Capital Structure: Theory, Policy and the Financing Decision

Branch: Economics & Corporate Completions (Finance-Plus Ring) · Node EC1.06 · ~18 focused hours · Standard mastery gate ≥ 85%

Two forging companies run the same two plants, sell to the same three axle makers, and earn the same ₹360 crore of operating profit. One owes its lenders ₹144 crore. The other owes ₹1,440 crore. On the arithmetic worked out later in these pages, the second is worth about ₹97.92 crore more than the first, which is 4.83% on a base of roughly ₹2,029 crore.

The gross tax shield on the extra ₹1,296 crore of debt is worth ₹326.20 crore. The extra expected cost of financial distress swallows ₹228.28 crore of it, or 69.98%. What survives is the residue of two large numbers fighting, and neither is observable. You will build both, from a rating table, a spread, a probability schedule you choose and defend, and a distress-cost fraction that the published research brackets rather than pins.

The financing question in its exact form is this. A business generates a stream of operating cash flows, and that stream can be sold to owners, to lenders, or to hybrid claimants in any proportion at all. Does the proportion change the total? Modigliani and Miller's answer in 1958 was no, in a world with no taxes, no bankruptcy, no information asymmetry and no agency conflict, and their proof is an arbitrage argument you can carry out with a pencil. Every real theory of capital structure since is an argument about which of those four assumptions to relax first, and about how much the total moves once you do.

By the end you can prove the irrelevance result by constructing the arbitrage, price the tax shield and reconcile it to the beta arithmetic the valuation phase already uses, build a trade-off curve numerically and locate its optimum, show that the curve is nearly flat for a full turn of leverage either side of that optimum, put numbers on the two agency conflicts that debt creates, compute the point at which debt beats equity on earnings per share and say why that point is the wrong criterion, and read a financing announcement as the information event it is.

One standing caution, because it applies to everything below. The optimum you locate is a statement about your assumptions, not a measurement of the firm. Move the distress-cost fraction from 40% of unlevered value to 25% and the optimum in the worked example jumps a full 0.6 turns of EBITDA. Widen every credit spread in the table by 50 basis points and a near-tie between two leverage points becomes a ₹50.43 crore gap. The corpus's habit of quoting a number with its provenance attached matters more here than almost anywhere, because the output looks like a measurement and is not one.

This Ring is additive institutional and CFA breadth. It presents the toolkit and the plumbing alongside, not in place of, the program's process-over-P&L, concentration-over-diversification philosophy.

Learning objectives

By the end you can:

1. Prove Modigliani and Miller's Proposition I without taxes by constructing the homemade-leverage arbitrage in both directions, and state the four assumptions the proof consumes.
2. Compute the levered cost of equity and the weighted average cost of capital at three or more leverage levels under MM Proposition II without taxes, and show that the cost of equity rises exactly fast enough to leave the weighted average flat.
3. Price the corporate tax shield, compute the levered value, the levered cost of equity and the weighted average cost of capital with taxes, and verify each by an independent route.
4. Reconcile the levered cost of equity to the Hamada relation, carry a non-zero debt beta through it, and quantify in basis points the error that assuming a zero debt beta introduces.
5. Work the Miller equilibrium in both tax regimes, compute the net gain per unit of debt after personal taxes, find the corporate rate at which that gain vanishes, and say why the sign of the answer is not robust to its inputs.
6. Build a static trade-off curve numerically: solve the coverage-to-spread fixed point, apply a stated distress-probability schedule and distress-cost fraction, locate the interior optimum, and quantify how flat the curve is near it.
7. Quantify the agency cost of equity using the free-cash-flow argument, and the two agency costs of debt (asset substitution and debt overhang) with payoff tables that show who gains, who pays, and how much simply evaporates.
8. State the pecking order's prediction for an equity issue, convert the announcement effect into money on a stated market capitalisation, and reconcile the ordering of financing sources with the ordering of capital uses without contradicting either.
9. Solve the EBIT-EPS indifference point algebraically, tabulate earnings per share under a debt plan and an equity plan across a range of operating profit, and demonstrate that maximising earnings per share and maximising value per share are different objectives that disagree.
10. Read a financing decision as a signal: name the average announcement effect of an equity issue, a debt issue, a buyback and a dividend initiation, size each in money, and say what an analyst may and may not infer from one.

Prerequisites & connections

Builds on. M3.03 owns the cost of capital: the risk-free rate, the bottom-up beta, the Hamada unlever-and-relever recipe, the synthetic-rating table that maps interest coverage to a credit spread, and the marginal-tax convention (India about 25.17%, the United States about 25% once state tax is added). This node borrows all of it, states nothing about beta estimation, and adds only the reason the Hamada relation is true and the correction it needs when the debt beta is not zero. M3.07 owns capital allocation: the conservation-of-value principle, the five uses of capital, and the proof that a buyback creates value only below intrinsic value. M2.02 owns the coverage and solvency ratios that feed the rating lookup. EC1.02 owns dividend policy, the Modigliani and Miller 1961 dividend-irrelevance baseline, share classes and index construction. E11.01 owns the credit lens: rating methodology, covenants, the recovery waterfall under Section 53 of the Insolvency and Bankruptcy Code, and the spread-equals-probability-times-loss slide rule. M4.01 owns the degree of operating leverage, the degree of financial leverage and their product.

Deliberate non-overlaps. Do not rebuild any of the above. The synthetic-rating table is used here as a lookup and is not re-derived; where a spread appears, M3.03's table produced it. The value arithmetic of a buyback belongs to M3.07 and is cited rather than repeated; what is added here is the leverage side of the same transaction, meaning the tax shield the borrowing creates and the distress cost it imports. Covenant drafting and the insolvency waterfall belong to E11.01; what is added here is the payoff table that shows why a covenant is worth paying for. Dividend policy belongs to EC1.02; what is added here is dividends read as a signal about private information, alongside the other three financing announcements. M4.01 owns the degree of financial leverage as a sensitivity measure; the EBIT-EPS indifference point taught here is the decision rule that sits on top of it.

Feeds forward. EC1.07 takes the agency conflicts named here and asks what governance machinery contains them. M3.09's modelling track consumes the speed-of-adjustment result directly, because a three-statement model that snaps the capital structure to target in year one omits interest the company will actually pay. E11.02's leveraged-buyout and special-situation work is this arithmetic run at the far right of the leverage grid. AA1.06's factor work meets it again from the other side, because leverage is one of the exposures a multifactor model prices.

Core teaching content

5. Personal taxes and the Miller equilibrium

Merton Miller's 1977 argument is that the corporate shield is not the whole tax story, because the investor who receives the interest pays tax on it too, usually at a higher rate than on equity income.

The net gain from leverage, per unit of permanent debt, is

G_L / D = 1 − (1 − t_c)(1 − t_ps) / (1 − t_pd)

t_c = corporate tax rate
t_ps = the investor's effective tax rate on equity income (dividends and capital gains blended)
t_pd = the investor's tax rate on interest income

The corporate-shield-only result is the special case where personal rates are equal, which collapses the expression to t_c.

