common, and flag the number as a regression placeholder that you'll rebuild bottom-up in the WACC build. Compute Re = clean INR Rf + β × India ERP.
• Infosys or TCS (exporter): find revenue by geography in the segment note of the latest annual report. Infosys also files a Form 20-F on SEC EDGAR, which is worth seeing once. First-order λ = the share of revenue earned outside developed markets, counting India + other emerging markets as exposed. For the large IT names, expect something like λ ≈ 0.2–0.35. That range is illustrative, so compute it from the actual split. Then firm ERP = mature ERP + λ × CRP and Re = clean INR Rf + β × firm ERP. What you get is an Indian-listed exporter priced, in rupees, at close to a developed-market premium.
• One US anchor (Microsoft or Coca-Cola): Re = DGS10 + β × US implied ERP. Label all three results with their currency and date, and refuse, in writing, to rank the INR numbers against the USD one (WE4(e)).

Step 6: log and memo (~30 min). Write the ERP card as one dated note in your knowledge system, in your toolkit's format. It should hold each value · source page · pull date and the three-line derivation. Add your rating-vs-CDS choice and the refresh rule. That rule is every January when Damodaran's data update arrives, plus a re-pull after a ±15% index move, a ratings action on India, or a rate shock. Then write a one-page memo (≤400 words) presenting the three costs of equity, the λ reasoning with the filing cited, and the NIFTY-cross-check diagnosis.

Scoring rubric. Be straight with yourself when you score. A criterion is only a 2 if it's complete, correct, dated, and sourced.

Pass: ≥12/16 with no criterion at 0. If you fall short, redo the weak steps before you move on to the full cost-of-capital build, which assumes this card exists and cites it.

Reading & resources

Core (do these).

• Damodaran, Investment Valuation (3rd ed.), ch. 4 (The Basics of Risk) and ch. 7 (Riskless Rates and Risk Premiums). The canonical treatment of this week's whole arc. [Paid] [Intermediate]
• Damodaran, "Equity Risk Premiums (ERP): Determinants, Estimation and Implications": his book-length paper, re-issued annually with fresh numbers and free on SSRN. Read the implied-ERP and country-risk sections now, skim the rest, re-download each year. [Free] [Intermediate]
• Damodaran's NYU class sessions on risk, the risk-free rate, and the ERP, free on YouTube (the Valuation course's early sessions; use the current year's playlist so the numbers are live). Watching him iterate an implied ERP is the video version of WE3. [Free] [Beginner→Intermediate]
• Brealey, Myers & Allen, Principles of Corporate Finance (13th ed.), ch. 7 (Introduction to Risk and Return) and ch. 8 (Portfolio Theory and the CAPM). The standard textbook path over the same ground with more practice. [Paid] [Beginner→Intermediate]
• Zerodha Varsity, Risk Management module: the portfolio-variance and beta chapters. They give you the same arithmetic in an Indian retail idiom, good for a second pass. [Free] [Beginner]

Going deeper.

• Fama & French (2004), "The Capital Asset Pricing Model: Theory and Evidence," Journal of Economic Perspectives: the definitive, readable statement of the empirical case against CAPM, by the prosecution. Free online. [Free] [Intermediate]
• Frazzini & Pedersen (2014), "Betting Against Beta": the flat-SML anomaly monetised, free at AQR's research library. [Free] [Advanced]
• Dimson, Marsh & Staunton, UBS Global Investment Returns Yearbook (annual; the summary edition is a free PDF on the UBS site), the survivorship corrective: 20+ markets since 1900. [Free] [Intermediate]
• McKinsey, Valuation (8th ed.): the cost-of-capital chapter, which is the practitioner counterweight on arithmetic-vs-geometric and premium choice. [Paid] [Intermediate]
• Fernandez, "Survey: Market Risk Premium and Risk-Free Rate", annual and free on SSRN. It shows what thousands of practitioners say they use, country by country. Read as sociology. [Free] [Beginner]
• MIT OCW 15.401 Finance Theory I: the portfolio-theory and CAPM lectures, if you want the frontier built with full rigour once. [Free] [Intermediate→Advanced]
• Mehra & Prescott (1985), "The Equity Premium: A Puzzle," Journal of Monetary Economics: the original statement that the historical premium is far larger than standard risk-aversion theory can justify (CAPM's fifth empirical failure, above). Short, seminal, and widely mirrored free online. [Free] [Advanced]
• Antti Ilmanen, Expected Returns (2011): the practitioner-scholar's full tour of every major risk premium, what drives each, and how they move through time. It's the natural deep well once your ERP machinery feels solid. [Paid] [Advanced]

Live data. This is the January list, so bookmark all of these.

