Market-Level Valuation, the Index Default and Your Own Policy

Phase 11 elective · The market-valuation and policy lab · ~16 focused hours · Mastery gate ≥ 85%

A ratio of 32.00 and a ratio of 23.21 describe the same index on the same afternoon. The first divides today's level by the average of ten years of earnings restated into today's rupees. The second divides it by last year's earnings. Neither is a mistake, and the 38 percent gap between them is the whole content of the disagreement: it says that the last three years of earnings ran far above the decade's own average, and it asks you whether you believe that is a new level or a good patch.

Most investors never compute either figure. They carry an impression instead. Shares feel expensive, or they feel cheap, and the impression tracks how recently the index went up rather than anything about price against earning power. An impression cannot be shown to be wrong, because it never committed to anything. A number with an error band around it can be shown to be wrong, and being capable of wrongness is the only property that makes a valuation useful.

So the first half of the work here is to put numbers on a market. There are five, and each can be built from public data in an afternoon. A cyclically adjusted earnings multiple. A market capitalisation set against national output. An index multiple read against its own past. An earnings yield placed beside a real bond yield rather than a nominal one. And an implied equity risk premium solved out of today's price.

Each gauge gets a current reading, a percentile inside its own history, and a written statement of what it licenses you to do.

The second half is the part that decides your outcome, and it is far less interesting than the first. The honest finding about market valuation is that the relationship between a starting multiple and the next ten years of real return is real, weak and useless for timing. On the fitted line used throughout the teaching below, a ten-point difference in the cyclically adjusted multiple moves the ten-year forecast by 3.20 percentage points a year, while one standard deviation of the residual is 3.90 points. A single unexplained residual is worth more than a 12.19-point swing in valuation. You are holding a prior, not a signal.

What survives that finding is a policy. Put the capital that has no named edge into a broad index at the lowest cost you can find. Hold an allocation band you wrote down before you needed it, with a floor as well as a ceiling. Let the gauge move you a little inside the band rather than in and out of the market. Decide in advance what number counts as enough, because a portfolio with no finish line converts every gain into a new reason to take risk.

The arithmetic ranks those two halves for you, and the ranking is not close. Tilting your equity weight by a full 15 percentage points on a correct valuation call is worth about 0.60 percentage points a year against a 4 percent equity risk premium. Indexing a core that would otherwise pay 150 basis points, against an index vehicle at 10, is worth 1.40 percentage points a year. The certain saving beats the uncertain call by a factor of 2.33, and the saving pays whether the call was right or not.

One warning about the data before you start. Every index series printed below is illustrative and internally consistent: the numbers behave the way the published series behave, and they were chosen so the arithmetic is reproducible from the tables on the page. They are not the market's actual readings on any date. The primary sources, and the protocol for rebuilding each gauge from live data with a check date attached, are given where each gauge is first constructed. A market gauge computed from a number nobody can trace is worse than no gauge at all, because it feels like evidence.

Learning objectives

By the end you can:

1. Construct a cyclically adjusted price-to-earnings ratio for either market from ten years of index earnings and a stated inflation series, and state the reading against its own history as a percentile rather than as an adjective.
2. Quantify the three standard objections to that ratio: the inflation adjustment being skipped, a single write-off year distorting the ten-year mean, and net buybacks inflating a per-share series relative to aggregate earnings.
3. Read a valuation-to-return relationship honestly, converting a fitted slope, a coefficient of determination and a residual standard deviation into a prediction band, and stating the effective sample size once overlapping windows are accounted for.
4. Compute a market capitalisation to gross domestic product ratio for both markets and repair its denominator, using the listed sector's share of value added in India and the foreign-revenue share of the index in the United States.
5. Separate a genuine multiple re-rating from a composition change, computing the index multiple as a market-value-weighted harmonic mean and showing the shift when sector weights move but no sector's own multiple does.
6. Demonstrate arithmetically that the Fed model is an inflation-illusion error, computing the earnings-yield spread against a nominal yield and against an exact real yield in a high-inflation and a low-inflation year, and showing that the ranking between the two years reverses.
7. Solve for the implied equity risk premium at index level using a two-stage cash-flow model with dividends plus net buybacks, and report its sensitivity to the terminal-growth convention and to the stage-one growth assumption.
8. Assemble a five-gauge market-temperature dashboard with each gauge's reading, its own-history percentile and a written licence statement, and resolve a disagreement between an absolute gauge and a relative one.
9. Operate a defensive allocation band with a floor and a ceiling, running a stated multi-year path, showing every rebalancing trade, and computing what the discipline costs in a bull market and saves in a drawdown.
10. Compare a systematic investment plan against a lump sum on a rising and a falling price path, computing the average cost as a harmonic mean and stating why the comparison is usually the wrong one for a salaried investor.
11. Derive the index default from arithmetic rather than belief: the zero-sum identity after costs, the fee-drag identity, the horizon effect on the share of active funds that beat, and the survivorship and backfill corrections to any fund database.
12. Size an active sleeve against an indexed core, computing the gross alpha the sleeve must earn to justify itself and writing the sentence that decides which capital is allowed into it.
13. Write a one-page investment policy with an allocation band, a rebalancing trigger, an enough number, a named edge for any active capital, and a review cadence you can follow inside a drawdown.

Prerequisites & connections

Builds on. M3.02 gave you the equity risk premium three ways, and the implied route computed here is that machinery pointed at an index rather than at a single company; the historical-versus-implied argument, the country-risk stack and the Damodaran two-stage iteration all belong there and are used, not rebuilt. M3.06 gave you multiples and the discipline that a numerator and a denominator must claim the same cash, which is what makes an index earnings yield comparable to a bond yield only after the bond yield has been made real. M7.05 gave you the cycle and the Kindleberger anatomy of a bubble, so you already know how a market gets to a high multiple and what the late stages look like from the inside. M9.02 gave you portfolio construction, the concentration case, the factor menu and the active-versus-passive debate. M0.03 gave you the fee-drag identity, which reappears here as the quiet arithmetic that beats every market call in the file.

Feeds forward. E11.06 (Event-Driven Special Situations and the Canon's Unclosed Drills) is the natural destination for any capital that survives the named-edge test set out below. Its section on the performance derby is the structural mirror of the index-default argument here. The same relative-return mandates and quarterly reporting that make indexing the right default for you are what leave event-driven mispricings on the table for someone else. M9.04's process capstone consumes the policy you write in the mini-project. M10.04's lifelong-practice audit is where the review cadence you set here gets its first genuine test. The dashboard row set built in section 4.8 is the valuation block of the program's macro dashboard artefact, and the two are meant to be maintained together.

The deliberate non-overlaps are worth naming one by one. M3.02 owns the equity risk premium, including the historical route, the implied route's derivation and the country-risk surcharge; it is cited and reused at index level, never re-derived. M9.02 owns the active-versus-passive argument as a question about market efficiency and the factor menu as a set of choices; what is added here is the policy that follows from the arithmetic, not a second pass at the debate. M7.05 owns the bubble anatomy and the seven historical episodes, so no episode is re-narrated. M0.03 owns the fee-drag identity and its derivation. PW1.01 owns the private-client investment policy statement as a professional document with return objectives, risk tolerance and the full constraint set; the one-page policy written here is the learner's own, aimed at a drawdown rather than at a client file. M9.01 owns position sizing and the ruin constraint. AL1.02 owns commodities and real assets, PW1.03 owns private debt and real assets, and AL1.04 owns the allocation case for alternatives; section 4.12 borrows their conclusions in two paragraphs and sends you to them for the mechanics.

Core teaching content

4.1 Why a market has a valuation, and why almost nobody computes one

A single company has a valuation because it has cash flows you can forecast and a claim you can price. An index has one for exactly the same reason. It is a portfolio of claims on aggregate corporate cash flow, and aggregate corporate cash flow is one of the more forecastable series in economics over long horizons, because it is tethered to national output and to the profit share.

The reason the market-level number is so rarely computed is not difficulty. It is that the number is embarrassing to hold. A single-stock valuation can be wrong quietly. A market valuation is a public position on whether everyone else is mistaken, it takes years to resolve, and in the meantime the price moves against you often enough that holding the view costs social capital as well as money.

So the useful framing is not "is the market expensive" but "what would I write down, and how wide is the band". Consider the difference in what the two produce. An impression produces a mood and no record. A number produces a reading, a percentile, a band, a date, and a decision rule attached in advance. Twelve months later the impression has quietly rewritten itself to match whatever happened, while the number is still sitting there with its date on it.

There is a second reason to insist on the band rather than the point estimate, and it is the one that separates this work from market forecasting. A point estimate invites a timing decision, which the evidence cannot support. A band invites an allocation decision inside a policy you already wrote, which the evidence supports comfortably. The gauges below are built to produce bands.

A market valuation also does something a stock valuation cannot. It tells you what the average rupee or dollar of your own portfolio is priced to deliver, which is the input to every question about how much you need, how long you must work, and what withdrawal rate survives. Nothing in a single-stock file answers those.