India after the Finance Act 2020 is the cleaner illustration, because the abolition of the dividend distribution tax moved dividends into the shareholder's own hands and made the personal layer visible. Take a top-bracket investor whose interest income is taxed at an effective 35.88%, and model equity income as 30% dividends at that same rate and 70% long-term capital gain at 12.5%, giving a blended equity rate of 19.514%. At a corporate rate of 25.17%:

G_L / D = 1 − (0.7483 × 0.80486) / 0.6412 = 0.060704

Just over six paise per rupee of debt, against the 25.17 paise the corporate shield alone advertised. The personal layer erased 75.88% of it. The corporate rate at which the gain disappears entirely is 1 − 0.6412/0.80486 = 20.334%, and India sits above that line, which is why the gain is positive but small.

The United States gives the more startling answer. At a top ordinary rate of 37% plus the 3.8% net investment income tax, interest is taxed at 40.8%, while qualified dividends and long-term gains face 20% plus the same 3.8%, so 23.8%. Using the federal corporate rate of 21%:

G_L / D = 1 − (0.79 × 0.762) / 0.592 = −0.016858

Negative. For that investor, at that corporate rate, permanent debt destroys about 1.69 cents of value per dollar rather than creating 21 cents. Add typical state corporate tax to reach the 25% marginal rate the valuation phase uses and the sign flips back to a positive 3.46 cents. The break-even corporate rate is 22.3097%, and the two rates tested here sit on opposite sides of it.

Take the honest lesson rather than the number. The Miller gain is a small difference between large quantities, so its sign is not robust to any of its inputs. Apply a deferral haircut to the capital-gains half of the equity return, which is what a buy-and-hold investor actually experiences, and the United States figure at a 25% corporate rate falls from positive 3.46 cents to negative 2.87 cents at a haircut factor of 0.7 and negative 7.09 cents at 0.5. Anyone who tells you the tax gain to leverage is 25 paise per rupee has priced the corporate layer and ignored the investor. Anyone who tells you it is zero has picked one plausible set of personal rates out of many.

11. Target versus actual, and the speed of adjustment

Firms behave as though they have a target capital structure and as though moving to it is expensive. Both halves matter.

The standard description is the partial-adjustment model: D_t − D_{t−1} = λ (D* − D_{t−1}), where D* is the target and λ is the fraction of the gap closed each year. Estimates of λ for United States firms have ranged widely across specifications, with Flannery and Rangan (2006) finding roughly a third of the gap closed annually and other studies finding substantially slower adjustment. The half-life follows directly as ln(0.5) / ln(1 − λ): 1.94 years at λ = 0.30, 3.11 years at λ = 0.20, and 6.58 years at λ = 0.10.

Adjustment is slow because it is costly. Issuing equity carries the announcement effect priced above. Repaying debt early triggers make-whole payments. Issuing debt into a weak market is expensive and sometimes impossible. Meanwhile the ratio drifts on its own, because a share price that doubles halves the debt-to-value ratio without anyone deciding anything.

For an analyst the practical consequences are two, and the second is where models actually break. A deviation from target is information: a firm well above target after an acquisition is telling you the next three years of free cash flow are spoken for, which constrains the dividend, the capital expenditure and the next deal. And a three-statement model that snaps the capital structure back to target in the first forecast year invents earnings. Sahyadri Freight Systems Ltd (synthetic) sits at 52% debt to value against a 35% target on a ₹5,000 crore enterprise value, borrowing at 8%. At λ = 0.25 the ratio traces 47.75%, 44.56%, 42.17%, 40.38% and 39.03% over five years, still 4.03 percentage points above target at the end. The interest a snap-to-target model omits totals ₹155.589844 crore over those five years, which is ₹116.427880 crore of after-tax earnings the model has invented from nothing.

12. Leverage, earnings per share, and the indifference point

Debt raises earnings per share whenever the operating profit is high enough, and the threshold is exact.

Compare two ways of raising the same money. Under the equity plan, share count rises by ΔN and interest does not change. Under the debt plan, share count is unchanged and interest rises by I. Setting the two earnings-per-share expressions equal, the after-tax factor cancels, and the indifference point is

EBIT* = I × (N + ΔN) / ΔN

Above EBIT* the debt plan produces higher earnings per share, below it the equity plan does, and at the point itself they are identical. The line for the debt plan is steeper because the same profit is spread over fewer shares, and it starts lower because of the interest, so the two lines cross exactly once.

Two things the indifference point does not tell you, both of which people read into it anyway. It says nothing about value, because it holds the discount rate fixed while the whole point of a leverage change is that the discount rate moves. And it says nothing about the probability of being on the good side of the line. Winslow Instrument Corp's indifference point sits at $143.75 million of operating profit; if operating profit is normally distributed with a mean of $170 million and a standard deviation of $45 million, the debt plan wins 72.0166% of the time and loses 27.9834% of the time. The 27.9834% is not a random 28% of years. It is the recession, and it is correlated with the covenant test, the refinancing window and the customer who stops paying.

Use the indifference point as a stress threshold. Compute it, then ask how far below current operating profit it sits and what would have to happen to get there. A firm 40% clear of it is comfortable. One whose indifference point sits above last year's actual operating profit is running an earnings-per-share argument it cannot support.

13. Recapitalisation arithmetic, and why accretion is not the test

A debt-funded buyback does two things at once, and they have to be scored separately.

It creates a tax shield, worth t × D on permanent debt. And it buys shares at a price, which transfers value between the holders who sell and the holders who stay, in a direction set entirely by price against intrinsic value. The valuation phase proves the second half. What the financing lens adds is that the first half funds a specific, computable amount of overpayment and no more.

Wexford Media Corp (synthetic) has an intrinsic equity value of $2,000 million across 100 million shares, so $20.00 a share, and trades at $28.00. It borrows $560 million at 6% and buys stock at the market price. The shield created is 0.25 × 560 = $140 million. Shares retired are 560/28 = 20 million. Intrinsic equity afterwards is 2,000 + 140 − 560 = $1,580 million across 80 million shares, or $19.750000 a share. The continuing holder is 25 cents, or 1.25%, worse off.

Earnings per share went the other way. Interest of $33.6 million takes net income from $150.0 million to $124.8 million, but the share count fell by a fifth, so earnings per share rise from $1.500000 to $1.560000, an accretion of 4.00%. Return on equity does better still, from 15.000000% to 28.363636%, because the buyback removed $560 million of book equity from the denominator. Three metrics improved and the owner is poorer.

The break-even price is where the shield exactly pays for the overpayment. Each dollar spent at price P buys 20/P dollars of value and t dollars of shield, so the recap is neutral when (P − 20)/P = 0.25, giving P = $26.666667. The tax shield funds a 33.333333% premium to intrinsic value, and not one cent more. Verify it the long way and the answer is the same: at $26.666667 the firm retires 21 million shares, leaving 79 million holding $1,580 million, which is exactly $20.000000 a share.

Run the identical recapitalisation at the intrinsic $20.00 and 28 million shares retire, leaving 72 million holding $1,580 million, or $21.944444 a share, a gain of 9.722222%. Same leverage, same shield, same interest expense, same accretion arithmetic, opposite verdict. The only variable that moved was the price.

