Cycles, Bubbles & Crises

Phase 7 · Week 61 · ~20 hours

Where it sits. In 2020 the United States had a recession that lasted two months. In 2022 it had two consecutive quarters of falling GDP and no recession at all. Fifth module of the macro phase, and the one where the machinery of the previous four comes alive as history repeating with variations. M7.01 built the economic machine and, crucially here, Dalio's three forces: productivity, the short-term debt cycle (~5–8 years), and the long-term debt cycle (~50–75 years). M7.02 gave you the central bank and its lender-of-last-resort fire hose. M7.03 gave you the yield curve, whose inversion is the single most famous leading indicator of the thing studied here. M7.04 gave you fiscal policy and the anatomy of sovereign crises. All of it now goes to work on the two questions a macro-aware analyst must be able to answer about any moment in time: where are we in the cycle, and is the system carrying dangerous excess? The teaching covers the anatomy of the business cycle and how professionals actually date it (the NBER method, not the "two quarters of falling GDP" folk rule); the indicator toolkit (leading, coincident, lagging) and how to build a cycle read from a panel of numbers; the PMI and why a survey of purchasing managers leads the hard data. Then the engine room: the credit cycle and Minsky's Financial Instability Hypothesis (why stability itself breeds the next crisis), the progression from hedge to speculative to Ponzi finance, the Minsky moment, and the Kindleberger bubble anatomy as a reusable template. That template then runs through seven of the great cycles and crises (1929, the 1970s and Volcker, Japan's lost decades, the dot-com bust, the 2008 GFC, the 2020 COVID shock, and the Asian crisis of 1997–98) before the analyst's payoff: sector rotation and the growth/inflation 2×2, the framework that connects "where are we in the cycle" to "what should I own." Throughout, one piece of honesty is enforced: cycles rhyme but do not repeat, timing the top or bottom is very nearly impossible, and the value of everything here is in recognizing regime and excess, not in calling turns.

Suggested hour budget. Core content (the cycle, the indicators, Minsky, the anatomy, the seven cases, the 2×2, India): 9h · worked examples (the cycle read, the hedge/speculative/Ponzi classification, mapping a bubble to the anatomy, the surprise-vs-level trade): 3h · practice set (indicator-panel staging, financing classification, case mapping, regime→asset): 4.5h · applied mini-project (place the current moment in the cycle and write a one-page regime note): 2h · flashcards + mastery check: 1.5h.

App link. This module is the content engine behind the Macro & Cycles branch's scenario-event system (competency C8). The "recession," "rate shock," "commodity shock," and "fraud reveal" events that fire mid-simulation in the portfolio simulator are this module's cycle turns and Minsky moments made playable, with consequences that reward the learner who recognized the regime and reduced risk before the trigger. Four drills seed here: a cycle-staging drill (an indicator panel in, the correct cycle stage out, timed), a financing-classification drill (a borrower's cash-flow profile in, hedge/speculative/Ponzi out), a bubble-anatomy drill (a case vignette in, the correct Kindleberger stage out), and a regime-to-allocation drill (a growth/inflation quadrant in, the winning and losing asset classes out). It hands the cycle stage and regime forward to the portfolio simulator's scenario seeds and to the M7.07 dashboard capstone, where reading indicators as rate-of-change versus expectations and mapping them onto the debt-cycle stage is the graded deliverable.

Learning objectives

By the end you can:

1. Describe the anatomy of the business cycle (expansion → peak → contraction → trough) and date it the way professionals do (the NBER method: depth, diffusion, and duration across real GDP, income, employment, and industrial production), and explain precisely why the popular "two consecutive quarters of falling GDP" rule is wrong, using 2020 (a two-month recession) and 2022 (two negative GDP quarters and no recession) as the proofs.
2. Classify any indicator as leading, coincident, or lagging, name the key members of each family (yield curve, building permits, new orders, the stock market, credit spreads lead; GDP, employment, industrial production coincide; the unemployment rate, CPI, unit labour costs lag), and build a coherent cycle read from a mixed panel.
3. Explain how a PMI is constructed (the diffusion-index formula % higher + 0.5 × % same, and the 50 line), compute one from its sub-components, and say why a survey of purchasing managers leads the hard data.
4. State Minsky's Financial Instability Hypothesis ("stability is destabilizing"), define hedge, speculative, and Ponzi finance precisely from the cash-flow-coverage test, classify a real lending situation, and explain what the Minsky moment is and what triggers it.
5. Apply the Kindleberger/Minsky bubble anatomy (displacement → boom/credit expansion → euphoria → distress/profit-taking → revulsion/panic) as a template to any mania, and identify which stage a described episode is in.
6. Run each of seven great cycles and crises through the anatomy (1929, 1970s stagflation and the Volcker cure, Japan's 1990s bubble and balance-sheet recession, the dot-com bust, the 2008 GFC, the 2020 COVID shock, and the Asian crisis of 1997–98), naming the specific mechanism and the durable lesson of each.
7. Use the growth/inflation 2×2 to name the four regimes and the asset classes that win and lose in each; explain sector rotation across the cycle; and explain the All-Weather insight that assets are priced off surprises relative to expectations, not levels — so a rate hike can rally bonds.
8. Characterize India's cycle: historically less endogenous-credit-driven than the US, with its own specific vulnerabilities (oil and the current account, FPI flows, the 2018 IL&FS/NBFC episode as India's clearest Minsky moment). And articulate honestly why recognizing regime and excess beats trying to call the turn.