4.9 Calibrating aggressiveness without forecasting a date

A gauge reading has to connect to an action or it is entertainment. The connection that survives the evidence is narrow: the reading sets your position inside a band you already committed to, and it sets the hurdle new capital must clear. It never sets a date, and it never takes you outside the band.

A workable ladder maps the composite percentile onto an equity weight:

The illustrative composite of 68.0 lands on 52 percent. Notice how little that is. The full range of the ladder is 30 percentage points of equity weight, so the maximum deviation from a 60 percent neutral is 15 points either way.

Which is where the arithmetic delivers its most useful result, and it is a deflating one. If equities out-earn bonds by 4 percentage points a year, a full 15-point tilt changes your expected return by 0.60 percentage points a year. The entire prize for being right about the market's valuation is 0.60 points, and it is only collected if you were right. Meanwhile, moving a core holding from a 150-basis-point active fund to a 10-basis-point index fund is worth 1.40 percentage points a year with no forecast required at all. The certain saving beats the uncertain call by a factor of 2.33.

A policy that separates the two is therefore doing something honest. The tilt is a small optimisation applied to a large, cheap, permanent position. Anybody whose valuation work is driving a larger share of their outcome than their cost discipline has the two backwards.

There is a second, better-founded use of the gauge, and it has nothing to do with the tilt. Use it to set expectations about the rate, not the timing. At a cyclically adjusted ratio of 32, the fitted forecast is 4.76 percent real a year. Feed that into a retirement plan and it changes the required savings rate, the safe withdrawal rate and the number of working years. Those are decisions with long horizons and no market-timing content, which is exactly where a weak-but-real ten-year relationship is worth having.

The discipline that separates a policy from a market call comes down to what was written before the reading arrived. A policy states the ladder, the band and the trigger in advance, and it is executed whether or not you feel like it. A market call arrives after the reading, adapts to it, and never states in advance what would make it wrong. If your allocation changed and you cannot point to the sentence you wrote before the change that mandated it, you made a call.

4.12 The index default, stated as arithmetic

The case for indexing the core does not rest on a claim that markets are efficient. It rests on four pieces of arithmetic, each of which holds whether or not markets are efficient.

The first is an identity, due to William Sharpe. Every share is owned by someone. If the passively managed rupees hold the market portfolio, then in aggregate the actively managed rupees must also hold the market portfolio, so before costs the average active rupee earns exactly the market return. After costs it earns the market return minus its costs. This is not an empirical claim that could turn out otherwise; it is bookkeeping.

Put numbers on it. A market returning 11.00 percent, passive vehicles costing 5 basis points all-in, active vehicles costing 95 basis points in fees plus 35 in trading. Passive nets 10.95 percent, the average active rupee nets 9.70 percent, and the gap of 1.25 percentage points is exactly the cost difference. Compounded, the passive rupee ends 12.00 percent ahead over ten years, 25.43 percent ahead over twenty and 40.48 percent ahead over thirty.

The second is the fee-drag identity, which says the share of terminal wealth a recurring fee consumes is 1 − ((1+g−f)/(1+g))^n and contains no wealth term at all. On 250,000 dollars at 8 percent gross for ten years, 25 basis points consumes 2.29 percent of terminal wealth (12,364 dollars), 100 basis points consumes 8.88 percent (47,943), and 200 basis points consumes 17.05 percent (92,019). Stretch the horizon and the effect becomes structural rather than annoying: on ₹50,00,000 at 12 percent gross for thirty years, a 50-basis-point fund and a 150-basis-point fund differ by ₹3,10,20,545, which is 6.20 times the whole amount originally invested and 620 times the ₹50,000 the fee difference cost in its first year.

The third is the horizon effect on the share of active funds that beat, and it falls straight out of the same two facts without any behavioural story. Model active gross returns as the market plus noise with an annual standard deviation of 6 percent and no skill, with the 1.25-point cost gap. Over one year the annualised noise is 6.00 percent and 41.75 percent of funds beat. Over five years the noise is 2.68 percent and 32.07 percent beat. Over ten it is 1.90 percent and 25.50 percent beat. Over twenty it is 1.34 percent and 17.57 percent beat. Nothing degrades; the cost is constant and the noise that could hide it shrinks with the square root of time.

The fourth is that every fund database overstates what investors actually earned. Take 300 funds at the start of a ten-year window with 35 percent closed or merged by the end, and assume the closers ran 2.50 percentage points a year behind. The survivor-only average overstates the cohort by 0.875 percentage points a year, which compounds to 9.10 percent of terminal wealth. Backfill adds more: a database that adds an incubated fund's history only after a good run, three years at 1.20 points above the truth, contributes another 0.36 points a year. Combined bias, 1.24 percentage points a year, which is almost exactly the size of the entire cost gap being measured.

The same bias corrupts the beat rate. If the true ten-year beat rate is 25.50 percent and every closure came from the losing tail, then among the survivors 39.23 percent appear to have beaten, an overstatement of 13.73 percentage points. Any published figure for how many funds beat their benchmark needs its survivorship treatment named before it means anything.

Persistence is the last leg and the weakest for the active case. The published scorecards that track this year's top-quartile funds into subsequent years find that the share still in the top quartile after several years is close to what chance would produce, and in some windows below it. Those figures are perishable and jurisdiction-specific; pull the current edition of the relevant scorecard for your own market and date it rather than quoting a remembered number.

4.14 Beyond equities, briefly

Two ideas from outside the equity market change the shape of a policy and both are owned elsewhere in the program.

The first is the endowment model: a portfolio that accepts illiquidity in return for a premium, on the reasoning that an investor with no near-term call on the capital is being paid to supply something the market is short of. The premium is real and it is smaller than the marketing suggests, because a meaningful part of what looks like return is return smoothing from infrequent, appraisal-based valuation, and because the dispersion between top and bottom managers in private assets is far wider than in public ones. The mechanics of measuring it, including the public market equivalent, belong to the private-markets sequence.

The second is the premia-across-assets frame: term premium, credit premium, equity premium and illiquidity premium are four distinct payments for four distinct discomforts, and a portfolio should hold them deliberately rather than by accident. An allocation that looks diversified by asset name is often concentrated by premium, because credit and equity are largely the same bet in different clothing.

Which is why the honest version of a multi-asset policy for an individual is short. Own the equity premium through a broad index. Own the term premium through high-grade government bonds sized to the floor and ceiling in your band. Own credit and illiquidity only where you can name what you are being paid for and what a bad outcome looks like. And treat any allocation you cannot describe in those terms as an allocation you do not understand.

The four pillars from William Bernstein's book of that name are the right closing frame for a policy. Theory, meaning the return you can expect and why. History, meaning what has actually happened to investors who did this before. Psychology, meaning the specific ways you will sabotage the plan. Business, meaning who is being paid by your decisions and how much. A policy that addresses only the first is the one most people write, and it is the one that fails in the second month of a drawdown.

4.15 Writing the policy

All of the preceding work produces one page, which you will read again under conditions you cannot currently imagine. Its quality is measured entirely by whether you follow it then. Six elements, each with a number in it.

An allocation band with a floor, a neutral and a ceiling, chosen from the gauge ladder and your own capacity to bear a drawdown. Write the floor first, because it is the one you will want to break.

A rebalancing rule with a trigger, either a collar in percentage points or a fixed calendar date, and a statement of which it is. A rule with two triggers is a rule you will apply selectively.

An enough number, defined as annual spending divided by a withdrawal rate, computed before you need it. At 12 lakh rupees of annual spending, 3.5 percent gives 3.43 crore and 3.0 percent gives 4.00 crore, a 16.7 percent difference produced entirely by a half-point assumption. At 70,000 dollars, 3.5 percent gives 2.00 million and 4.0 percent gives 1.75 million.

An index core with the vehicle named, the total expense ratio written down, and the date you checked it.

A named edge for any active capital, with all three clauses answered in writing, or the sentence "no active sleeve" instead.

A review cadence: quarterly for the gauges, annually for the policy itself, and a rule that the policy may only be amended on a review date and never within thirty days of a decision it would have prevented.

The last clause is the one that does the most work and the one most often left out. A policy you can rewrite at will is not a policy. The thirty-day rule costs you nothing when you are calm and saves you from the only decisions that matter.

Common mistakes & how experts think differently

1. Comparing a valuation ratio's level across markets or across eras. The temptation is obvious: 32 is bigger than 21, so India is dearer than it was and dearer than somewhere else. The definitions differ in earnings treatment, in index composition and in accounting convention, and the composition arithmetic in section 4.5 shows a 9.95 percent multiple change from weights alone. An expert reads the percentile of a gauge inside its own history in its own market, never the level against a foreign or historical one.

1. Quoting a cyclically adjusted ratio without naming the earnings definition. As-reported and operating earnings differ by enough to move the ratio several points, and a series that switches definitions midway is not a series. The habit that fixes it is to write the definition into the same cell as the number, so it travels with it.