Common mistakes & how experts think differently

1. Calling debt cheap because its rate is lower. The question is settled by the weighted average before against the weighted average after, and setting the cost of debt beside the cost of equity never reaches it. Brandywine's cost of equity climbs from 12% to 21% across the no-tax table while the weighted average never leaves 12%. An expert quotes the blend and names what moved it.
2. Using book-value weights. The MM arithmetic is defined on market values throughout, and a book weight will reliably make a levered firm look more levered than it is once the equity has appreciated. The check is that levered value, equity plus debt at market, must equal after-tax operating profit divided by the weighted average cost of capital. If those two disagree, look at the weights first.
3. Relevering beta with a zero debt beta on a high-yield issuer. The Hamada relation assumes the debt carries no market risk. Carrying an implied debt beta of 0.4 through the arithmetic cut Brandywine's relevered cost of equity by 0.857143 percentage points at a debt-to-equity ratio of 0.571429 and by 3.000000 percentage points at 2.0. On a stressed issuer the error runs to hundreds of basis points and always in the direction of undervaluing the equity.
4. Quoting the tax shield as the tax gain. The corporate shield is the gross number. The Miller expression nets the investor's own tax position against it, and on plausible top-bracket inputs it removed 75.88% of the shield in India and reversed its sign in the United States at the federal corporate rate. Quote the corporate shield and the personal layer as a pair, or say which one you are quoting.
5. Reading the trade-off optimum as a target. The curve is flat: 0.1163% separates the peak from a point 0.6 turns away, and a 15-point change in the distress-cost assumption moves the peak itself by that much. An expert reports the plateau rather than the peak, and treats a company anywhere inside it as having no capital-structure problem.
6. Treating the coverage-to-spread lookup as a single-valued function. It is a step function with a circular input, so more than one rating can be self-consistent at high leverage. Chandori at three turns supports both B+ at a 3.00% spread and B at 3.70%, depending on which end you iterate from. State the convention. An unstated convention is an assumption you cannot audit later.
7. Confusing the pecking order of sources with the ranking of uses. They are orthogonal, and the apparent contradiction is a category error. The allocation ranking asks what a rupee should be spent on. The pecking order asks where the rupee should come from. A firm applies both, and only has to choose when a project clears the hurdle and equity is the only source left.
8. Selecting on earnings per share. Chandori's earnings per share rise 58.6198% from 1.2 turns to 3.0 turns and its return on equity rises 34.5708 percentage points, while value per share peaks at 1.8 turns and is ₹2.888011 lower at 3.0 turns, which is ₹57.760213 crore of value on 20 crore shares. Accretion is an arithmetic statement about swapping one funding source for another and carries no information about value.
9. Ignoring the agency conflicts because the firm is currently healthy. Both conflicts are options on default, so they price at nearly zero for an investment-grade issuer and dominate everything once default is live. The analytical move is to ask what the covenant package would be worth if leverage doubled, not what it is worth today.

The thread through all nine is that a capital-structure number is the joint product of an observed price and a model you built. The market gave you the share price, the credit spread and the rating. The optimum, the shield, the expected distress cost, the announcement effect and the implied probability were manufactured, from assumptions you chose. Say which is which, on the same page as the number.

Worked examples

Every figure below was recomputed in Python before it was written down. All entities are explicitly synthetic. All curves and schedules are illustrative but internally consistent, and every perishable input carries the source that settles it.

Worked example 1: The trade-off curve and the recapitalisation (India, ₹)

Chandori Forgings Ltd (synthetic) makes crankshaft and axle forgings from two plants in Maharashtra. It earns ₹480 crore of EBITDA, ₹120 crore of depreciation and amortisation, and therefore ₹360 crore of operating profit. Its marginal tax rate is 25.17%, being the 22% rate under section 115BAA of the Income-tax Act plus a 10% surcharge and a 4% cess (check against the Act before real use; checked 2026-09-07). Its unlevered cost of equity is 13.5%. It has 20 crore shares and ₹1,600 crore of book equity, and net debt of ₹144 crore, which is 0.3 turns of EBITDA. The board is considering a move to 1.2 turns.

The INR benchmark rate is taken as 6.50%, which is the valuation phase's 4.50% clean rupee rate plus its 2.00% India country default spread. Credit spreads come from the small-firm column of the synthetic-rating table (illustrative early-2026 vintage; the live sheet is refreshed each January and must be pulled before real use). The distress-probability schedule below is stated, not measured, and is shaped on published cumulative default rates by rating. The cost of distress if it occurs is taken as 40% of unlevered value, direct plus indirect.

Solution. Step 1, the unlevered value. V_U = 360 × (1 − 0.2517) / 0.135 = 269.388 / 0.135 = ₹1,995.466667 crore. The distress cost if it happens is 0.40 × 1,995.466667 = ₹798.186667 crore.

Step 2, solve the coverage-to-spread fixed point at each leverage level. At 3.0 turns, debt is ₹1,440 crore. Start at the narrowest spread of 0.60%, so r_D = 7.10% and coverage is 360 / (1,440 × 0.071) = 3.521, which is a BB+ band carrying 2.00%. Retry at r_D = 8.50%: coverage is 2.941, a B+ band carrying 3.00%. Retry at r_D = 9.50%: coverage is 2.632, still B+. Stable. Iterating upward from the widest spread instead lands on B at 3.70% with coverage 2.451, which is also self-consistent; the convention here is to iterate downward from the best rating and take the first stable band.

Step 3, the grid.

Distress probabilities used: AAA 0.4%, A+ 1.4%, A- 3.6%, BB+ 13.0%, B+ 29.0%, and BB 19.0%, which the widened-spread run at the end of this example needs. Value per share is (V_L − 144) / 20, the price a continuing holder ends up with once the announcement is fully anticipated and the buyback executes at that price.

Step 4, read the optimum and the plateau. The maximum is ₹2,184.2007 crore at 1.8 turns. The 2.4-turn point is ₹2.539947 crore below it, which is 0.1163%. The board's 1.2-turn target is ₹54.929493 crore below, which is 2.5149%. Across the full grid, total value moves 7.1276%.

Step 5, the earnings-per-share and return-on-equity table at three points. Each row assumes the new debt funds a buyback executed at the anticipated price.

Step 6, verify three ways. Value per share computed as (V_L − D) / N after the buyback reproduces the buyback price exactly at every column, for example (2,184.2007 − 864) / 12.941871 = 102.010037. The weighted average cost of capital is NOPAT / V_L, so its minimum must coincide with the value maximum, and it does at 1.8 turns. And the value an earnings-per-share maximiser gives up, 102.010037 − 99.122027 = ₹2.888011 on 20 crore shares, is ₹57.760213 crore, which equals 2,184.2007 − 2,126.4405 to six decimals.

Read it. An earnings-per-share maximiser picks 3.0 turns, where earnings per share are 58.6198% higher than at 1.2 turns and return on equity is 34.5708 percentage points higher, and hands ₹57.76 crore of value to the shareholders who sold. The honest counterweight is that ₹2.89 a share is 2.83% of the price, and the model's own error bars are wider than that. Move the distress-cost fraction to 25% and the optimum moves to 2.4 turns; move it to 55% and the optimum holds at 1.8 turns with the 2.4-turn point ₹30.6760 crore behind. Widen every spread by 50 basis points and the 2.4-turn row falls one notch to BB, on a 2.90% widened spread and a 19.0% distress probability, so the optimum stays at 1.8 turns while the near-tie becomes a ₹50.4311 crore gap.

