Equity risk premium

Everything now funnels into one number. Rf is observable. Beta is estimable. The ERP, the price of one unit of average equity risk, is neither. It has to be extracted, and how you extract it is one of the few really contested choices in valuation. Small differences matter enormously. For a β = 1 company with Rf = 4%, moving the ERP from 4% to 6% moves the cost of equity from 8% to 10%. Through a Gordon terminal value at g = 4%, that moves value per rupee of next-year cash flow from 25× to 16.7×. A third of the valuation, from one input.

Three families of estimate exist. You can dismiss the first quickly for our purposes. Surveys simply ask people, as in Fernandez's annual global survey of academics and practitioners or Graham-Harvey's CFO survey. They're useful as sociology, but survey answers lag prices, cluster around round numbers, and measure sentiment as much as required return. Know they exist, and don't build on them. The real contest is historical versus implied.

How is the historical equity risk premium calculated?

Historical estimates average the past. Take a long series of realised stock returns, subtract the risk-free return, average the difference. Three choices steer the answer without announcing themselves:

1. The window. 1928–present (Damodaran's default US table), post-1974, the last 20 years? Longer windows shrink statistical noise but average across dead regimes (gold standards, wars, 15% inflation). Short windows are noise. There's no "right" answer, only knowing which one you picked.
2. The risk-free proxy. Bills or bonds? Premia over T-bills run higher than over 10-year bonds. Use the bond. Your valuations discount decade-scale cash flows against a decade-scale Rf, and the premium has to be built against the same instrument you'll add it to. That gives you a consistency rule: never take a premium built over bills and stack it on a bond yield.
3. Arithmetic or geometric mean. This is the bootcamp's distinction, now with money attached. The arithmetic mean of annual excess returns answers "expected premium in one independent year." The geometric answers "premium actually compounded over the whole period." Arithmetic exceeds geometric by ≈ σ²/2, which for equities (σ ≈ 20%) is about 2 points. On US data for 1928–2024, stocks over 10-year T-bonds come to ≈ 6.5–7% arithmetic and ≈ 5% geometric (Damodaran's Historical Returns on Stocks, Bonds and Bills page; as of mid-2026). Same history, two valid numbers, 2 points apart. That gap matches the σ²/2 drag. Which should you use? For discounting long horizons, most appraisers, Damodaran included, lean geometric over long windows. McKinsey's authors argue for arithmetic adjusted downward for estimation issues and land near 5% too. In practice the camps converge on a mature-market historical ERP of ≈ 4.5–5.5%. State your choice, then apply it everywhere.

What are the traps in the historical equity risk premium?

Trap 1: the precision illusion. Annual equity returns have σ ≈ 20%. The standard error of a mean is σ/√N, so even with 97 years of data it's 20/√97 ≈ 2.0 points. Your "6.5%" carries a 95% confidence band of roughly ±4 points. The true long-run premium could plausibly be 2.5% or 10.5%. A century of data can't pin this number to the decimal your spreadsheet displays. Anyone quoting a historical ERP to two decimals is reporting typography.

Trap 2: survivorship. Why does everyone quote US history? Because the US spent the 20th century becoming the world's dominant market, a fact known only afterward. Averaging the winner's past overstates what an investor could have expected in advance. The corrective evidence is the Dimson-Marsh-Staunton database, published annually in the UBS Global Investment Returns Yearbook (formerly Credit Suisse, with a free summary edition on the UBS site). Across 20+ markets since 1900, world equity premia run roughly a point or more below the US figure, and that's among markets that survived the century at all. Investors in St. Petersburg (1917) and Shanghai (1949) held claims that went to zero and vanished from every "long-run average" you'll ever be quoted.

So the US historical number lies at the optimistic edge of the distribution, away from the centre.

Trap 3: India's history cannot carry the load. India's Sensex series starts in 1979 and is partly reconstructed. Then the economy changed species in 1991, inflation regimes shifted twice, and the indices are survivor-heavy through decades of churn. A "historical Indian ERP" computed from this is a short, noisy sample drawn from three different economies wearing one name. Nobody can fix that defect, and it's why the profession estimates India's premium structurally: start from a mature market's premium and add a measured country surcharge, as the country-risk build does. File the pattern. For any emerging market, prefer structure over its own thin history.

The equity risk premium II: the implied route (Damodaran's method)

The historical method reads the rear-view mirror. The implied method reads the windscreen. Today's prices already contain the market's required return, so solve for it.