1. Treating a fitted forecast as a point prediction. The line says 4.76 percent real a year at a ratio of 32. The residual says the honest interval is −1.66 to 11.18. An expert quotes the band first and the point estimate second, because a reader who hears only the point estimate will convert it into a timing decision the instrument cannot support.

1. Subtracting a nominal bond yield from an earnings yield. The arithmetic in section 4.6 shows the error equals expected inflation, which was 9.40 points in one illustrative year and 2.38 in another, and reverses the ranking between them. The expert's reflex is to ask what the real yield is before any comparison to an earnings yield is allowed to happen.

1. Reporting an implied premium without its terminal convention. Moving the terminal growth rate by 150 basis points swings the illustrative Indian premium from 1.85 to 0.46. When 92 percent of the answer sits in a perpetuity, the perpetuity's assumption is the answer. An expert states the convention, the cash-yield definition and the terminal share in the same breath as the premium, and compares series rather than levels.

1. Treating the capitalisation-to-output ratio as a cross-country valuation. The denominator measures the whole economy and the numerator prices the listed part, and the two adjustments in section 4.4 flip the India-versus-United States sign. The expert uses it against one market's own history and separately tracks whether the listed share of the economy is rising.

1. Remembering the ceiling in a defensive band and forgetting the floor. A ceiling stops greed and a floor stops capitulation, and capitulation is the one that ends careers. The band is written when calm precisely because the floor will feel indefensible at the moment it binds.

1. Believing a wide band is a rebalancing rule. Section 4.10 computes that a 25 to 75 band takes 5.98 years of a 28.6 percent equity run to bind from a 50/50 start. An expert holds the wide band as a ruin guard and adds a narrow collar if they want the discipline to generate trades.

1. Selling systematic investment as a way to beat a lump sum. It loses in expectation by 4.76 percent over a year under the stated model, and 2.09 percent after interest on idle cash. Its actual justification is that it removes twelve chances a year to talk yourself out of investing, and the dip-waiting rule that fires zero times in twelve months is the comparison that matters.

1. Reading a fund league table without asking what happened to the funds that are missing. A 35 percent closure rate with a 2.50-point shortfall in the closers overstates the cohort by 0.875 points a year and inflates the beat rate from 25.50 percent to 39.23 percent. The expert's first question about any performance statistic is which funds were in the sample at the start and which are in it now.

1. Sizing an active sleeve without computing what it must earn. A 10 percent sleeve needs 6.40 points of gross alpha a year to lift the portfolio by half a point. Most people who would fail that test have never calculated it, which is why the sleeve feels harmless.

1. Spending more effort on the market call than on the cost line. The full tilt is worth 0.60 percentage points a year and only if correct; the fee decision is worth 1.40 points and is certain. An expert does the cheap, certain thing first and treats the valuation work as what it is, a modest refinement to a position that was going to be held regardless.

Worked examples

Every figure below was recomputed in Python before it was written down. Index series are illustrative and internally consistent; they are shaped to behave the way the published series behave and are not any market's actual readings on any date. Each example names the primary source you would use to rebuild it with live data.

Worked example 1: A cyclically adjusted ratio and an implied premium for an Indian index (India, ₹)

You have ten fiscal years of earnings per index unit for an illustrative broad Indian index, a consumer price index on a FY2017 base, and a current level of 26,000. Build the cyclically adjusted ratio, compare it with the trailing multiple, then solve the implied equity risk premium on the same index and state where the two gauges agree.

Solution. Step 1, restate each year's earnings into FY2026 rupees. The deflator is the current price index divided by that year's.

Step 2, average and divide. The ten-year mean of real earnings is 812.4447. The ratio is 26,000 ÷ 812.4447 = 32.0022. The trailing multiple is 26,000 ÷ 1,120 = 23.2143.

Step 3, read the gap. The ratio sits 1.3786 times the trailing multiple, and the whole of that gap is growth: nominal earnings compounded at 12.5975 percent over the nine intervals and real earnings at 7.4790 percent, against price-index growth of 4.7623 percent a year.

Step 4, the check that catches the commonest error. Average the nominal earnings instead and the mean is 686.50, giving a ratio of 37.8733. Skipping the deflator overstates the reading by 5.8711 points, 18.3459 percent, in a period whose price level rose 52.00 percent.

Step 5, place it in its own history. Against ten prior annual readings of 20.1, 21.4, 24.8, 26.1, 19.6, 34.6, 30.4, 26.7, 29.8 and 31.2, the current 32.00 has nine below it, the 90th percentile, against a prior-decade mean of 26.47.

Step 6, the implied premium on the same index. Dividend yield 1.30 percent and net buyback yield 0.25 percent give a first-year cash flow of 26,000 × 0.0155 = 403.00. Grow it at 12.00 percent for five years, then at the risk-free rate of 6.90 percent in perpetuity, and solve for the rate that reproduces 26,000. It is 8.7540 percent, so the implied premium is 1.8540 percentage points. Verification: discounting the five explicit flows and the terminal value at 8.7540 percent returns 26,000.0000, a residual below 1e-6.

Step 7, the caveat that comes with it. The five explicit years contribute 1,966.76 of value, 7.56 percent of the level. The terminal value contributes 24,033.24, 92.44 percent. Move the terminal growth rate down by 150 basis points and the premium falls from 1.85 to 0.46 points; move the five-year growth rate by 200 basis points either way and it moves only between 1.73 and 1.98. The convention nobody argues about dominates the forecast everybody argues about.

Read it. Both gauges say the same thing in different units: at the 90th percentile of its own decade on one and at a thin 1.85-point premium on the other, the index is priced for a poor real return rather than a good one. They disagree about how poor, and the disagreement is not resolvable from the data, because the premium's answer is 92 percent a function of a terminal assumption. Neither number says anything at all about next year. If the earnings series switched definition anywhere in the ten years, or the price index base was misread, every figure in the table is wrong in a way that will not look wrong.

Worked example 2: The same two gauges for a US index, the Fed model demolished, and the ten-year fee drag (US, $)

Build the cyclically adjusted ratio for an illustrative United States index on as-reported earnings, solve its implied premium, then show arithmetically that comparing an earnings yield to a nominal bond yield reverses the ranking between two years. Close by pricing ten years of fees on a stated portfolio.

Solution. Step 1, the ratio. Earnings of 110, 132, 139, 94, 198, 172, 193, 209, 230 and 250 against a price index of 100.0, 102.4, 104.3, 105.6, 110.6, 119.5, 124.2, 128.0, 131.5 and 134.8 restate to 148.28, 173.77, 179.65, 119.99, 241.32, 194.02, 209.47, 220.10, 235.77 and 250.00. The mean is 197.2377, so at a level of 6,800 the ratio is 34.4762 and the trailing multiple is 27.2000, a factor of 1.2675. Against ten prior readings of 26.4, 28.9, 24.1, 27.6, 33.8, 37.2, 28.4, 31.9, 33.6 and 35.9, eight sit below, the 80th percentile.

Step 2, what the 2020 year is doing. The 94 in 2020 is an as-reported figure carrying impairments. Had real 2020 earnings been 200 units lower still, the ratio would read 38.3665 rather than 34.4762 with no change in price. The general form: nine normal years at 100 and one at −50 lift the ratio by 17.6471 percent.

Step 3, the implied premium. Dividend yield 1.25 percent plus net buyback yield 1.80 percent gives 3.05 percent, so the first-year cash flow is 207.40. Grow at 10.50 percent for five years then at the risk-free rate of 4.25 percent. Solved rate 8.0304 percent, implied premium 3.7804 percentage points, with the terminal value supplying 85.22 percent of the level. The Gordon one-liner gives 3.05 + 4.25 = 7.30 percent and a premium of 3.05 points, within 0.73 of the full solve.

Step 4, the Fed model in a high-inflation year. Earnings yield 12.00 percent, nominal ten-year yield 13.90 percent, expected inflation 9.00 percent. The exact real yield is (1.139 ÷ 1.09) − 1 = 4.4954 percent. The Fed-model spread is 12.00 − 13.90 = −1.90 points, so stocks look expensive. The correct spread is 12.00 − 4.4954 = +7.50 points, so stocks look very cheap. The error is 9.4046 points, which equals expected inflation plus its cross term with the real yield.

Step 5, the same in a low-inflation year. Earnings yield 3.40 percent, nominal yield 1.50 percent, expected inflation 2.40 percent, exact real yield −0.8789 percent. Fed spread +1.90 points; correct spread +4.28 points; error 2.3789 points.

Step 6, the reversal. The Fed model ranks the low-inflation year above the high-inflation year by 3.80 points. The real-yield comparison ranks the high-inflation year above the low-inflation year by 3.23 points. The gap between the two errors, 7.03 points, tracks the 6.60-point gap in expected inflation almost exactly. Inverting the nominal yields gives Fed-model "fair" multiples of 7.19 and 66.67; using the real yields plus a 4 percent premium gives 11.77 and 32.04.