Worked example 2: The financing choice and the announcement cost (US, $)

Winslow Instrument Corp (synthetic) has 40 million shares trading at $50.00, no debt, and needs to raise $300 million for a capacity expansion. Under the equity plan it issues shares at $50.00. Under the debt plan it issues ten-year notes at 6.25%. The marginal tax rate is 25%.

Solution. Step 1, the two plans. The equity plan adds 300/50 = 6.000000 million shares for a total of 46.000000 million. The debt plan adds 300 × 0.0625 = $18.750000 million of annual interest and leaves the share count at 40.000000 million.

Step 2, the indifference point. EBIT* = 18.75 × 46/6 = $143.750000 million. Check it directly: 40 × 143.75 = 5,750 and 46 × (143.75 − 18.75) = 5,750, a residual of zero.

Step 3, earnings per share across a range.

Step 4, how often the debt plan wins. If operating profit is normally distributed with a mean of $170 million and a standard deviation of $45 million, the indifference point sits at z = (143.75 − 170)/45 = −0.583333, so the debt plan wins on earnings per share 72.0166% of the time and loses 27.9834% of the time. Normality is doing real work in that figure, and operating profit in a cyclical business has a fatter left tail than a normal curve allows, so read 72.0166% as the optimistic end of the range.

Step 5, the announcement effect the pecking order predicts. Winslow's pre-announcement market value is 40 × 50 = $2,000 million. At the roughly negative 3% average abnormal return reported for United States seasoned equity offerings, the equity plan costs $60.000000 million on announcement, which is 20.00 cents on every dollar raised. The debt plan's announcement effect is approximately zero.

Step 6, put the two costs beside each other. The debt plan's tax shield on permanent debt is 0.25 × 300 = $75.000000 million. If the extra leverage moves Winslow to a rating carrying a 2.5% probability of distress and distress costs 40% of the $2,000 million unlevered value, the expected distress cost is $20.000000 million. Net of both, and adding back the $60 million announcement cost the debt plan avoids, the debt plan is ahead by $115.000000 million. The distress probability that would erase that advantage is 16.875%, which is a deep-junk rating.

Read it. The comparison people run is the earnings-per-share table in Step 3, and it is the least important of the six steps. The two numbers that decide the question, the $60 million announcement cost and the $75 million shield, are both invisible in it. And the 27.9834% of outcomes where the debt plan loses on earnings per share are the same outcomes in which the covenant is tested and the refinancing window is shut, so they are worth more than their probability.

Worked example 3: Proposition II reconciled to the beta arithmetic (US, $)

Brandywine Valve Corp (synthetic) earns a perpetual $100 million of operating profit. Its unlevered cost of equity is 12%, it borrows at 6%, the risk-free rate is 4% and the equity risk premium is 5%. Show Proposition II with and without taxes at three leverage levels, and reconcile the levered cost of equity to the relevering formula the valuation phase uses.

Solution. Step 1, the implied betas. β_U = (12 − 4)/5 = 1.600000 and β_D = (6 − 4)/5 = 0.400000. The debt beta is not zero, which is what Step 4 turns on.

Step 2, no taxes. V_U = 100/0.12 = $833.333333 million at every leverage level. At D = 500, equity is $333.333333 million, D/E = 1.500000, and r_E = 0.12 + 0.06 × 1.5 = 21.000000%. Net income of 100 − 30 = $70.000000 million over $333.333333 million gives the same 21.000000%. The weighted average, at E/V = 333.333333/833.333333 = 0.4 and D/V = 0.6, is 0.4 × 0.21 + 0.6 × 0.06 = 12.000000%, unchanged.

Step 3, with taxes at 25%. V_U = 75/0.12 = $625.000000 million. At D = 500, V_L = 625 + 0.25 × 500 = $750.000000 million, equity is $250.000000 million, and D/E = 2.000000. Then r_E = 0.12 + 0.06 × 0.75 × 2 = 21.000000% and the weighted average is (1/3) × 0.21 + (2/3) × 0.06 × 0.75 = 10.000000%. Two checks: r_0 (1 − tD/V_L) = 0.12 × (1 − 0.166667) = 10.000000%, and 75 / 0.10 = 750.000000.

Step 4, the relevering reconciliation.

The second column of costs of equity is exactly what Proposition II produced in Step 3, which is the independent route. The overstatement has a closed form, β_D (1 − t)(D/E) × ERP, which returns 0.857143 and 3.000000 percentage points on the two rows.

Read it. The Hamada relever is a special case, not a law, and its special case is that lenders bear no market risk. That is nearly true of an investment-grade issuer and badly false of anything trading at a wide spread. The correction is one extra term and it always cuts the cost of equity, which means the standard recipe systematically undervalues levered equity, most where leverage is highest.

Worked example 4: Asset substitution, priced (US, $)

Braddock Chemicals Inc (synthetic) has $100 million of debt maturing in one year and one investment decision left before then. The safe project delivers $115 million or $125 million with equal probability. The risky project delivers $40 million or $190 million with equal probability. Assume risk neutrality and a zero discount rate so the arithmetic is visible.

Solution. Step 1, the payoff table. Debt receives min(assets, 100) and equity receives max(assets − 100, 0).

Step 2, the identity. The lenders lose $30.000000 million. Of that, $25.000000 million reaches the owners, which is 83.3333%, and $5.000000 million evaporates, which is 16.6667%. Equity gain plus value destroyed equals debt loss, exactly.

Step 3, who actually pays. Lenders who can foresee the switch will not lend $100 million against a $100 million face value. They will lend $70.000000 million, an implied yield of 100/70 − 1 = 42.857143%. A covenant that credibly rules the switch out restores the debt to $100.000000 million, so it is worth $30.000000 million to the borrower.

Read it. Equity in a levered firm is a call option, and option holders are paid for volatility, so an owner facing a wall of debt has a rational reason to prefer a worse project. The conflict costs the borrower rather than the lender, because the lender prices it in advance. That is also why the conflict vanishes for a firm nowhere near default: with $100 million of debt against assets that cannot fall below $115 million, the option is deep in the money and its volatility sensitivity is close to zero.

Worked example 5: Debt overhang and three ways out (India, ₹)

Deccan Roadways Ltd (synthetic) operates a fleet of 900 trucks and owes ₹800 crore, maturing in a year. Its assets will be worth ₹500 crore or ₹1,100 crore with equal probability. A depot automation project costs ₹100 crore of fresh equity today and adds ₹150 crore to asset value in both states.

Solution. Step 1, without the project. Firm value is ₹800.000000 crore, debt is worth (500 + 800)/2 = ₹650.000000 crore, and equity is worth (0 + 300)/2 = ₹150.000000 crore.

Step 2, with the project. Assets become ₹650 crore or ₹1,250 crore. Firm value is ₹950.000000 crore, debt is worth (650 + 800)/2 = ₹725.000000 crore, and equity is worth (0 + 450)/2 = ₹225.000000 crore.

