Second of four pieces examining the discount-rate stack.
An analyst building a discount rate has to put a single equity risk premium on the page. The three places that figure usually comes from are the long-run history of realised returns, the premium implied by today's market price, and the average that other practitioners say they use. The reflex is to treat these as three readings of one quantity, pick whichever is nearest to hand, and move on.
They are not three readings of one quantity. The historical premium answers what equities paid above the risk-free asset over a chosen window and market. The implied premium answers what discount rate reconciles today's index level with an assumed path of cash flows. The survey premium answers what other people report using. A figure that does not say which of those questions it answered has not measured the premium; it has hidden an assumption inside a number.
That is the spine of this piece. Three numbers, three questions, one figure on the page: a discount rate that names which question it answered survives the committee, and one that does not survives only until the first sceptical analyst asks. The argument that follows is about method, not about which value to adopt. There is no per cent figure here that a reader should lift into a live model.
Three numbers, or three questions?
The first move is to stop asking which premium is correct and start asking which question the model needs answered. The three families are not competing estimates of the same parameter. They are answers to different questions that happen to share a unit.
This matters because the silent input does the real work in each case. The number on the page is the visible output; the assumption that produced it sits behind the figure and rarely travels with it into the committee pack. Naming the family is the first half of disclosing that assumption. The second half, the parameter choices within the family, is where the figure actually moves.
The equity premium is also not a fixed property of markets. Mehra and Prescott showed that the realised premium in US data is far larger than standard consumption-based asset pricing predicts for plausible levels of risk aversion, a result that has resisted tidy resolution for four decades [6]. The practical lesson is narrow but firm: the historical number is an observation about one realised path, not a constant of nature an analyst can read off and trust to hold.
What did equities actually pay?
The historical, or realised, equity risk premium is the mean of equity return minus risk-free return measured over a long window. The reference series for this family is the Dimson, Marsh and Staunton data assembled in the UBS Global Investment Returns Yearbook, which tracks long-run premia across more than 20 markets since 1900 [4]. The institutional summary of the latest edition reports that equities have outpaced both bonds and bills over the 125-year record by a wide margin [5].
Four choices sit inside that single realised number, and each one moves it. The first is the market: a premium drawn from one surviving national market reads differently from a global premium. The second is the window: the start and end dates of the sample change the mean, and a window chosen to include or exclude a particular regime is an assumption wearing the costume of a fact. The third is the base instrument. The premium measured against long government bonds is not the premium measured against short bills, because the bond carries a term premium and a different reinvestment profile; quoting an equity risk premium without naming whether it is an equity-bond or an equity-bill premium leaves the number undefined [4].
The fourth choice is the averaging convention. The arithmetic mean is the average of single-period premia and is the larger figure; the geometric mean is the compound annualised premium and describes realised compounding [2]. The gap between them widens with return volatility. The arithmetic mean is the one consistent with discounting a single expected future cash flow, which is what a discount rate does, so the choice is not cosmetic.
One discipline guards this whole family: survivorship. A premium estimated from a market selected because it survived overstates the forward premium, because the markets that failed or were interrupted are absent from the average. The DMS dataset was built partly to correct exactly this bias in earlier US-only, survivor-only series [4]. The realised premium is a fact about a chosen window in a chosen market; it is not, on its own, a forecast.
What discount rate does today's price imply?
The implied, or forward, premium takes a different route. Rather than averaging the past, it solves for the discount rate that sets the current index level equal to the present value of expected future cash flows, then subtracts the risk-free rate. In its single-stage form this is the Gordon growth model rearranged for the cost of equity; Damodaran's published series uses a multi-stage free-cash-flow-to-equity variant, with cash flow defined as dividends plus buybacks, and updates the implied number as prices and inputs move [1].
The appeal of the implied premium is that it is conditional on today's price, not on a window an analyst chose. It is internally consistent: the figure is whatever makes the present-value identity hold. That consistency is also its boundary. The method is only as defensible as two inputs: the payout assumption (whether buybacks are counted alongside dividends) and the expected growth rate fed into the cash-flow path [2]. Change the growth input and the implied premium moves, because the same price now reconciles with a different discount rate.
This is the section where a house forecast can smuggle itself in unnoticed. The growth assumption is an input the analyst supplies, and it is tempting to treat the resulting implied premium as objective because the arithmetic is mechanical. It is not objective; it is conditional. The honest disclosure records the growth and payout assumptions alongside the figure, so a reader can see the premium is the output of a stated view rather than a reading taken from the market. The implied premium describes the discount rate the market is currently applying given those assumptions. It is not a prediction of returns, and presenting it as one overstates what the method delivers.
What does everyone else say they use?