Step 7, ten years of fees on 250,000 dollars at 8 percent gross. Gross terminal value is 539,731.25. The share of terminal wealth consumed is 1 − ((1+g−f)/(1+g))^n.

Moving from 25 to 100 basis points costs 35,578.97; moving from 100 to 200 costs a further 44,075.92. The 100-basis-point drag expressed as an annualised equivalent is 0.9259 percent a year.

Read it. The 8.88 percent that a 100-basis-point fee takes over ten years is certain, arrives whether the market call was right or wrong, and is set by a decision you make once in an afternoon. The valuation work in steps 1 to 3 produces a forecast whose 90 percent band is 12.83 points wide. Ranked by expected value per hour of effort, the fee decision wins by a distance. If the expected-inflation figure in steps 4 and 5 came from a source you cannot name and date, redo them; the entire demonstration rests on it.

Worked example 3: Market capitalisation to output for both markets, with the denominator problem quantified (India ₹ and US $)

Compute the ratio for both markets, then repair each denominator and show what happens to the cross-market comparison.

Solution. Step 1, India raw. Listed market capitalisation of ₹460 lakh crore against nominal output of ₹360 lakh crore gives 127.78 percent.

Step 2, India's own history. Prior-decade readings of 76.1, 78.6, 83.0, 79.9, 56.2, 103.0, 111.0, 96.3, 130.8 and 130.8 percent put the current reading eighth of ten, the 80th percentile, against a prior mean of 94.57 percent.

Step 3, India's denominator repair. On an illustrative assumption that listed companies account for 35 percent of value added, with agriculture and the unorganised sector making up much of the rest, the ratio against listed-sector output is 460 ÷ (360 × 0.35) = 365.08 percent.

Step 4, the United States raw. Market capitalisation of $62.0 trillion against output of $30.5 trillion gives 203.28 percent.

Step 5, the United States numerator repair. On an illustrative 30 percent foreign-revenue share, the domestic claim is 62.0 × 0.70 = $43.4 trillion, so the ratio is 142.30 percent. The equivalent operation on the denominator is 30.5 ÷ 0.70 = $43.5714 trillion, giving 142.30 percent by an independent route.

Step 6, the comparison, before and after.

Read it. The sign flipped without a single fact about either market changing. Both adjustments are estimates and both carry wide uncertainty, so neither adjusted figure should be quoted as a valuation; what they establish is that the raw cross-country comparison carries no information at all, because the two denominators are measuring different things. The gauge is legitimate against one market's own history, which is how step 2 uses it, and there its message is specific: the listed sector's capitalisation has outrun the economy for several years. To find out whether that is a rising listed share or a rising multiple, track listed-sector revenue against nominal output, since the first moves it and the second does not.

Worked example 4: A 25 to 75 band across a five-year path, with every trade (India, ₹)

A book of ₹20,00,000 starts at 50 percent equity and 50 percent high-grade bonds. Equity returns run 37.6, 23.0, 33.4, 28.6 and 21.0 percent over five years, a geometric mean of 28.5704 percent; bonds return 7.0 percent each year. Run two policies: a bare 25 to 75 band with no rebalancing inside it, and the same band plus a 45 to 55 collar that rebalances to 50/50 on breach.

Solution. Step 1, Policy A, the bare band.

Zero trades in five years. The ceiling never binds even after a run of that size.

Step 2, how long it would take. Starting at 50/50, the equity-to-bond ratio must triple to reach a 75 percent weight. At a 28.5704 percent geometric equity return against 7.0 percent bonds, that takes ln(3) ÷ ln(1.285704 ÷ 1.07) = 5.9822 years. On the other side, driving a 50/50 book to the 25 percent floor takes ln(3) ÷ ln(1.07 ÷ 0.80) = 3.7779 consecutive years of a 20 percent equity decline.

Step 3, Policy B, the collar.

Ending book: equity ₹2,300,499.17, bonds ₹2,300,499.17, total ₹4,600,998.34.

Step 4, the cost of the discipline. Policy A ends at ₹4,915,774.70 and Policy B at ₹4,600,998.34. The collar cost ₹314,776.37, which is 6.8415 percent of Policy B's terminal value, over a five-year period in which equities did nothing but rise.

Step 5, run both through a crash. Add three years at −37.0, −22.0 and +26.0 percent for equity with bonds still at 7.0 percent.

Read it. The collar bought 10.8396 percentage points less drawdown and, over this particular eight-year path, ₹320,980.84 more terminal wealth, which is 7.6162 percent of the collared book's own ending value and 8.2441 percent more than the uncollared book ended with. Over a different path it would have lost, and it lost on the first five years of this one. The insurance framing is the honest one: the premium is visible every year and the payout is rare and large. What the arithmetic settles is the narrower question of what the bare 25 to 75 band does in ordinary conditions, and the answer is nothing at all, for years at a time. It is a guard against ruin and against being fully out, and it needs a tighter rule inside it to generate any discipline in a normal decade.

Worked example 5: Systematic investment against a lump sum, rising and falling (India, ₹)

₹10,000 a month for twelve months, ₹1,20,000 in total. Compute the average cost per unit on a rising path from ₹100 to ₹155 in ₹5 steps and on a falling path from ₹100 to ₹45 in ₹5 steps, both against deploying the whole ₹1,20,000 at the first month's price. Idle cash earns 6.5 percent a year.

Solution. Step 1, the rising path. Units bought total 959.011077, so the average cost is 120,000 ÷ 959.011077 = ₹125.128899. The arithmetic mean of the twelve prices is ₹127.500000 and the harmonic mean is ₹125.128899, identical to the average cost by construction. The advantage over the arithmetic mean is ₹2.371101, 1.8949 percent of the average cost.

Step 2, the falling path. Units total 1,759.765029, so the average cost is ₹68.190922. The arithmetic mean is ₹72.500000 and the harmonic mean is again exactly the average cost. The advantage is ₹4.309078, 6.3191 percent of the average cost, larger because the prices are more dispersed.

Step 3, against the lump sum. A lump sum at ₹100 buys 1,200.000000 units on either path.

Step 4, credit the idle cash. The plan leaves 660,000 rupee-months of cash undeployed, which at 6.5 percent earns ₹3,575.00. The rising-path shortfall narrows to ₹33,778.28 and the falling-path advantage widens to ₹28,764.43.

Step 5, the expected value under a stated model. At an expected monthly total return of 0.90 percent, equivalent to 11.3510 percent a year, one rupee invested at the start grows to 1.113510 by month twelve. The average rupee in the plan is invested for 6.5 months and grows to 1.060475, a shortfall of 4.7629 percent. Adding interest on the idle cash lifts it to 1.090266 and cuts the shortfall to 2.0874 percent.

Step 6, the comparison that actually applies. Run a rule that deploys only after a 10 percent fall from the running peak, against the rising path. The trigger fires in zero of twelve months and the rule deploys ₹0 while the price rises 55 percent.

Read it. Averaging in loses to a lump sum in expectation, and the loss is small. The claim it should be sold on is step 6: against the alternative most people actually run, which is waiting for a better price, a rule that buys every month regardless is overwhelmingly better because the waiting rule never fires. The harmonic-mean advantage in steps 1 and 2 is real but is not a free lunch; it is arithmetic that follows from buying a fixed rupee amount, and it says nothing about whether the terminal value is good. Note also that the model in step 5 assumes a positive expected monthly return; that assumption is doing all the work, and it is the assumption the valuation gauges are least able to confirm.

Worked example 6: The zero-sum identity and the survivorship correction (US $, with an India ₹ tail)

A market returns 11.00 percent. Passive vehicles cost 5 basis points all-in; active vehicles cost 95 basis points in fees plus 35 in trading. Compute the aggregate outcome, the share of active funds that beats by horizon, and the correction a fund database needs.

Solution. Step 1, the identity. If passive holds the market, active in aggregate also holds the market, so before costs the average active dollar earns 11.00 percent. Passive nets 11.00 − 0.05 = 10.95 percent. The average active dollar nets 11.00 − 1.30 = 9.70 percent. The gap is 1.2500 percentage points, which is exactly the cost difference and not an empirical result.

Step 2, compound it. The passive dollar ends 11.9971 percent ahead over ten years, 25.4335 percent over twenty and 40.4819 percent over thirty.

Step 3, the share that beats. Model active gross returns as the market plus noise with an annual standard deviation of 6.00 percent and no skill. The annualised noise shrinks with the square root of the horizon while the cost gap does not.

Step 4, survivorship. Of 300 funds at the start of a ten-year window, 35 percent close or merge, and the closers run 2.50 percentage points a year behind. The survivor-only average overstates the cohort by 0.35 × 2.50 = 0.8750 percentage points a year, which compounds to 9.1027 percent of terminal wealth over the decade.

Step 5, backfill. A database that adds an incubated fund's history only after a good run contributes three years at 1.20 points above the truth, spread over ten years: 0.3600 percentage points a year. Combined database bias is 1.2350 points a year, almost exactly the size of the cost gap the study is trying to measure.