Step 3, who gets what. The project's net present value to the firm is 150 − 100 = ₹50.000000 crore. The owners' equity rose ₹75.000000 crore and they must fund ₹100.000000 crore, leaving them ₹25.000000 crore worse off. The lenders gained ₹75.000000 crore for nothing, which is 50.0000% of the value the project creates. A rational owner declines a positive-value project.

Step 4, the write-down that unlocks it. Reduce the face value to F. For 500 < F ≤ 650, the owners' gain from the project is (800 − F)/2, which reaches ₹100 crore at F = ₹600.000000 crore. That works, but it costs the lenders: their claim falls from ₹650.000000 crore to ₹600.000000 crore. They give up ₹50.000000 crore and the owners end ₹100.000000 crore ahead, being the ₹50.000000 crore of net present value plus ₹50.000000 crore transferred out of the lenders' claim.

Step 5, the new-money route. Let the lenders fund the ₹100 crore themselves with a claim senior to the existing debt. Assets are ₹650 crore or ₹1,250 crore, so the new money is repaid in full in both states. The residual after it is ₹550 crore or ₹1,150 crore, so the old debt is worth (550 + 800)/2 = ₹675.000000 crore and the equity is worth (0 + 350)/2 = ₹175.000000 crore. Total is 100 + 675 + 175 = ₹950.000000 crore, which ties. The old lenders are ₹25.000000 crore better off and the owners are ₹25.000000 crore better off, and the two gains sum to the ₹50.000000 crore of net present value.

Read it. Overhang is the correct decision for the owners given the claim structure, which is why exhortation never fixes it and a change in the claim structure always does. Of the three routes, the concession costs the lenders ₹50 crore, the coordination pays them ₹25 crore, and a ring-fence achieves the same thing through a separate vehicle where the law permits it. Priority new money is the mechanism behind debtor-in-possession financing and behind the interim finance provisions of India's insolvency framework.

Worked example 6: The Miller equilibrium in two regimes (India ₹ and US $)

Compute the net gain from permanent debt after personal taxes for a top-bracket investor in each market, and find the corporate rate at which the gain vanishes.

Solution. Step 1, India. Interest is taxed at an effective 35.88%, being a 30% slab rate with a 15% surcharge and a 4% cess (a top-bracket illustration; surcharge and cess rates are set annually by the Finance Act, check before use; checked 2026-09-07). Equity income is modelled as 30% dividends, taxed at the same 35.88% since the Finance Act 2020 moved dividends into the shareholder's hands, and 70% long-term capital gain at 12.5% under the Finance (No. 2) Act 2024. The blended equity rate is 0.30 × 0.3588 + 0.70 × 0.125 = 0.195140. At a corporate rate of 25.17%:

G_L / D = 1 − (1 − 0.2517)(1 − 0.195140)/(1 − 0.3588) = 1 − 0.602277/0.641200 = 0.060704

Step 2, the United States. Interest faces 37% plus the 3.8% net investment income tax, so 40.8%. Qualified dividends and long-term gains face 20% plus the same 3.8%, so 23.8%, and no deferral haircut is applied in the base case. At the 21% federal corporate rate:

G_L / D = 1 − (0.79 × 0.762)/0.592 = 1 − 1.016858 = −0.016858

Step 3, both regimes side by side.

Step 4, size it. On $300 million of permanent debt, the United States figures are negative $5.057432 million at a 21% corporate rate and positive $10.388514 million at 25%, against the $75.000000 million the corporate shield alone would have claimed.

Read it. The net gain from debt once personal taxes are counted is a small residue of large offsetting quantities, so its sign is not robust to any input. Apply a deferral factor of 0.7 to the capital-gains half of the equity return and the United States figure at a 25% corporate rate goes from positive 3.4628 cents to negative 2.8691 cents. Two working conclusions survive that fragility. The corporate shield overstates the tax case for debt for a taxable individual investor, everywhere. And the shield is close to its full stated value for a tax-exempt investor such as a pension fund or a charitable endowment, which is one reason clienteles for high-yield debt look the way they do.

Worked example 7: A debt-funded buyback above intrinsic value (US, $)

Wexford Media Corp (synthetic) has an intrinsic equity value of $2,000 million across 100 million shares. The stock trades at $28.00. Net income is $150 million, book equity is $1,000 million, and the tax rate is 25%. The board borrows $560 million at 6.0% and repurchases stock at the market price.

Solution. Step 1, the two effects. The shield created is 0.25 × 560 = $140.000000 million. Shares retired are 560/28 = 20.000000 million, leaving 80.000000 million.

Step 2, intrinsic value per share afterwards. Intrinsic equity is 2,000 + 140 − 560 = $1,580.000000 million, so 1,580/80 = $19.750000 a share against $20.000000 before, a fall of $0.250000, or 1.2500%.

Step 3, trace the money. Sellers received $28.00 for $20.00 of value, twenty million times, so $160.000000 million left the continuing holders. The shield returned $140.000000 million. The net leak of $20.000000 million spread over the surviving 80 million shares is exactly $0.250000 a share, which reconciles Step 2 by an independent route.

Step 4, what the reported metrics did. Interest is 560 × 0.06 = $33.600000 million, so net income falls from $150.000000 million to $124.800000 million. Earnings per share rise from $1.500000 to $1.560000, an accretion of 4.0000%. Return on equity rises from 15.000000% to 28.363636%, a gain of 13.3636 percentage points, because the buyback removed $560 million of book equity from the denominator.

Step 5, the break-even price. Each dollar spent at price P acquires 20/P of value and 0.25 of shield, so the recap is value-neutral when (P − 20)/P = 0.25, giving P = $26.666667, a premium of 33.3333% to intrinsic value. Verify the long way: at that price 21.000000 million shares retire, leaving 79.000000 million holding $1,580 million, which is $20.000000 a share.

Step 6, the same recap at intrinsic value. At $20.00 the same $560 million retires 28.000000 million shares, leaving 72.000000 million holding $1,580 million, or $21.944444 a share, a gain of 9.7222%.

Read it. The tax shield is real and it buys a specific, computable amount of overpayment room, which is 33.3333% at a 25% tax rate. Beyond that price the recapitalisation transfers more to the sellers than the government hands back. Three reported metrics improved in the base case and the owner was poorer, which is the reason the buyback test is price against value and not accretion.

Worked example 8: Speed of adjustment and what a snap-to-target model invents (India, ₹)

Sahyadri Freight Systems Ltd (synthetic) has an enterprise value of ₹5,000 crore, a stated target of 35% debt to value, and an actual ratio of 52% after a debt-funded acquisition. It borrows at 8% and is taxed at 25.17%. Take the annual speed of adjustment as 0.25.

Solution. Step 1, the half-life. ln(0.5)/ln(0.75) = 2.409421 years. For reference, λ = 0.30 gives 1.943358 years, λ = 0.20 gives 3.106284 years, and λ = 0.10 gives 6.578813 years.

Step 2, the path, and the interest a snap-to-target model omits.