Prerequisites & connections

Builds on. M7.01 is the spine: the short- and long-term debt cycles are the cycles anatomized here, and "spending = money + credit" is why a credit contraction is a recession; Minsky is the microfoundation of Dalio's debt cycle. From M7.02, the central bank's lender-of-last-resort role (Bagehot) is the crisis function every case study turns on, and "read the surprise, not the print" is the seed of the surprise-vs-level idea in §4.8; the policy response is often the difference between a crash and a depression. From M7.03, the yield-curve inversion is the marquee leading indicator (§4.3), and the "rates are the gravity of valuation" idea explains why bubbles inflate when rates are low; inflation types (cost-push vs demand-pull) are the whole story of the 1970s. From M7.04, fiscal policy is the other half of the 2008/2020 response, debt-deflation and r vs g connect to the crash mechanism, and the anatomy of sovereign crises directly sets up the Asian '97 case. From M3.05, reverse DCF ("what is priced in?") and duration are exactly the tools for judging a bubble's valuation excess (dot-com) and for understanding why long-duration assets lead the cycle. From M2.02/M2.07, leverage, coverage ratios, and the credit lens are the raw material for classifying hedge/speculative/Ponzi. From M6.04, Marks's pendulum of sentiment and the behavioral biases (herding, recency, narrative) are the psychology inside the euphoria stage.

Feeds forward. M7.06 takes up FX, capital flows, and the impossible trinity: the Asian '97 case here is the bridge, deliberately left at the water's edge of "currency and maturity mismatch" so M7.06 can formalize the sudden-stop mechanism, the eurodollar plumbing, and the global financial cycle. M7.07 is the dashboard capstone, and it inherits the indicator taxonomy (leading/coincident/lagging), the PMI, the credit-cycle stage, and the regime 2×2 that the dashboard is built to read; the "regime memo" you write there is the cycle read made permanent. M9 (portfolio, risk, and process): "where are we in the cycle" sets the exposure dial, "is there excess" sets the risk budget, and Marks's "you can't predict, you can prepare" is the temperament installed here, which means reducing risk when the margin of safety has vanished, without pretending to know the day. M8 (the rapid-analysis system): every teardown from here on carries a one-line macro overlay, the cycle stage and regime the business is operating in.

The one-sentence version of this module. Economies and markets move in cycles that are driven, above all, by the expansion and contraction of credit, and because stability itself makes lenders and borrowers bolder (Minsky), every long calm builds the leverage and the Ponzi finance that the next shock detonates through a recognizable anatomy (Kindleberger); the analyst's job is not to time the turn, which is nearly impossible, but to read where in the cycle the economy is (from leading, coincident, and lagging indicators and the growth/inflation regime) and how much excess the system carries, and to set risk accordingly.

Core teaching content

4.7 Seven cycles and crises, run through the anatomy

Now the payoff: the same template applied seven times. Each case gives you the displacement, the excess, the trigger/Minsky moment, the crash, the response, and the durable takeaway: the lesson. Read them as a set: together they teach that the costumes always change (a new technology, a new financial instrument, a new country) while the anatomy stays the same, which is exactly why the next crisis fools people who memorized the last one. Figures are round, order-of-magnitude-correct, and historical (these are dated events, not live values).

4.7.7 The Asian crisis, 1997–98: the emerging-market currency and maturity mismatch (bridge to M7.06)

The template for a different family of crisis, not an equity bubble but a currency and external-debt crisis, and the deliberate bridge to M7.06. Displacement: the "Asian miracle" of the early-mid 1990s, when fast-growing "tiger" economies (Thailand, Indonesia, South Korea, Malaysia) ran currencies pegged to the US dollar alongside open capital accounts, a combination that attracted floods of foreign short-term dollar-denominated capital chasing high local returns with (apparently) no currency risk, funding property and investment booms.