Step 6, the beat rate corrected. If the true ten-year beat rate is 25.5010 percent and every closure came from the losing tail, the survivor-only rate is 25.5010 ÷ 0.65 = 39.2322 percent, an overstatement of 13.7313 percentage points.

Step 7, the India tail. Regular and direct plans of the same Indian fund differ by roughly 100 basis points. On ₹50,00,000 at 12 percent gross for thirty years, a 50-basis-point plan ends at ₹13,09,83,329.55 and a 150-basis-point plan at ₹9,99,62,784.48, a difference of ₹3,10,20,545.07. The drags are 12.5610 percent and 33.2690 percent of terminal wealth respectively, and both figures are independent of the amount invested.

Read it. Steps 1 and 2 are bookkeeping and hold in any market at any level of efficiency. Step 3 needs a distributional assumption, so treat the specific percentages as illustrative and the shape as the durable part: a constant cost against shrinking noise means the beat rate falls with horizon, mechanically. Steps 4 to 6 mean that any published statistic about active performance is an overstatement until its survivorship treatment is named, and the size of the overstatement is comparable to the entire effect being measured. What none of this establishes is that no manager has skill; it establishes what the average dollar earns and what the odds look like before you have identified one.

Worked example 7: Reading the dashboard when the gauges disagree (India ₹, with a US $ contrast)

Assemble the five gauges for the illustrative Indian index, compute each percentile, take the composite to the ladder, and resolve the disagreement.

Solution. Step 1, the fifth gauge, which needs building. The nominal ten-year government yield is 6.90 percent and expected inflation is 4.50 percent, so the exact real yield is (1.069 ÷ 1.045) − 1 = 2.2967 percent. The trailing earnings yield is 1 ÷ 23.2143 = 4.3077 percent. The spread is 2.0110 percentage points.

Step 2, the table.

For the two gauges where a low reading means dear, the percentile is inverted before it enters the table.

Step 3, the composite. Mean 68.00, median 80.00, range 20 to 90, spread 70 percentile points.

Step 4, the ladder. A composite of 68.00 falls in the 60-to-80 band, which sets equity at 52 percent, eight points below a 60 percent neutral and well inside a 25 to 75 band.

Step 5, resolve the disagreement. Four gauges are absolute and one is relative. The relative gauge reads 20 because the real bond yield is 2.2967 percent, low against its own decade, so equities look tolerable beside a poor alternative. The absolute gauges read 80 to 90 because the price is high against earnings and against output. Both are correct and they answer different questions.

Step 6, the second disagreement. The trailing multiple's 60th percentile against the cyclically adjusted ratio's 90th says current earnings sit far above their own ten-year average, by a factor of 1.3786. That is either a durable level shift or a cycle peak, and the test is observable: check the profit share of national income and index-level margins against their own histories. If both are at records, weight the cyclically adjusted reading. If neither is, weight the trailing one.

Step 7, the United States contrast. The same five gauges on the illustrative United States inputs give a cyclically adjusted ratio at the 80th percentile and an implied premium of 3.7804 points, which is thin but far from the Indian 1.8540. The gap is mostly the risk-free rate: 4.25 percent against 6.90 percent. Holding the United States inputs fixed and moving the risk-free rate from 4.25 to 6.00 percent moves the premium only from 3.78 to 3.56 points, because the price is fixed and most of the rate move passes into the solved discount rate.

Read it. A 70-point spread across five gauges is the normal state, not a failure. The composite is a summary that hides the informative part, so report the range beside it always. The action the table licenses is an eight-point tilt inside a band, worth about 0.32 percentage points a year at a 4 percent equity premium, which is smaller than the fee decision sitting elsewhere in the same portfolio. The action it does not license is a date, an exit, or any claim about the next twelve months.

Practice set

Work each problem fully before reading its solution. A calculator is assumed; numeric answers within about ±1 percent score as correct unless the problem states otherwise. All series are illustrative and internally consistent.

P1 (guided). Build a five-year cyclically adjusted ratio. Earnings per index unit for five years are 52, 58, 44, 61 and 70. The price index for the same years is 100, 104, 108, 113 and 118. The index level is 1,450. Compute the restated earnings, the mean, the cyclically adjusted ratio and the trailing multiple, and say why the two differ.

Solution. Deflators are 1.18, 1.134615, 1.092593, 1.044248 and 1.000000, giving restated earnings of 61.3600, 65.8077, 48.0741, 63.6991 and 70.0000. The mean is 61.788176. The ratio is 1,450 ÷ 61.788176 = 23.4673. The trailing multiple is 1,450 ÷ 70 = 20.7143. The ratio is higher because the mean of a rising real series sits below its latest value, so the cyclically adjusted denominator is smaller than the trailing one. A five-year window is too short to span a cycle and is used here only to keep the arithmetic visible.

P2 (guided). The missing deflator. Using P1's earnings, compute the ratio without the inflation adjustment and state the overstatement in points and percent.

Solution. The nominal mean is (52 + 58 + 44 + 61 + 70) ÷ 5 = 57.0000, so the naive ratio is 1,450 ÷ 57 = 25.4386. That is 1.9713 points above the correct 23.4673, an overstatement of 8.40 percent, produced by an 18 percent rise in the price level across the window. Scale the window to ten years and the price level to a 52 percent rise, as in Worked Example 1, and the overstatement grows to 18.35 percent.

P3 (guided). Read a fitted line honestly. On the line ten-year annualised real return = 15.00 − 0.32 × ratio, with a residual standard deviation of 3.90 points, state the point forecast and the 90 percent band at a ratio of 25.00, and convert both ends into a cumulative real multiple over ten years.

Solution. The point forecast is 15.00 − 0.32 × 25.00 = 7.000 percent a year. The 90 percent band is 7.000 ± 1.645 × 3.90 = 0.585 percent to 13.415 percent. Cumulative real multiples are 1.00585^10 = 1.060× and 1.13415^10 = 3.521×. So the honest statement is that a rupee's real value in ten years is somewhere between roughly unchanged and three and a half times, with a central estimate near double. Anyone who reports only the 7.000 percent has thrown away the part of the answer that determines what you should do with it.

P4. The write-off distortion. A ten-year earnings series runs at 120 a year except for one year at −60. Compute the ten-year mean, the mean of a clean series at 120 throughout, and the percentage by which the ratio is inflated at an unchanged price.

Solution. Clean mean 120.0000. Distorted mean = (9 × 120 − 60) ÷ 10 = (1,080 − 60) ÷ 10 = 102.0000. The ratio is inflated by 120 ÷ 102 = 1.176471, or 17.6471 percent, at an unchanged price. This is the same factor as Worked Example 2's nine-at-100-and-one-at-minus-50 case, because both have the bad year at −0.5 times the normal level.

P5. The composition effect. A market has a high-multiple sector at a price-to-earnings ratio of 50 and everything else at 16. Compute the index multiple at a 20 percent weight and at a 35 percent weight, and state how much of any observed re-rating is composition.

Solution. The index multiple is the market-value-weighted harmonic mean, 1 ÷ Σ(wᵢ ÷ PEᵢ). At 20 percent: 1 ÷ (0.20 ÷ 50 + 0.80 ÷ 16) = 1 ÷ 0.054 = 18.5185. At 35 percent: 1 ÷ (0.35 ÷ 50 + 0.65 ÷ 16) = 1 ÷ 0.0476250 = 20.9974. The rise is 2.4789 points, 13.3858 percent, with no sector's own multiple changing. An index that re-rated from 18.5 to 21.0 over the period in question re-rated by exactly zero.

P6. The Fed model, one case. Earnings yield 4.80 percent, nominal ten-year yield 7.10 percent, expected inflation 4.50 percent. Compute the exact real yield, both spreads and the error, and state the verdict each spread implies.

Solution. Exact real yield = (1.071 ÷ 1.045) − 1 = 2.4880 percent. Fed-model spread = 4.80 − 7.10 = −2.3000 points, implying equities are expensive against bonds. Correct spread = 4.80 − 2.4880 = +2.3120 points, implying equities are modestly cheap. The error is 4.6120 points, which is expected inflation of 4.50 plus the 0.1120 cross term. The two spreads carry opposite signs, so this is not a precision problem.

P7. Solve an implied premium with the Gordon one-liner. An index sits at 3,200 with a total cash yield to shareholders of 2.10 percent and a terminal growth rate of 5.00 percent, against a risk-free rate of 4.30 percent. Then repeat for an index at 18,500 with a cash yield of 1.45 percent and a terminal growth rate of 6.70 percent, against a risk-free rate of 6.70 percent.

Solution. Expected return equals cash yield plus growth. First: 2.10 + 5.00 = 7.1000 percent, so the implied premium is 7.10 − 4.30 = 2.8000 points. Second: 1.45 + 6.70 = 8.1500 percent, premium 8.15 − 6.70 = 1.4500 points. The second market's thin premium is a direct consequence of a low cash yield against a high risk-free rate, and it is the reason the buyback yield must be included; omit a 0.5-point buyback yield and the premium reads 0.95 rather than 1.45, a third lower.