Step 3, the totals. Over five years the omitted interest is ₹155.589844 crore and the invented after-tax earnings are ₹116.427880 crore. The ratio is still 4.0342 percentage points above target at the end of year five.

Read it. The error compounds twice: it inflates every year's earnings, and it does so most in the early years that carry the most weight in a discounted cash flow. Put the target structure in the terminal value if you like, but the explicit forecast period has to carry the actual debt on its actual schedule. The deviation itself reads separately and just as usefully: a firm 17 percentage points above target is telling you where the next three years of free cash flow are going before management says a word.

The eight examples restated at full precision, so any figure above can be reproduced or contradicted: unlevered value of Chandori 1,995.466667; levered value at 1.8 turns 2,184.200747; value per share at 1.2, 1.8 and 3.0 turns 99.263563, 102.010037 and 99.122027; earnings per share at those three points 15.204770, 16.968574 and 24.117772; Winslow's indifference operating profit 143.750000 and its earnings per share there 2.343750; the probability the debt plan wins 0.720166; Brandywine's levered cost of equity at D/E = 0.571429 14.571429% against the zero-debt-beta relever's 15.428571%; the Miller gain per unit of debt 0.060704 in India, negative 0.016858 in the United States at a 21% corporate rate and 0.034628 at 25%; Braddock's debt loss 30.000000 against a value loss of 5.000000; Deccan's lender gain 75.000000 against an owner outlay of 100.000000; Wexford's post-recap intrinsic value per share 19.750000 and its break-even repurchase price 26.666667; and Sahyadri's five-year omitted interest 155.589844.

Practice set

Work each problem fully before reading its solution. A calculator is assumed. Numeric answers within about 1% score as correct unless the problem states otherwise. All entities are synthetic, all schedules are illustrative, and the first four problems carry more scaffolding than the rest.

P1 (guided). The arbitrage, placed. An unlevered firm earns a perpetual 96 with an unlevered cost of equity of 12%. Its levered twin carries 400 of perpetual debt at 5% and the market prices the levered twin at 880. There are no taxes. Construct the trade that captures the mispricing on a 1% stake, and state the profit.

Solution. The unlevered firm is worth 96/0.12 = 800.000000, so the levered twin is overpriced by 80.000000. Its equity is 880 − 400 = 480.000000, and 1% of it is worth 4.800000, paying 0.01 × (96 − 20) = 0.760000 a year. Sell that stake for 4.800000 and borrow 4.000000 personally at 5%, giving 8.800000 of cash. Buy 1% of the unlevered firm for 8.000000, which pays 0.960000 a year, and pay 0.200000 of personal interest, leaving 0.760000. Identical income, identical risk, and 0.800000 left over, which is 1% of the 80.000000 mispricing.

P2 (guided). Proposition II without taxes. A firm's unlevered cost of equity is 11.5%, it borrows at 5.5%, and it runs a debt-to-equity ratio of 0.80. There are no taxes. Compute the levered cost of equity and the weighted average cost of capital.

Solution. r_E = 0.115 + (0.115 − 0.055) × 0.80 = 0.163000, so 16.3000%. Equity is 1/(1 + 0.80) = 0.555556 of the total and debt is 0.444444, so the weighted average is 0.555556 × 0.163 + 0.444444 × 0.055 = 0.115000, exactly the unlevered cost of equity. The blend is unmoved because the rise in the cost of equity was calibrated to offset the cheaper debt precisely.

P3 (guided). Proposition II with taxes, verified twice. Operating profit is 250 in perpetuity, the unlevered cost of equity is 13%, the cost of debt is 7%, the tax rate is 25%, and the firm carries 400 of permanent debt. Compute the unlevered value, the levered value, the cost of equity and the weighted average cost of capital, and verify the last two independently.

Solution. V_U = 250 × 0.75/0.13 = 1,442.307692. V_L = 1,442.307692 + 0.25 × 400 = 1,542.307692, so equity is 1,142.307692 and D/E = 0.350168. Then r_E = 0.13 + (0.13 − 0.07) × 0.75 × 0.350168 = 0.145758, so 14.5758%. The weighted average is (1,142.307692/1,542.307692) × 0.145758 + (400/1,542.307692) × 0.07 × 0.75 = 0.121571, so 12.1571%. First check: r_0 (1 − tD/V_L) = 0.13 × (1 − 0.25 × 400/1,542.307692) = 12.1571%. Second check: 250 × 0.75 / 0.121571 = 1,542.307692, the levered value you started from.

P4 (guided). The relevering correction. A firm's unlevered beta is 0.95, its debt-to-equity ratio is 0.60, the tax rate is 25%, the risk-free rate is 4.5%, the equity risk premium is 5.5% and it borrows at 7.2%. Compute the levered beta and the cost of equity both with a zero debt beta and with the debt beta carried through, and state the error in basis points.

Solution. The debt beta is (0.072 − 0.045)/0.055 = 0.490909. With a zero debt beta, β_L = 0.95 × (1 + 0.75 × 0.60) = 1.377500 and r_E = 4.5 + 1.3775 × 5.5 = 12.076250%. Carrying the debt beta, β_L = 0.95 + (0.95 − 0.490909) × 0.75 × 0.60 = 1.156591 and r_E = 10.861250%. The overstatement is 121.500000 basis points, which the closed form β_D (1 − t)(D/E) × ERP = 0.490909 × 0.75 × 0.60 × 0.055 reproduces exactly.

P5. The Miller gain at a different personal-tax pair. A corporate rate of 25.17% applies. The investor faces 15% on equity income and 30% on interest income. Compute the net gain per rupee of debt, the break-even corporate rate, and the gain on ₹500 crore of permanent debt against the corporate-shield-only figure.

Solution. G_L/D = 1 − (0.7483 × 0.85)/0.70 = 0.091350, so 9.1350 paise per rupee. The break-even corporate rate is 1 − 0.70/0.85 = 17.6471%, and 25.17% sits above it. On ₹500 crore the net gain is ₹45.675000 crore, against the ₹125.850000 crore the corporate shield alone would have claimed. The personal layer removed 63.7068% of the advertised shield, and it removed less than in the top-bracket case because this investor's spread between the two personal rates is narrower.

P6. Solve a coverage-to-spread fixed point. Operating profit is ₹240 crore and debt is ₹900 crore. The benchmark rate is 6.50% and the small-firm rating table maps coverage above 3.5 to BB+ at 2.00%, coverage of 3.0 to 3.5 to BB at 2.40%, and coverage of 2.5 to 3.0 to B+ at 3.00%. Iterating downward from the narrowest spread, find the stable rating, spread, cost of debt, coverage and interest.

Solution. Start at 0.60%, so r_D = 7.10% and coverage is 240/(900 × 0.071) = 3.756, giving BB+ at 2.00%. Retry at 8.50%: coverage is 3.137, giving BB at 2.40%. Retry at 8.90%: coverage is 2.996, giving B+ at 3.00%. Retry at 9.50%: coverage is 240/85.500000 = 2.807018, still B+. Stable at B+, a 3.00% spread, a 9.50% cost of debt, coverage of 2.807018 and interest of ₹85.500000 crore. The first pass overstated the rating by three notches, which is why the iteration is not optional.