Its logic is the time-value lessons run backwards. A DCF takes cash flows and a rate and outputs a price. But the market shows you the price every day, and cash flows can be estimated. That leaves one unknown, the discount rate. The r that makes expected cash flows discount to today's index level is the return the market is collectively pricing itself to deliver, the market's IRR. Subtract the risk-free rate, and what's left is the premium investors are demanding right now:

Index level today = PV of expected future cash flows to shareholders, discounted at r
solve for r  →  implied ERP = r − Rf

How do you calculate the implied equity risk premium?

Warm up with the Gordon one-liner, which is the growing perpetuity inverted. Say a market trades at level P and will pay cash CF₁ to shareholders next year, growing forever at g. Then P = CF₁/(r−g) rearranges to r = CF₁/P + g. Expected return = cash yield plus growth. A stable market at 1,000 paying 35 next year and growing 4% forever prices in 3.5% + 4% = 7.5%. Against a 4% bond, that's an implied ERP of 3.5%. You can now read a market's temperature on the back of an envelope.

Damodaran's full implementation adds realism in three places, and Worked Example 3 executes every step with an iteration table:

1. Cash flows = dividends + buybacks. In the US, buybacks have long exceeded dividends as the route cash takes to reach shareholders. Counting dividends alone understates the cash yield and therefore the implied premium. It's a classic error, and it once made US stocks look absurdly expensive to dividend-only models. Read both off the index's aggregates. In India, buybacks are a thinner (and tax-shifting) channel, so dividends carry more of the yield, and the overall payout is lower because reinvestment is higher.
2. A growth stage, then a terminal state. Grow the cash flows at analysts' aggregate earnings-growth expectations for ~5 years, then set terminal growth equal to the risk-free rate. That rule clocks long-run growth to the economy, because a market's cash flows can't outgrow the economy forever. Using Rf as the cap is Damodaran's convention, and the terminal value page derives the g ≤ economy logic fully.
3. Solve by iteration. There's no closed form, so you try a rate, compute PV, compare it with the index, adjust. Two trials and a straight-line interpolation get you within a few basis points. You'll do this by hand in WE3, and it takes ten minutes.

Strengths and weaknesses of the implied equity risk premium

Why go to that trouble? Because the implied route has four strengths:

• It's forward-looking and always current. No 1928 anywhere. It updates with every price tick and every revision to expected cash flows.
• It has no survivorship problem. Today's price is today's price, and nobody's dead market contaminates it.
• It moves the right way, when it matters most. Because price is in the denominator, a crash raises the implied ERP, so the market starts offering more per unit of risk just when fear peaks. Damodaran's monthly series shows ≈ 2% at the start of 2000, the dot-com peak. That was his lowest reading, with equities priced to deliver almost nothing over bonds. Compare ≈ 6.4% entering 2009, near the GFC trough, the highest to that date. His site's front page carries the current reading, and these values come from his published series (check against the Implied ERP page). A historical average fed the same 2008 would have fallen, with one more bad year in the mean. That implies, absurdly, that equity risk got cheaper to bear after the crash. The implied premium is the only one of the three that behaves like a price.
• It makes your valuations internally consistent. Discounting at a premium implied by today's market means your DCF disagrees with the market only where you intend it to, in the cash flows of your specific company. A stale macro input can't slip a disagreement in by accident.

It has real weaknesses too, and the first is that the output inherits every input. Optimistic analyst growth forecasts inflate it. Buyback run-rates may not last. Are peak-cycle buybacks a permanent yield? Its terminal convention is a modelling choice. And it prices the average stock, so single names still need beta. Deepest of all, using it as the premium assumes, without saying so, that the market's overall level is fair. If you believe the whole market is in a bubble, the implied ERP is the bubble's own discount rate. This cuts both ways and divides good practitioners. Camp Damodaran uses the current implied ERP, stays consistent with the market you actually transact in, and expresses company views through cash flows. Camp Normalized uses a long-run ~5% so a manic market can't lower your hurdle. In exchange, your "undervalued" flags partly just re-express your macro view. Both are defensible. Pick one policy, write it into your valuation workflow, state it in every memo. Our default, following Damodaran, is current implied, refreshed monthly-to-quarterly, logged with its date.

Cash-flow news vs risk-premium news

There's one more use for this method. Run it on the changes. When the index falls 20% while expected cash flows fall 5%, the implied ERP jumps, because the market is repricing risk as well as earnings. Decomposing a selloff into "cash-flow news vs risk-premium news" is an institutional-grade habit, and after WE3 you'll own the machinery for it.