P8. Terminal-convention sensitivity. On the illustrative Indian inputs the two-stage solve gives an implied premium of 1.8540 points with the terminal growth rate set to the 6.90 percent risk-free rate, and 0.4566 points with it set 150 basis points lower. State the percentage change in the premium and explain why the sensitivity is so large.

Solution. The premium falls by 1.3974 points, a 75.37 percent reduction, from a 150-basis-point change in one convention. The cause is the perpetuity denominator: the solved rate sits only 1.85 points above the terminal growth rate, so the terminal value carries 92.44 percent of the index level and small changes in r − g move it enormously. The practical rule is that an implied premium computed with a terminal share above about 85 percent is a statement about the convention, and only the change in the series over time is safe to read.

P9. Repair a capitalisation-to-output ratio. A market's listed capitalisation is 3.4 trillion against output of 2.1 trillion, and listed companies account for 40 percent of value added. Compute the raw and repaired ratios and say what each one is a statement about.

Solution. Raw = 3.4 ÷ 2.1 = 161.90 percent. Repaired = 3.4 ÷ (2.1 × 0.40) = 3.4 ÷ 0.84 = 404.76 percent. The raw ratio is a joint statement about the valuation of the listed sector and the share of the economy that is listed, and those move independently. The repaired ratio isolates valuation but inherits all the uncertainty in the 40 percent estimate, so it should be used to demonstrate that the raw cross-market comparison is meaningless rather than quoted as a valuation in its own right.

P10. A band breach and two possible trades. A book holds ₹31,50,000 of equity and ₹8,50,000 of bonds against a 25 to 75 band with a 60 percent neutral. Compute the current weight, the sale that brings it to the ceiling, and the sale that brings it to neutral.

Solution. Total ₹40,00,000, so the weight is 31.5 ÷ 40 = 78.7500 percent, above the ceiling. To reach 75 percent, equity must fall to ₹30,00,000, a sale of ₹1,50,000. To reach the 60 percent neutral, equity must fall to ₹24,00,000, a sale of ₹7,50,000. Which one is correct is a policy question you answer before the breach: rebalancing to the edge trades least and breaches again on the next move, while rebalancing to neutral trades most and buys the longest interval before the next trade. A rule that decides at the moment of breach is not a rule.

P11. The average cost of a systematic plan. Six monthly contributions of ₹6,000 buy at unit prices of ₹200, ₹160, ₹125, ₹160, ₹200 and ₹250. Compute the units, the average cost, the arithmetic mean price, the ending value at ₹250 and the value a lump sum of ₹36,000 at ₹200 would have reached.

Solution. Units = 30 + 37.5 + 48 + 37.5 + 30 + 24 = 207.000000. Average cost = 36,000 ÷ 207 = ₹173.913043, which is the harmonic mean of the six prices. Arithmetic mean = ₹182.500000, so the plan's cost advantage is ₹8.586957, 4.9375 percent of the average cost, on the same basis as Worked Example 5. Ending value = 207 × 250 = ₹51,750.00. Lump sum = (36,000 ÷ 200) × 250 = 180 × 250 = ₹45,000.00. The plan wins here by ₹6,750 because the path dipped 37.5 percent before recovering; on a monotonic rise it would have lost.

P12. Fee drag at three horizons. Using 1 − ((1+g−f)/(1+g))^n, compute the share of terminal wealth consumed by (a) 75 basis points at 12 percent gross over 20 years, (b) 150 basis points at 10 percent gross over 25 years, and (c) 30 basis points at 8 percent gross over 40 years.

Solution. (a) (1.1125 ÷ 1.12)^20 = 0.874259, so the drag is 12.5741 percent. (b) (1.085 ÷ 1.10)^25 = 0.709457, drag 29.0543 percent. (c) (1.077 ÷ 1.08)^40 = 0.894701, drag 10.5299 percent. Case (c) is the instructive one: 30 basis points sounds trivial and consumes more than a tenth of a lifetime's terminal wealth, because the identity depends on the fee and the horizon and contains no wealth term at all.

P13. The beat rate at two horizons. With a cost gap of 1.25 percentage points and active gross returns distributed around the market with an annual standard deviation of 6.00 percent and no skill, compute the share of funds beating the index over five years and over fifteen.

Solution. At five years the annualised standard deviation is 6.00 ÷ √5 = 2.6833 percent, so z = 1.25 ÷ 2.6833 = 0.465847 and the share beating is 1 − Φ(0.465847) = 32.0662 percent. At fifteen years it is 6.00 ÷ √15 = 1.5492 percent, z = 0.806872, share 20.9870 percent. The decline needs no story about skill decaying; the cost is constant while the noise that could disguise it shrinks with the square root of time.

P14. Correct a league table for survivorship. Of a starting cohort, 28 percent of funds closed or merged over fifteen years, and the closers ran 3.00 percentage points a year behind the survivors. Compute the annual bias in the survivor-only average and its compounded effect, then state what you would ask the table's publisher.

Solution. Annual bias = 0.28 × 3.00 = 0.8400 percentage points a year. Compounded over fifteen years, 1.0084^15 − 1 = 13.3685 percent of terminal wealth. The question for the publisher is not whether they adjusted but which of the two adjustments they made: whether dead funds are included at their actual returns to the closure date, and whether any fund's pre-listing history was backfilled. A table that answers neither is unusable, because the bias is comparable in size to the effect it reports.

P15. Size an active sleeve. A book runs a 25 percent active sleeve costing 175 basis points against an index core costing 7 basis points. Compute the break-even gross alpha the sleeve must earn, and the gross alpha needed to lift the whole portfolio by 0.40 percentage points a year.

Solution. Break-even = 175 − 7 = 1.6800 percentage points of gross alpha, independent of the sleeve's weight, because it is set by the cost difference on the capital the sleeve displaced. For a 0.40-point portfolio lift, the sleeve's net contribution must be 0.40 ÷ 0.25 = 1.60 points on its own capital, so the required gross alpha is 1.68 + 1.60 = 3.2800 percentage points a year. Sustaining 3.28 points of gross alpha annually is a professional-grade record; the arithmetic is the honest test of whether the sleeve should exist.

P16. Compute an enough number and its sensitivity. Annual spending is ₹18,00,000. Compute the corpus implied by a 3.25 percent withdrawal rate and by a 4.00 percent rate, and state the difference as a percentage.

Solution. At 3.25 percent: 18,00,000 ÷ 0.0325 = ₹5,53,84,615.38. At 4.00 percent: ₹4,50,00,000.00. The lower rate raises the target by ₹1,03,84,615.38, 23.08 percent. The withdrawal rate is not a market forecast but a judgement about sequence risk and longevity, and it moves the finish line by a quarter, which is why the policy has to state it explicitly rather than leave it implied.

Applied mini-project

Write the one-page policy you will still be able to follow in the twentieth month of a drawdown.

Six artefacts, each of which must pass an explicit test before it counts. Use a spreadsheet or a notebook, show every tie-out rather than asserting it, and put a retrieval date on every perishable input. Label any illustrative series as illustrative.

Artefact 1, the gauge table (2 tests). Build all five gauges for one market from primary sources: a cyclically adjusted ratio, a capitalisation-to-output ratio, an index multiple against its own history, an earnings yield less a real bond yield, and a two-stage implied premium. Test 1: the deflated earnings series reproduces its ten-year mean to six decimals, and the same ratio computed without the deflator is reported alongside so the inflation adjustment's size is visible. Test 2: the solved discount rate, fed back through the two-stage model, reproduces the index level to within 1e-6, and the terminal value's share of the level is stated next to the premium.

Artefact 2, the percentiles and the disagreement (1 test). For each gauge, ten prior annual readings and the current reading's percentile inside them, with the direction of dearness stated for each. Test 3: report the composite, the median and the range, and write one paragraph naming the largest disagreement in the table and which underlying question each side of it answers.

Artefact 3, the band and the trigger (2 tests). State a floor, a neutral and a ceiling, and a rebalancing trigger that is either a collar in percentage points or a fixed calendar date. Test 4: run the band across a stated five-year path and list every trade, with the ending allocation. Test 5: run the same path with the trigger removed and report the difference in terminal wealth and in maximum drawdown, so the discipline's cost and benefit are both quantified rather than assumed.

Artefact 4, the enough number (1 test). Annual spending, a withdrawal rate, and the corpus that follows, plus the same corpus at withdrawal rates 75 basis points either side. Test 6: compute the years to the target from your current corpus and saving rate at a stated real return, then recompute with 100 basis points of fees added, and report the difference in years.

Artefact 5, the cost line (1 test). Every vehicle you hold, its total expense ratio, the date you checked it, and the blended cost of the whole book. Test 7: apply the fee-drag identity at your own horizon and report the share of terminal wealth your current blended cost consumes, alongside the share a plausible cheaper alternative would consume.