P7. One point on a trade-off curve. Unlevered value is ₹1,200 crore, the tax rate is 25.17%, debt is ₹400 crore, the distress probability at the implied rating is 3.6%, and distress costs 40% of unlevered value. Compute the tax shield, the expected distress cost, the levered value and the net gain per rupee of debt.

Solution. Shield 0.2517 × 400 = ₹100.680000 crore. Expected distress cost 0.036 × 0.40 × 1,200 = ₹17.280000 crore. Levered value 1,200 + 100.68 − 17.28 = ₹1,283.400000 crore. Net gain per rupee of debt is (100.68 − 17.28)/400 = 0.208500, so 20.85 paise, which is 82.84% of the gross shield. That fraction is what falls as leverage rises, and it is the whole shape of the curve.

P8. The indifference point, second set. A firm with 25 million shares can raise $200 million by issuing 5 million new shares at $40, or by issuing notes at 7%. The tax rate is 25%. Find the indifference operating profit and the earnings per share there, then compare the plans at $40 million and $90 million of operating profit.

Solution. Interest is 200 × 0.07 = $14.000000 million, the equity plan gives 30.000000 million shares, so EBIT* = 14 × 30/5 = $84.000000 million and earnings per share there are 84 × 0.75/30 = $2.100000. At $40 million: equity plan $1.000000, debt plan $0.780000, so equity wins. At $90 million: equity plan $2.250000, debt plan $2.280000, so debt wins. The crossing is unique because the debt line is steeper and starts lower.

P9. A buyback at three prices. A firm's intrinsic equity value is $300 million across 50 million shares. It borrows $60 million at a 25% tax rate and repurchases stock. Compute intrinsic value per share afterwards at repurchase prices of $5, $8 and $10, and find the break-even price.

Solution. Intrinsic value is $6.000000 a share and the shield is 0.25 × 60 = $15.000000 million, so intrinsic equity afterwards is 300 + 15 − 60 = $255.000000 million at any price. At $5: 12.000000 million shares retire, leaving 38.000000 million, so $6.710526 a share, up 11.8421%. At $8: 7.500000 million retire, leaving 42.500000 million, so $6.000000, unchanged. At $10: 6.000000 million retire, leaving 44.000000 million, so $5.795455, down 3.4091%. The break-even price is 6/(1 − 0.25) = $8.000000, which the middle row confirms.

P10. Asset substitution, second set. A firm owes $80 million in one year. The safe project pays $90 million or $110 million with equal probability; the risky one pays $30 million or $160 million. Build the payoff table and state who gains, who pays and how much evaporates.

Solution. Safe: firm value 100.000000, debt 80.000000, equity 20.000000. Risky: firm value 95.000000, debt (30 + 80)/2 = 55.000000, equity (0 + 80)/2 = 40.000000. The switch destroys 5.000000 of firm value, raises equity 20.000000 and cuts debt 25.000000. Of every dollar the lenders lose, 80 cents reaches the owners and 20 cents disappears. The owners take the risky project, and the lenders who saw it coming already charged for it.

P11. Debt overhang, second set. A firm owes ₹600 crore maturing in a year against assets worth ₹400 crore or ₹900 crore with equal probability. A project costs ₹80 crore of fresh equity and adds ₹120 crore in both states. Will the owners fund it, and who takes the value?

Solution. Without: debt (400 + 600)/2 = ₹500.000000 crore, equity (0 + 300)/2 = ₹150.000000 crore. With, on assets of ₹520 crore or ₹1,020 crore: debt (520 + 600)/2 = ₹560.000000 crore, equity (0 + 420)/2 = ₹210.000000 crore. The project's net present value is 120 − 80 = ₹40.000000 crore. The owners gain ₹60.000000 crore and must fund ₹80.000000 crore, so they are ₹20.000000 crore worse off and decline. The lenders would have gained ₹60.000000 crore, which is 50% of the ₹120 crore created, for nothing.

P12. Sizing an announcement effect. An Indian issuer with a ₹12,000 crore market capitalisation announces a ₹1,500 crore equity raise. Take the announcement effect as negative 2.8%. Compute the loss in rupees and per rupee raised, and compare it with the corporate tax shield a debt raise of the same size would have created at 25.17%.

Solution. The loss is 12,000 × 0.028 = ₹336.000000 crore, which is 336/1,500 = ₹0.224000 per rupee raised, so 22.4 paise. The shield on ₹1,500 crore of permanent debt is 0.2517 × 1,500 = ₹377.550000 crore. The two are within about 11% of each other, so on these inputs the announcement cost of the equity route is nearly as large as the entire corporate tax advantage of the debt route, before any distress cost is deducted. Note the assumption doing the work: the shield is a perpetuity and the announcement effect is a one-time hit, so comparing them directly is generous to the equity route.

P13. Speed of adjustment. A firm sits at 48% debt to value against a 30% target, with an annual adjustment speed of 0.20. Compute the half-life, the ratio at the end of each of the next three years, and roughly how long until the gap falls to 2 percentage points.

Solution. Half-life is ln(0.5)/ln(0.80) = 3.106284 years. Year one: 0.48 − 0.20 × 0.18 = 44.400000%, a gap of 14.400000 points. Year two: 41.520000%, a gap of 11.520000. Year three: 39.216000%, a gap of 9.216000. The gap falls to 2 points when 0.80^n × 18 = 2, so n = ln(2/18)/ln(0.80) = 9.846686 years. A forecast that assumes the target is reached inside a five-year explicit period is asserting an adjustment speed roughly twice this one.

P14. Read the financing decision. A profitable, low-growth Indian manufacturer with net debt of 0.4 turns of EBITDA and a decade of unbroken free cash flow announces a large debt-funded tender buyback at a 15% premium to the market price. State what each of the four theories in this file predicts about the announcement, name the single fact that would decide whether the board is right, and say what you would ask for.

Solution. The trade-off model says a firm at 0.4 turns with stable cash flow is far left of any plausible optimum, so the shield gained is worth more than the distress cost imported, and the move creates value. The free-cash-flow argument says the same thing for a different reason: a mature firm with no reinvestment opportunities and a decade of accumulating cash has an agency problem, and debt service is the bonding device. The pecking order is silent on a repurchase but notes that management is doing the opposite of issuing equity, which is a positive signal about private information. The signalling literature predicts a positive announcement return, larger for a fixed-price tender than for an open-market authorisation. The single deciding fact is the price against a conservatively estimated intrinsic value, because at a 15% premium the shield funds only up to a 33.6% premium at a 25.17% tax rate and everything above that is a transfer to the sellers. Ask for the board's own intrinsic-value estimate and its method, the reinvestment pipeline the cash would otherwise have funded, and the pro-forma coverage ratio at trough EBITDA rather than at last year's.

Applied mini-project

Place one real company on its own trade-off curve, and say what would move it.

Pick one listed company, in either market, with publicly disclosed debt and a traded or rated credit. Build the curve from its own numbers, place it, and write the recommendation. Every input is sourced and dated. Three artefacts, each with a numbered test it must pass before it counts.