The fatal structure, a double mismatch (this is the whole lesson): local banks and firms borrowed in US dollars, short-term, and lent/invested in local currency, long-term. That is a currency mismatch (owe dollars, earn baht/rupiah) stacked on a maturity mismatch (owe short, hold long). It is Minsky's speculative/Ponzi structure (§4.5) with a lethal FX twist.

Trigger/Minsky moment: as the US dollar strengthened and current-account deficits widened, speculators attacked the Thai baht peg; Thailand exhausted its reserves defending it and was forced to float (July 2, 1997), and the baht collapsed. The self-reinforcing cascade: as the local currency fell, the dollar debts exploded in local-currency terms, since a firm that owed $10m now owed far more baht to service the same dollar loan. That drove mass insolvency, which triggered capital flight, which drove the currency down further, which deepened the insolvency (a currency-flavoured debt-deflation, and a textbook "sudden stop" of capital inflows). Contagion spread across the region; the Indonesian rupiah lost ~80%, Indonesia's GDP fell ~13% in 1998, and South Korea needed a ~$58bn IMF bailout. The IMF programs attached harsh austerity conditions that remain deeply controversial (many argue they worsened the collapse). Underneath it all sat the impossible trinity (M7.06): a country cannot simultaneously hold a fixed exchange rate, free capital movement, and an independent monetary policy. The tigers tried to run all three, and the crisis was the trinity collecting its debt.

Lesson: the emerging-market crisis is a currency/maturity-mismatch crisis, not primarily an equity bubble; "denomination is destiny" (M7.04), and the toxic recipe is fixed exchange rate + open capital account + large short-term foreign-currency debt. This is the sudden-stop template that M7.06 formalizes.

India's response, a lesson learned: India was largely spared, precisely because it had not fully opened its capital account (foreign short-term debt was limited and the rupee was managed, not pegged-and-defended). India internalized the lesson permanently, which is a large part of why, to this day, it maintains a managed capital account, a managed float, and very large FX reserves as insurance against exactly this kind of sudden stop (its near-miss in the 2013 taper tantrum, §4.9, showed the vulnerability is real even for India).

4.10 The honest truth about timing: recognize regime and excess, do not call the turn

A guardrail now, and it matters as much as any framework above, because those frameworks are seductive, and the seduction is dangerous. Everything you have learned makes it feel as though you should be able to time the cycle. You cannot, and believing you can is how analysts blow up.

Cycles rhyme, but they do not repeat. Every cycle has a different displacement (a new technology, a new instrument, a new country), a different excess (margin in 1929, land in Japan, dot-com valuations, subprime CDOs, NBFC paper in India), and a different trigger. The anatomy is constant; the costume is always new, which is exactly why people who "learned the lesson" of the last crisis get blindsided by the next one (the generals who fought subprime were unprepared for a pandemic; the ones who fought inflation in the 1970s were unprepared for the deflation of 2008). Pattern-matching the shape is powerful; pattern-matching the specifics fails.

Timing the top or bottom is very nearly impossible, and the attempt is ruinous. Three reasons, each worth internalizing:

• Euphoria can last far longer than solvency. "The market can remain irrational longer than you can remain solvent" (often attributed to Keynes; no source in his writings has been found). The Nasdaq nearly quadrupled in the three-plus years after Greenspan's 1996 "irrational exuberance" warning; countless investors who correctly identified the dot-com and housing bubbles were early — and being early, especially with leverage or shorts, is indistinguishable from being wrong until you are bankrupt.
• The exact triggers are unforecastable. No one can know which minor event, on which day, flips sentiment. The structure can be maximally fragile for months before the match is struck.
• The turns are only visible in hindsight (§4.2): the peak and trough are dated after the fact, sometimes years after, so "we are at the top" is never knowable in real time, only guessable.