Artefact 6, the sleeve decision. Either the sentence "no active sleeve", or a named edge with all three clauses answered: what the edge is, who is on the other side and why they are selling, and what written evidence you have that you possess it. If a sleeve exists, compute the gross alpha it must earn to break even and to lift the whole portfolio by half a point.

Deliverable. One page carrying the six elements from section 4.15, plus a working file behind it with the seven test residuals, every input's source and retrieval date, and a closing paragraph naming the single assumption your policy is most sensitive to with the sensitivity quantified.

Rubric (21 points; pass at 17 with no zero on any row).

Rows score 2 at the stated standard and 3 at the stronger one. Any artefact whose test fails scores zero for that row regardless of how well the page reads, because a policy that rests on an unverified number is a policy you will abandon the first time the number is questioned.

Reading & resources

Each item here is an anchor to verify against, never a source to copy a number from without tracing it yourself.

Valuation at market level:

• Robert J. Shiller, Irrational Exuberance, the book that put the cyclically adjusted ratio into general use, and the source of the long-run data set. [Paid] [Intermediate]
• Robert Shiller's online data set (Yale, "U.S. Stock Markets 1871-Present and CAPE Ratio"), the primary source for the United States series, downloadable and monthly. [Free] [Intermediate]
• Aswath Damodaran, Equity Risk Premiums: Determinants, Estimation and Implications, updated annually, plus his monthly implied-premium series; the reference implementation for section 4.7. [Free] [Intermediate to Advanced]
• Elroy Dimson, Paul Marsh and Mike Staunton, Triumph of the Optimists, the long-run cross-country record of equity returns and the best correction to any single market's history. [Paid] [Intermediate]
• Jeremy Siegel, Stocks for the Long Run, the case for the equity premium's persistence, worth reading against Dimson, Marsh and Staunton rather than instead of them. [Paid] [Beginner to Intermediate]

Policy, indexing and costs:

• Benjamin Graham, The Intelligent Investor, chapters 4 and 5 for the defensive policy and the 25-to-75 band. [Paid] [Beginner]
• John C. Bogle, Common Sense on Mutual Funds, the fee-drag argument at book length by the person who acted on it. [Paid] [Beginner to Intermediate]
• William F. Sharpe, The Arithmetic of Active Management, 1991, two pages that settle section 4.12's first argument permanently. [Free] [Beginner]
• Charles D. Ellis, Winning the Loser's Game, the practitioner's version of the same argument. [Paid] [Beginner]
• William J. Bernstein, The Four Pillars of Investing, the frame used to close section 4.14. [Paid] [Beginner to Intermediate]

Primary sources for the data you will actually use:

• NSE Indices factsheets and daily index statistics for Indian index-level earnings, price-to-earnings and price-to-book, and dividend yield. [Free] [Beginner]
• MOSPI for the Indian consumer price index and the national accounts, and RBI's Handbook of Statistics on the Indian Economy for the long series. [Free] [Beginner to Advanced]
• FRED (Federal Reserve Bank of St. Louis) for United States price indices, Treasury yields and inflation-linked yields, and BEA for national accounts. [Free] [Beginner]
• S&P Dow Jones Indices for index-level earnings, buyback and dividend data, and for the foreign-revenue share of the index. [Free to metered] [Intermediate]
• SPIVA scorecards and the Persistence Scorecards, published by market including India, for current beat rates and persistence; pull the current edition and date it rather than quoting a remembered figure. [Free] [Beginner to Intermediate]
• AMFI for Indian fund expense ratios and direct-versus-regular plan data, and the fund's own scheme information document for the fee schedule. [Free] [Beginner]

Do this, not just read: build Worked Example 1's table in a spreadsheet, then break it deliberately. Change the price-index base year from FY2017 to FY2020 without rescaling the series and watch the ratio move while every input still looks correct. Ten minutes of watching a base-year error produce a plausible wrong answer teaches more about why the source and the date travel with the number than any warning can.

Case Lab: run XL-11.5. The Market's Own Record: Valuation Series, Primary Sources after Worked Examples 1 to 3. It sends you to the published series themselves and asks for the two gauges rebuilt from them, with the retrieval date carried per cell and the cross-country comparison adjusted rather than quoted. The write-before-peek rule applies to every published figure it names.

The drill itself, in full. Rebuild two gauges for one market from the published long-run series rather than from the illustrative tables here. First a cyclically adjusted ratio, from the downloadable earnings and price-index data. Then an implied premium, from the index level, the cash yield and your own stated terminal-growth convention. Report both with the retrieval date and the earnings definition attached, and show the same ratio computed without the deflator alongside.

Now write down every place your reconstruction disagrees with the published figure, and the reason for each. The usual causes are a different earnings definition, a different price-index base, a different buyback treatment and a different terminal cap. Naming the cause matters more than closing the gap, and the write-before-peek rule applies to the published number.

The Modern Analyst's Addendum

The teaching above builds every gauge by hand, which is how you learn it and how the gate tests it. What follows is how a working analyst amplifies the same skill. It adds; it never replaces.

AI-Augment this skill

{
  "skill": "Auditing a fund performance claim, a league table or a backtest for survivorship, backfill and cost treatment before any of it reaches a decision",
  "use": "Have an assistant turn a performance factsheet, scorecard or league table into a structured table fast, then generate the arithmetic checks you will run yourself: gross-to-net through the stated fee schedule, cumulative-to-annualised, the count of funds at the start of the window against the count at the end, and the benchmark's own definition (total return or price return, gross or net of withholding, same currency, same period-end). Ask separately for a list of every figure in the document that cannot be verified without data the document does not contain, because that list is usually where the claim lives.",
  "tools": ["Document AI grounded on the factsheet or scorecard you already hold", "Chat assistants with code execution for the recomputation and the beat-rate model", "The index provider's own methodology document and the fund's regulatory filings as the primary sources"],
  "prompt": "From this document, extract into a table: period, gross return, net return, benchmark return, number of funds in the universe at the start of the period, number at the end, and the stated fee schedule. Quote the survivorship and backfill treatment verbatim if the document states one, and write 'not stated' if it does not. Then list every arithmetic relationship among those columns that I can verify independently, and every figure I cannot verify without data this document does not contain.",
  "verify": "Do not accept the extracted table. Trace each row back to the document yourself, because document tools misread dense financial tables and a transposed row is invisible downstream. Recompute net from gross through the quoted fee schedule and the annualised figures from the cumulative ones by hand. Then apply the corrections you can compute: the closure rate times the closers' shortfall gives the survivorship bias in percentage points a year, and any beat rate quoted on survivors alone is inflated by roughly one over the survival rate. Where the document does not state its survivorship treatment, the correct output is 'unusable as stated', not a discounted number. Date-check the period against the claim being made, and cite the document and page for every figure that leaves this exercise.",
  "diy": "The gate is unaided. You state and apply the zero-sum identity, compute the fee-drag share of terminal wealth from the identity rather than from a calculator someone else wrote, derive why the beat rate falls with horizon from a constant cost against shrinking noise, and compute the survivorship and backfill corrections from a stated closure rate and shortfall. A tool can assemble the table; it cannot tell you that a 1.24-point database bias is the same size as the 1.25-point effect the study claims to measure, which is the finding that decides whether the table means anything.",
  "market": "US"
}

Modern Data Analysis

By hand first. You restated ten years of earnings into current money and averaged them. You solved a two-stage model for a discount rate by interpolation and checked that it reproduced the index level. You computed an exact real yield from a nominal one and an inflation expectation, and watched the sign of a spread flip. You ran a band across an eight-year path and priced what the discipline cost and what it saved.

All of that stays in your hands, because the gauges are small and the errors are silent.

What code adds is not the arithmetic. It is three things the page cannot give you: a series rather than a reading, a percentile computed against a real history rather than an illustrative one, and a refresh that happens on a schedule instead of when you remember.

Today's workflow. Three changes, and the first is a piece of algebra worth stating in closed form because the module performs it repeatedly.

The cyclically adjusted ratio is a price divided by a deflated mean, so its sensitivity to the deflator is exact and easy to bound. Write the ratio as P ÷ mean(Eₜ × CPI_now ÷ CPIₜ). Every restated earnings figure scales linearly with the current price index, so a misread current base moves every restated figure and the mean by the same factor and the ratio by its reciprocal. A 2 percent error in the current price index is a 2 percent error in the ratio, in the opposite direction, and it is invisible in the output because the series still looks smooth. Which is why the check that matters is not on the ratio but on two individual deflated cells recomputed by hand.

The implied-premium solve has a second closed-form guard. Since the Gordon one-liner sets r = cash yield + terminal growth, and the full two-stage solve differs from it only by the present value of the stage-one growth in excess of terminal growth, the two should never be far apart when the stage-one window is short. On the illustrative Indian inputs they differ by 0.30 points and on the United States inputs by 0.73. A solve that lands three points away from its own Gordon approximation has a bug, almost always in the terminal-value discounting exponent.