• Damodaran Online (pages.stern.nyu.edu/~adamodar): home-page implied ERP (monthly), histimpl (implied-ERP history), ctryprem (default spreads, multiplier, country ERPs), histretSP (US historical returns 1928–present). [Free]
• Musings on Markets, Damodaran's blog. Its January "Data Update" series is a free annual masterclass on these very numbers. [Free]
• FRED (fred.stlouisfed.org): series DGS10, US 10-year Treasury. [Free]
• RBI DBIE (dbie.rbi.org.in) and CCIL (ccilindia.com): 10-year G-sec yields, primary sources. [Free]
• niftyindices.com, for NIFTY factsheets and daily reports: index P/E, P/B, dividend yield. [Free]
• SEC EDGAR (sec.gov/edgar) and company annual reports: revenue-by-geography notes for λ. [Free]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. That is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "Expected return and dispersion, the two-asset correlation result, systematic versus diversifiable risk, the efficient frontier, CAPM and beta, and both routes to the equity risk premium",
  "use": "The risk of this module is not arithmetic error, it is smooth confidence about a contested model — so use a tool where it argues and never where it asserts. (1) Ask it to state the empirical case against CAPM and the practitioner's defence, each in its strongest form with the specific papers, and check every citation. §4.7 already gives you both halves; the exercise is to see whether the tool reproduces them honestly or flattens them into a shrug, which teaches you how much to trust it on anything contested. (2) Ask it to enumerate the estimation choices behind any published beta — index, frequency, window, adjustment, currency, dividend treatment — before you use a number off a screener; a beta is not a fact, it is the output of six decisions. (3) For the implied ERP, ask it to lay out the cash-flow build (dividends plus buybacks, growth stage, terminal convention) and then compute the iteration yourself. (4) Ask it to construct the peer set for a bottom-up beta and to find each peer's debt, equity and tax rate in the filings — assembling those inputs is document work, and unlevering them is not.",
  "tools": ["Chat assistants — Claude, ChatGPT for the CAPM debate, estimation-choice enumeration and the ERP build's structure", "Document/PDF AI — for peer capital structures, tax rates and buyback aggregates from filings", "Coding assistants — for the regression, the standard errors and the simulation below"],
  "prompt": "State the empirical case against the CAPM as its strongest critics make it, with the specific papers and their samples, and then the strongest defence of using it anyway for an ex-ante required return. Keep both to equal length and do not conclude. Separately, list every estimation choice that goes into a published equity beta and, for each, the direction and rough magnitude of its effect. Do not give me a beta for any company.",
  "verify": "Two rules. First, no beta enters a valuation without its estimation choices stated and its uncertainty attached — the Quantitative lens shows the 95% interval on a five-year monthly beta is about ±0.34, which is not a rounding issue. Second, every ERP is logged with its date, its method and its source: a historical premium and an implied premium are different quantities that answer different questions, and this program's default is the current implied premium refreshed monthly-to-quarterly. Recompute any implied ERP from the index level and cash-flow aggregates yourself; the number is an IRR and its inputs are contestable, so an unshown ERP is an unusable one. A model asked for 'the beta of X' will produce a plausible number with no window, no index and no interval (Guardrail 1-5).",
  "diy": "The gate is unaided: you compute covariance, beta and the CAPM required return from a returns table by hand, build the historical and implied premiums, and state CAPM's failures without notes. Nothing in this callout is admissible in that room."
}

Modern Data Analysis

By hand first. You computed a beta from a six-quarter table (deviations, products, covariance, variance) and iterated the implied ERP to a few basis points with a calculator. Don't skip that. It's what makes the standard error below mean something, so it doesn't just sound like a caveat. The code adds scale and, more importantly, error bars, the one thing the hand version can't give you.

Today's workflow. A modern beta build is three lines of regression and thirty lines of care. Pull total-return series for the stock and a stated index, align the calendar, choose and record the frequency and window, and run the ordinary least squares regression. Then report the standard error alongside the slope. It comes free from the same regression, and it changes what the number can be used for. Once that's a function, the estimation choices become a grid you can actually inspect. The same company against NIFTY 50 and NIFTY 500, on two-year weekly and five-year monthly data, is four betas, and seeing their spread once teaches more than any warning. The bottom-up build then becomes a loop: regress each peer, unlever with each peer's own debt-to-equity and tax rate, average the asset betas, relever at the subject's structure. You can measure why it works, too, because averaging k independent estimates divides the standard error by √k. For the implied ERP, write the index cash-flow build as a function and solve for the internal rate of return with a root-finder. Then run it monthly and keep the series. The implied method's best gift is decomposing a selloff into cash-flow news versus risk-premium news, and that needs history you kept.