Artefact 1, the inputs sheet (2 tests). Assemble, from primary documents only: operating profit and EBITDA for the last full year, net debt at the most recent balance-sheet date, the share count, the market price on a stated date, the marginal statutory tax rate, and the company's actual cost of debt from a traded bond yield or the borrowing note. Then state an unlevered cost of equity, built the way the valuation phase builds one, and a distress-cost fraction with one paragraph defending it for this industry. Test 1: every figure carries a document name, a page or note reference and a date. Test 2: your implied cost of debt at current leverage sits within 150 basis points of the rate the company discloses paying; if it does not, say why, and use the disclosed rate.

Artefact 2, the curve (3 tests). Build a grid of at least six leverage levels spanning the company's current level. At each, solve the coverage-to-spread fixed point, apply your distress-probability schedule, and compute the tax shield, the expected distress cost, the levered value and the value per continuing share. Test 3: at the company's current leverage, your model's levered value is within 20% of its actual enterprise value; report the gap and name what explains it rather than tuning inputs until it closes. Test 4: the weighted average cost of capital implied by each levered value reaches its minimum at the same row where value peaks. Test 5: re-run the entire grid at two other distress-cost fractions and at spreads 50 basis points wider, and report where the optimum moves in each case.

Artefact 3, the recommendation (2 tests). One page. State where the company sits against the optimum, in turns of EBITDA and in value per share, and the width of the plateau within which you cannot tell the difference. Then give the recommendation, which may legitimately be that leverage is not this company's problem. Test 6: name the two facts that would change the answer, with the direction and size of the change each would produce. Test 7: state one thing the model cannot see here, and what you would read to see it.

Deliverable. One workbook or notebook plus the one-page recommendation. The workbook carries the sourced inputs sheet, the grid, the three sensitivity runs and the seven test results.

Rubric (18 points; pass at 14 with no zero on any row).

Any artefact whose test fails scores zero for that row regardless of how well it is written, because a curve built on an unsourced spread is a drawing rather than an analysis.

Reading & resources

• Franco Modigliani and Merton Miller, "The Cost of Capital, Corporation Finance and the Theory of Investment", American Economic Review, 1958. The original arbitrage proof, and shorter than its reputation suggests. [Free] [Intermediate]
• Merton Miller, "Debt and Taxes", Journal of Finance, 1977. The personal-tax layer and the equilibrium that carries his name. [Paid] [Advanced]
• Michael Jensen, "Agency Costs of Free Cash Flow, Corporate Finance and Takeovers", American Economic Review, 1986. The free-cash-flow argument in nine pages. [Free] [Intermediate]
• Stewart Myers and Nicholas Majluf, "Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have", Journal of Financial Economics, 1984. The pecking order derived rather than asserted. [Paid] [Advanced]
• Stewart Myers, "The Capital Structure Puzzle", Journal of Finance, 1984. The honest statement of why the trade-off model over-predicts leverage. [Paid] [Intermediate]
• Richard Brealey, Stewart Myers and Franklin Allen, Principles of Corporate Finance, the chapters on capital structure. The clearest textbook treatment of the propositions and the frictions. [Paid] [Beginner to Intermediate]
• Aswath Damodaran, Applied Corporate Finance, the optimal-capital-structure chapters, plus his Current Data page for the ratings, interest-coverage and default-spread sheet the trade-off arithmetic here consumes. The spreads are refreshed each January and must be pulled live. [Paid book, free data] [Intermediate]
• Tim Koller, Marc Goedhart and David Wessels, Valuation, the capital-structure and payout chapters. The practitioner's version, with the conservation-of-value discipline attached. [Paid] [Intermediate to Advanced]
• Gregor Andrade and Steven Kaplan, "How Costly is Financial (not Economic) Distress?", Journal of Finance, 1998. The best-known attempt to separate distress costs from the economic shock that caused the distress, and the source of the ranges quoted for indirect costs. [Paid] [Advanced]
• Murray Frank and Vidhan Goyal, "Testing the Pecking Order Theory of Capital Structure", Journal of Financial Economics, 2003, read alongside Lakshmi Shyam-Sunder and Stewart Myers, Journal of Financial Economics, 1999. Two papers that disagree, which is why they are more useful together. [Paid] [Advanced]
• Reserve Bank of India, Handbook of Statistics on the Indian Economy, and the Clearing Corporation of India and Financial Benchmarks India pages for the government securities valuation curve. The India anchor for any benchmark rate used in a rupee build. [Free] [Beginner to Advanced]
• United States Treasury daily par yield curve rates, and the Financial Industry Regulatory Authority's bond data for traded corporate yields. The dollar anchors, and the way to replace a synthetic spread with a real one. [Free] [Beginner]
• Company filings: the borrowing note in an Indian annual report discloses actual rates on recent borrowings, and a United States 10-K's long-term debt note does the same. This is the single most under-used free source in the whole exercise. [Free] [Intermediate]

Do this, not just read. Rebuild Worked example 1's grid in a spreadsheet, solving each row's cost of debt with Goal Seek. Then break it deliberately: force the rating one notch better than the coverage supports and watch the optimum slide right by a full turn while every other number on the page stays plausible. Ten minutes of watching a wrong rating quietly move an optimum will teach you more about this model's fragility than any amount of reading about it.

Teach it back & journal

Feynman prompt. Sit down with someone who believes debt is cheap because its interest rate is lower than the cost of equity, and take that belief away in ten minutes without using the word irrelevance. Start with the no-tax table: same firm, three leverage levels, cost of equity climbing from 12% to 21%, weighted average never moving off 12%. Make them compute one row themselves, net income over market equity, so the number arrives in their own handwriting. Then add taxes, let them watch the blend finally fall, and ask where the money came from. If they say the government, they have it. Then hand them the Miller expression and let them find that the same investor pays tax on the interest received, and that on top-bracket rates the gain shrank to about six paise per rupee in India and changed sign in the United States.

Close on the part that matters most. Show them the trade-off grid and ask them to point at the optimum. Then move the distress-cost fraction from 40% to 25% in front of them and watch the optimum move 0.6 turns. If their next question is about the assumption rather than the answer, they have understood the whole exercise, and you can stop.

Journal prompt. For the next month, every time you meet a financing decision in the wild, a rights issue, a debt-funded buyback, a convertible, a dividend initiation or a large acquisition paid for in paper, write four lines before you form any opinion. Line one names what the firm issued and what that claim's payoff sensitivity says about management's private information. Line two sets the price beside your own estimate of value and works out how much premium the tax shield would have funded at this company's marginal rate. Line three places the firm on its own leverage plateau, roughly, and says whether the move would take it off. Line four records what the earnings-per-share arithmetic alone would have told you, had you looked at nothing else.

At the end of the month, count three things: the decisions defended publicly on earnings-per-share accretion, the decisions that disclosed a view about intrinsic value, and the notes of your own that leaned on a schedule or a fraction you assumed rather than sourced. The third count is usually the one that stings.

Then write one page on this. The uncomfortable finding here is that the trade-off curve is flat, so the single most-taught question in corporate finance usually has an answer wide enough to drive a truck through. That does not make the exercise useless; it changes what the exercise is for. Write down what you now think capital-structure analysis is good for, which friction does the most work in the companies you follow, and what you would have to observe to change your mind.