Build it as a panel: one row per market per quarter, with columns for each gauge, its percentile, the underlying inputs, and the retrieval date of each input. That structure is what turns five readings into a history you can compute a percentile against, and it is the only way the dashboard becomes an instrument rather than a snapshot. Two implementation points matter. Store the raw inputs, not the computed gauges, so a definition change can be re-run across the whole history. And store the retrieval date per cell rather than per row, because index earnings are revised for quarters after the fact and a panel that silently absorbs revisions will show you a history nobody could have acted on.

Tools and sources. pandas for the dated panel and the percentile computation, numpy for the solver, matplotlib for the gauge-against-its-own-history chart that makes a percentile legible. India: NSE Indices for index-level earnings and multiples, MOSPI for the consumer price index and national accounts, RBI's Handbook for the long series and the government yield curve, AMFI for fund expense ratios. United States: the Shiller data set for the long reconstruction, FRED for price indices and Treasury yields including the inflation-linked series, BEA for national accounts, S&P Dow Jones Indices for index earnings and buyback data. Every pull carries its date, and the date lives in the cell.

Verify. Prove the engine against the worked figures before it touches live data. The Indian build must reproduce a ten-year real mean of 812.4447, a ratio of 32.0022, a naive nominal ratio of 37.8733 and a trailing multiple of 23.2143. The United States build must reproduce a mean of 197.2377, a ratio of 34.4762 and a trailing multiple of 27.2000. The two-stage solver must return 8.7540 percent on the Indian inputs and 8.0304 percent on the United States inputs, each reproducing its index level to within 1e-6, with terminal shares of 92.44 percent and 85.22 percent. The real-yield function must return 4.4954 percent from a 13.90 percent nominal yield and 9.00 percent expected inflation, and −0.8789 percent from 1.50 percent and 2.40 percent. The fee-drag identity must return 8.8828 percent at 100 basis points, 8 percent gross and ten years. The band simulator must produce zero trades and a terminal ₹4,915,774.70 on Policy A and three trades and ₹4,600,998.34 on Policy B. Every one of those reproduces exactly. One convention to hold straight: the real yield uses the exact Fisher form (1+y)/(1+π) − 1 and not the subtraction y − π, and in the high-inflation case the two differ by 40 basis points, which is enough to move a percentile.

Quantitative lens

One quantitative fact governs everything in this file and it is the one most often dropped when the finding is repeated. A relationship with a coefficient of determination of 0.36 and a residual standard deviation of 3.90 percentage points is a prior, not a signal, and the arithmetic that proves it is one division.

Divide the residual standard deviation by the slope: 3.90 ÷ 0.32 = 12.19 ratio points. One standard deviation of unexplained variation in a ten-year outcome is worth more than the whole distance from a ratio of 20 to a ratio of 32. The instrument therefore cannot rank two markets that differ by less than about 12 ratio points. Most of the comparisons anybody wants to make are much closer than that.

The overlapping-window problem makes the estimate itself shakier than its published precision suggests. Across 1,740 months from 1881 to 2026 there are 1,620 overlapping ten-year forward windows and about 14 non-overlapping decades. Treating the overlapping windows as independent understates standard errors by roughly √(1,620 ÷ 14) = 10.76 times, which takes a naive t-statistic of 8.0 down to about 0.74. The correct treatment uses an autocorrelation-consistent estimator rather than this crude division and lands in the same neighbourhood. What the arithmetic establishes is not that the relationship is absent but that the effective sample supporting it is about a dozen observations, which is a small number to hang a strategy on.

Then the finding that decides how you use the dashboard. The gauges are not independent evidence, and treating them as though they are is the same error the survivorship correction catches in fund data. Four of the five absolute gauges share a numerator: the index price. When the price rises, the cyclically adjusted ratio rises, the trailing multiple rises, the capitalisation-to-output ratio rises and the implied premium falls, all at once, and none of those four is confirming the others. The composite's apparent agreement is mostly one variable wearing four hats. Which is why the range across the gauges is more informative than the mean, and why the fifth gauge, the only one whose reading can move against the other four because it carries the bond yield, is the one worth watching for genuine disagreement.

Now put the two halves together and price the whole enterprise. The maximum tilt the ladder permits is 15 percentage points of equity weight either side of neutral. Against a 4 percent equity risk premium that is worth 0.60 percentage points a year if the tilt is in the right direction, and it costs the same if it is not. The instrument driving the tilt has an effective sample of about a dozen decades and cannot distinguish markets closer than 12 ratio points. Against that, moving a core position from 150 basis points to 10 is worth 1.40 percentage points a year, requires no forecast, and cannot go the wrong way. The ratio is 2.33 to one in favour of the boring decision.

Honest limits, stated at length because a confident number does real damage here. The coefficients used above are illustrative and shaped to resemble the published relationship rather than measured from it; rebuild them from the Shiller data set before quoting any of them, and expect your own fit to differ. The residual standard deviation is itself estimated from the same small effective sample, so it carries an uncertainty the arithmetic above does not propagate. The fitted line assumes the relationship is stable across regimes, which is doubtful: accounting standards, the buyback channel, the tax treatment of dividends and the sectoral composition of the index have all changed enough that a nineteenth-century observation and a twenty-first-century one are not draws from the same process. The 4 percent equity risk premium used to price the tilt is an assumption, and at 2 percent the tilt is worth 0.30 points and at 6 percent it is worth 0.90, so the ranking against the fee decision survives across that whole range but the margin does not. The independence problem above is described qualitatively rather than modelled; a proper treatment would estimate the covariance among gauge percentiles and would almost certainly shrink the composite's information content further. And none of the above says the market's level is unknowable. The level is estimable and the timing is not, which is the whole distinction between a policy and a call.

Where this goes next: galaxy cross-links

• AA1.04 Capital Market Expectations: the institutional version of everything here, where the gauges become forward-looking inputs to an allocation process with a formal framework and a documented set of assumptions.
• AA1.07 Active Management and the Fundamental Law: the information-ratio arithmetic behind section 4.13's required-alpha table, and the formal treatment of why breadth and skill trade off.
• AA1.08 Index Construction, ETFs and Equity Implementation: what an index actually is, how it is maintained, and the tracking-difference and securities-lending details that separate a 5-basis-point vehicle from a 25-basis-point one.
• QM1.02 Hypothesis Testing and Regression: the standard-error machinery behind section 4.3's overlapping-window correction, including the autocorrelation-consistent estimators that do the job properly.
• QM1.04 Returns, Statistical Measures and the Discipline of Inference: geometric against arithmetic means, the harmonic mean that appears in both the index multiple and the systematic-plan average cost, and the inference limits on small effective samples.
• E11.06 Event-Driven Special Situations and the Canon's Unclosed Drills: where capital that passes the named-edge test actually goes, and the structural account of why the mispricings persist.

Teach it back & journal

Feynman prompt. Explain to somebody who has just been told that the market is expensive why that sentence is not yet a claim about anything. Use the sentence "a valuation is a price divided by an earning power, and both halves need a definition before the quotient means anything" and then build one gauge in front of them, out loud, from the ten-year table. Make them do the deflator arithmetic on two rows themselves, because that is where every silent error in this work lives. Then show them the same series averaged without the deflator, watch the ratio move from 32.00 to 37.87, and make them say what changed. The answer they have to reach is that nothing about the market did.

Then hand them the harder half. Give them the fitted line and the residual, and ask what they would do differently if the ten-year forecast were 4.76 percent rather than 9 percent. If they answer with a trade, you have not finished. Keep going until the answer is about a savings rate, a withdrawal rate, a retirement date or an allocation inside a band, because those are the decisions a weak ten-year relationship can actually inform. When they ask, unprompted, "what is the band around that", you have installed the instrument rather than the number.

Close on the ranking. Put the 0.60 percentage points a year that a perfect valuation call is worth beside the 1.40 that a cost decision is worth, and let them sit with the fact that the boring one is bigger and certain. Most people find this genuinely unwelcome, and the discomfort is the sign it landed.

Journal prompt. For one month, every time you meet a statement that a market is cheap or expensive, in a published note, a headline, a podcast or your own head, write down four things before you agree or disagree. Which gauge, if any, is behind it. What its reading is and against what history. What earnings definition and what date. And what the person saying it would do differently if they were wrong.

Then answer in writing at the end of the month. Count the ones that came with a number attached, and the ones that named the history the number was being read against. Count separately the relative comparisons against bonds that were wearing the clothes of an absolute claim. Where you could not find the underlying gauge at all, write down what you did with the statement and what you should have done.

Then write a page on the part that is about you rather than about markets. Your policy will be tested twice: once when the gauges read expensive and prices keep rising for three more years, and once when they read cheap and your portfolio is down 40 percent. Write down now, in advance, what you expect to want to do in each of those two situations, in your own words, with the reasoning you expect to find persuasive at the time. Then write the sentence in your policy that will stop you. Date it. Put it where you will find it, not where you will file it. The floor in your allocation band and the thirty-day amendment rule are the only two devices in this entire elective that are aimed at that moment, and they are the only two whose value you will be in no condition to reassess when it arrives.