Tools & sources (IN + US). Use pandas for the return series, numpy/statsmodels for the regression and its standard errors, and scipy.optimize.brentq for the ERP solve. Indian price history for NSE and BSE comes from the exchanges' own bhavcopy archives, or via the community-maintained jugaad-data package. Check that it still works. The older nsepy has had no release since March 2020, and reports of it breaking against NSE's rebuilt site are common. Add the NIFTY 50 and NIFTY 500 total-return series for the index choice, G-sec yields from FBIL and RBI's DBIE for a rupee risk-free rate, MOSPI CPI for the real-nominal work, and Damodaran's country-risk-premium and implied-ERP pages at pages.stern.nyu.edu, re-pulled and dated. For the US, use price history via yfinance or pandas-datareader, FRED (DGS10, DGS30) for the risk-free rate, Damodaran's monthly implied-ERP series, and the Ken French data library for the size, value, profitability, investment and momentum factor returns behind CAPM's criticisms.

# Beta's standard error — R3: recomputed in-session; synthetic returns, stated parameters
import numpy as np
sig_s, sig_m, beta, n = 0.30, 0.18, 1.0, 60        # annual vols, true beta, 5y of monthly obs
sig_e = sig_s * np.sqrt(1 - (beta * sig_m / sig_s) ** 2)          # residual vol = 24.0%
se = (sig_e / np.sqrt(12)) / ((sig_m / np.sqrt(12)) * np.sqrt(n))  # 0.172 analytically
print(round(se, 3), round(1 - 1.96 * se, 2), round(1 + 1.96 * se, 2))     # 0.172 0.66 1.34
rng = np.random.default_rng(302)                                   # empirical confirmation
rm = rng.normal(0, sig_m / np.sqrt(12), (20_000, n))
rs = beta * rm + rng.normal(0, sig_e / np.sqrt(12), (20_000, n))
c = lambda x: x - x.mean(1, keepdims=True)
bhat = (c(rm) * c(rs)).sum(1) / (c(rm) ** 2).sum(1)
print(round(bhat.std(), 3), [round(np.percentile(bhat, q), 2) for q in (2.5, 97.5)])  # 0.177
print([round(bhat.std() / np.sqrt(k), 3) for k in (1, 3, 5, 8, 15)])   # bottom-up: SE / sqrt(k)

Verify. Reproduce this week's worked one-liners before trusting any pipeline: a 0.55-beta staples business at a 4.2% risk-free rate and a 4.3% premium prices at 6.6%, a 1.7-beta chipmaker at 11.5%, a 1.0-beta rupee lender at 4.5% + 7.1% = 11.6%, and the Gordon warm-up gives 3.5% + 4% = 7.5%. Then verify the inputs, because that's where errors creep in here. The risk-free rate must be currency-matched to the cash flows, the index must be a total-return series or the beta inherits a dividend bias, the return frequency and window must be recorded, and an implied ERP must carry its date and its terminal convention. Treat any beta that arrives without a standard error as a number somebody chose instead of measured. On the allocation line, check a Sharpe ratio of 0.270 from a 4.2% risk-free rate, an 8.52% market return and a 16% market standard deviation. Check optimal market weights of 84.4% at a risk aversion of 2 and 28.1% at 6, giving certainty equivalents of 6.02% and 4.81%, and indifference-curve slopes of 0.270 at both optima. A 9.5%-return, 22%-risk portfolio's own line carries a Sharpe ratio of 0.241 and reaches 7.45% where the market line reaches 7.85%.

Quantitative lens

CAPM's scorecard lists beta's instability as the third empirical failure, and the full cost-of-capital build promises a bottom-up repair. You can compute both the size of the problem and the size of the repair, and doing so turns two pieces of received practice into arithmetic.

How wide is a regression beta? The standard error of an ordinary least squares slope is the residual volatility divided by the market's volatility and the square root of the sample size. Take a stock with 30% annual volatility against an index with 18%, with a true beta set at 1.0, which implies an R² of 0.36 and a residual volatility of 24%, both entirely typical. From five years of monthly returns the standard error of the estimated beta is 0.172, so the 95% confidence interval is roughly [0.66, 1.34], a width of 0.67. Simulating twenty thousand such regressions confirms it: mean 1.000, standard deviation 0.177, 2.5th-to-97.5th percentile [0.65, 1.34]. From only two years of monthly data the interval widens to [0.47, 1.53]. Five years of weekly data narrows it to [0.84, 1.16]. That's better, but it picks up more microstructure noise and non-synchronous trading in less liquid names, so choosing a frequency is a trade-off, and there's no free lunch in it.