The survey premium is the central tendency and dispersion of the premia that practitioners, companies and academics report using, gathered by questionnaire. The standing reference is the Fernandez survey, which collects reported market risk premia and risk-free rates across dozens of countries each year [3]. Read carefully, it answers a question the other two families do not: what is the consensus practice in a given market.
The instructive feature of the survey family is the dispersion. Within a single market, the reported figures span a range, not a point, which means a survey median tells an analyst what is conventional rather than what is correct [3]. The composition of the respondent pool shapes the result, so a survey reflects who answered as much as what the premium is.
The survey premium has a legitimate use and an illegitimate one. The legitimate use is as a reasonableness check: a chosen premium that sits far outside the reported range invites a question, and the analyst should be ready with an answer. The illegitimate use is to treat the median as the premium, importing a consensus number with none of the disclosure the other families demand. Consensus is not a measurement.
Why do the three disagree, and which disagreement matters?
The three families will rarely produce the same figure, and the gaps between them are informative rather than embarrassing. A wide gap between the realised and implied premia is not evidence that one is wrong; it is a signal about the assumptions each one rests on, the window on one side and the growth path on the other.
The disagreement that matters is the one inside the analyst's own choice. Picking the historical premium commits the model to a window, a market, a base instrument and an averaging convention. Picking the implied premium commits it to a growth and payout assumption. Picking the survey premium commits it to a respondent pool and a tolerance for dispersion. The figure the committee sees is identical in form across all three; the commitment behind it is not. The right response to the disagreement is not to average the three into a blended number that obscures every assumption at once. It is to choose the family that matches the decision and to state what the choice commits the model to.
What should a discount rate disclose?
The output of an equity-risk-premium decision is not a figure. It is a figure plus a disclosure. Six choices make a premium defensible, and each one should be visible in the model documentation: which family (historical, implied or survey), which market, which window, which base instrument (bond or bill), which averaging convention (arithmetic or geometric), and which growth assumption if the implied route was taken. An analyst who can defend those six choices has a discount rate that holds up under questioning. An analyst quoting a premium from memory has one that does not.
This is a modelling-documentation problem before it is a finance problem. The discipline lives in the assumption register and the source map, not in the cell that holds the number, which is why it belongs to construction rather than to opinion. The firm's financial modelling service builds the disclosure into the model, so the six choices are recorded where they are made and the question is answered before a reviewer asks it. That posture, naming the question before quoting the number, is the same independent challenge to prevailing assumptions that runs through GIVE Analytics' engagements in financial services, where institutional investors expect a discount rate to survive their own scrutiny.
Where this sits in the discount-rate stack
The equity risk premium is one layer of a discount rate, not the whole of it. Beneath it sits the country-risk premium, the subject of the country-risk premium piece earlier in this series, which carries the same instruction to name the construction method before quoting the figure. Above the equity premium, the next layer adds the adjustments for size and illiquidity that a private-asset discount rate needs; that layer is the subject of a later piece in this series.
The premium does not stay in isolation either. Once chosen and disclosed, it feeds the cost of equity and then the weighted average cost of capital, which is where the equity premium feeds a cross-border WACC becomes the operative question. The disclosure travels with it: an investment committee reading a discount rate is reading a chain of assumptions, and the equity risk premium is one link in that chain. The committee framing, the way an assumption is read and challenged at the decision point, is the subject of how investment committees read returns. The figure is the easy part. The question it answered is the part worth recording.
Notes
1. Aswath Damodaran, Historical Implied Equity Risk Premiums (US data table, 1960 to 2025), NYU Stern, updated January 2026. https://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile/histimpl.html
2. Aswath Damodaran, Equity Risk Premiums (ERP): Determinants, Estimation and Implications, NYU Stern, 2022 edition. https://pages.stern.nyu.edu/~adamodar/pdfiles/papers/ERP2022Formatted.pdf
3. Pablo Fernandez, Diego Garcia de la Garza and Lucia Fernandez Acin, Survey: Market Risk Premium and Risk-Free Rate used for 96 countries in 2024, IESE Business School, 2024. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4754347
4. Elroy Dimson, Paul Marsh and Mike Staunton, UBS Global Investment Returns Yearbook 2024 (summary edition), UBS / London Business School / Cambridge Judge, 2024. https://storage.googleapis.com/ni_library_storage/Research/ubs-global-investment-returns-yearbook-2024-summary-edition.pdf
5. Cambridge Judge Business School, Report: stocks have far outperformed over the past 125 years (Yearbook 2025 write-up), 2025. https://www.jbs.cam.ac.uk/2025/report-stocks-have-far-outperformed-over-the-past-125-years/
6. Rajnish Mehra and Edward C. Prescott, The Equity Premium: Why is it a Puzzle?, NBER Working Paper 9512, 2003. https://www.nber.org/system/files/working_papers/w9512/w9512.pdf