What does that width cost a valuation? Take this week's rupee build: a 4.5% risk-free rate, a 7.1% India premium, and a company whose cash flows grow at 5% in perpetuity. At beta 1.0 the cost of equity is 11.60% and the perpetuity value is proportional to 1/(11.60% − 5%). At the interval's low end, beta 0.66, the cost of equity is 9.19% and the value is 1.58× the base. At the high end, beta 1.34, the cost of equity is 14.01% and the value is 0.73×. The 95% interval on a single five-year monthly beta therefore spans a 2.2-fold range of intrinsic value, from one input, before you have made a single judgment about the business. So does a half-point of discount-rate refinement matter? Usually not. But the beta interval isn't a half-point.

And why does the bottom-up beta work? Because averaging independent estimates divides the standard error by the square root of their number. Averaging three peer betas takes the standard error from 0.177 to 0.102. Five peers take it to 0.079, and eight peers to 0.063, a 95% interval width of 0.25. Fifteen peers get it to 0.046. Don't mistake the bottom-up procedure (regress the peers, unlever each with its own capital structure, average the asset betas, relever at the subject's) for a stylistic preference for cleanliness. It's a variance-reduction technique with a factor-of-√k payoff, and it's the highest-return statistical move available anywhere in the cost-of-capital build.

Now the limits, and the last one matters most. The √k gain assumes the peers' residuals are independent, and they aren't. Firms in one sector share shocks, so the effective number of independent observations is smaller than the peer count, and the real gain is less than √k. It's still large and still the right move. A wider peer set buys statistical precision and spends economic comparability. You refused that same trade-off for peer tables when you studied benchmarking, and the answer here is the same. A peer has to qualify on business grounds first. The confidence interval above assumes a stable true beta over the window, whereas leverage and business mix change. So part of what a long window buys in precision it gives back in staleness, and that part is a bias no standard error will reveal. And the interval is a statement about estimation noise only. It says nothing about whether beta is the right measure of risk at all. That's the deeper argument from CAPM's scorecard, and no amount of data solves it.

Do it in code: run the regression with statsmodels so the standard error, R² and residual diagnostics arrive alongside the slope. Compute the beta on the full grid of index, frequency and window choices and report the range. Then build the bottom-up beta as an average of unlevered peer betas and carry its standard error into the cost-of-equity range you show. See G1/QM1.02 and G2/DA1.05.

Where this goes next: galaxy cross-links

• G1 · QM1.02 Hypothesis Testing & Regression. What an OLS slope's standard error is, why it shrinks with √n, and how to read the diagnostics that tell you a beta regression is misspecified rather than merely noisy.
• G1 · AA1.01 Mean-Variance Optimization & the Efficient Frontier. The efficient-frontier picture computed instead of drawn, including how violently an optimiser reacts to estimation error in the very inputs you estimate here.
• G2 · DA1.05 Statistics in Code. Covariance matrices, bootstrapped confidence intervals and the estimation-choice grid, done properly instead of by eye.
• G1 · QM1.01 Probability, Distributions & Estimation. Expected values, variance and the sampling distribution of an estimate, which is what turns "beta is unstable" into a number you can put beside the beta.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "Why the market won't pay you for risks you could have avoided, and how to read its price for the ones you can't." You have to land three ideas without formulas. (1) Why holding many stocks makes each company's private disasters stop mattering while recessions still do. Use a concrete image, such as many stalls in one bazaar sharing one monsoon. (2) Why competition among diversified buyers strips any extra reward off avoidable risk. (3) How, if you know what the market expects companies to pay out and you can see today's price, you can work backwards to the return the market is demanding, and why that number jumping upward is the sound of fear. Test it on someone, then rewrite whatever you had to explain twice.

Journal prompt. Three entries, dated. (1) Your policy, in writing. Decide which ERP your valuations will use: current implied (Camp Damodaran) or a normalised ~5% (Camp Normalized). Pick a default-spread route for India, rating or CDS, and a refresh cadence. Commit, because every memo from the cost-of-capital build onward cites this policy. (2) The two risk languages. In one paragraph each, describe a situation where beta is the right risk lens and one where Buffett's permanent-loss lens is. Then say plainly which you reach for by temperament and what that blind spot could cost you. (3) Your own diversification. Count where your actual wealth is (salary, home, any holdings, human capital included). Estimate how much of your risk is one country's systematic factor. No action is required yet. Phase 9 acts on it, and today you only have to see it.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.