So what is all of this for? Two things, and only two:

1. Recognizing the regime, roughly where in the cycle you are and which growth/inflation quadrant you're in, so you set reasonable expectations (a company modelled on late-cycle peak margins is mispriced) and don't fight the tape (don't be maximally bullish when every leading indicator has rolled over and every defensive lens says late-cycle). This is Marks's "taking the temperature of the market": you can't know what tomorrow holds, but you can know whether you're in a cheap, fearful, early-cycle environment or an expensive, euphoric, late-cycle one, and tilt accordingly.
2. Recognizing excess: the Minsky migration toward Ponzi finance, the euphoria markers, the leverage, the compressed spreads, the narrative unmoored from numbers. Spot those and you can reduce risk when the margin of safety has vanished, without any pretence of knowing the day it ends. You don't need to call the top to sell an asset that is obviously in stage-3 euphoria on 40× leverage; you need only recognize that the risk/reward has become terrible and act on that, accepting that it may keep rising after you step aside.

Marks's formulation is the one to carry forever: "You can't predict. You can prepare." The analyst's edge is not a crystal ball; it is (a) knowing where you are in the cycle well enough to size risk sensibly, and (b) recognizing dangerous excess well enough to be cautious when others are greedy and aggressive when others are fearful — a temperament, not a timing model. This is the direct handoff to M6 (Marks, behavioral discipline) and M9 (position sizing, risk, when to reduce exposure): the whole point of reading the cycle is not to trade it but to survive it and to keep your judgement calibrated to the weather. Recognize the regime; respect the excess; never bet the farm on the turn.

Common mistakes & how experts think differently

1. "Two consecutive quarters of falling GDP is a recession." The most common macro error there is. It is a folk rule, not the NBER definition, and it fails in both directions: 2020 was a recession without two quarters (too short, too deep) and 2022 had two negative quarters without a recession (jobs and incomes rose). Experts read breadth (depth × diffusion × duration across employment, income, production, sales), not a single GDP rule, and treat the word "technical recession" as a flag that someone is using the wrong definition.

2. Watching lagging indicators to make forward decisions. Beginners wait for the unemployment rate to fall before turning bullish and for inflation to peak before relaxing, but both lag the cycle by many months, so acting on them puts you a half-cycle behind. Unemployment rises into the recovery; inflation peaks after the economy peaks. Experts act on leading indicators (yield curve, permits, new orders, spreads), use coincident indicators to know where they are, and treat the lagging ones the crowd fixates on as confirmation, not signal.

3. Reading the level instead of the surprise (and the rate of change). "Growth is 6%, that's great" is a beginner's sentence; the market already priced 6%, so what moves it is 6% versus the 7% that was expected (a negative surprise) or the fact that growth decelerated from 8%. A rate hike can rally bonds if it's smaller than priced. Experts read every indicator as rate-of-change versus expectations, the single most important habit in macro-aware analysis, and the reason the M7.07 dashboard is built that way.

4. Believing "the displacement is real, so it can't be a bubble." The internet was real; Japanese industry was real; house-price stability was real. The kernel of truth is exactly what makes a bubble dangerous: it recruits the smartest people and silences the skeptics. Experts separate "is the technology/story real?" (often yes) from "is the price sane?" (often no). Being right about the internet in 2000 and buying Cisco at 150× earnings were two completely different bets, and only the second one mattered for your money.

5. Confusing solvency with liquidity (missing the Minsky structure). A speculative or Ponzi unit can be "profitable" and still die, not because it's insolvent but because it can't refinance when credit freezes. Banks, NBFCs, and any borrow-short/lend-long structure are liquidity-dependent, and the risk is not on the income statement but in the maturity and refinancing profile. Experts look at how an entity is financed (can it survive a frozen credit market?), not just whether it's currently making money. That is the question that flagged IL&FS, Bear Stearns, and every '97 tiger before they fell.

6. Assuming the last crisis's playbook fits the next one. Fighting the previous war. The 1970s were an inflation problem; 2008 was a deflationary credit-collapse problem requiring the opposite response; 2020 was an exogenous shock, not a credit bust at all. An analyst who "knows" that crises are deflationary got wrecked in 2021–22; one who "knows" they're inflationary got wrecked in 2008. Experts diagnose the nature of each shock (endogenous credit? external supply shock? inflation regime? deflation regime?) before reaching for a response.

7. Trying to time the top or bottom. The seductive error everything above warns against. Recognizing a bubble does not license shorting it or selling everything on a date: euphoria outlasts solvency, and the turn is only datable in hindsight. Experts use the cycle to tilt risk and set expectations, reducing exposure as excess builds and adding as fear peaks, while explicitly disclaiming any ability to call the day. "You can't predict, you can prepare."

8. Ignoring the credit cycle because "our banks are fine." The amplitude of a downturn is set by leverage and the credit structure, which build invisibly in the calm (everything is repaid on schedule right up to the Minsky moment). Complacency about credit because defaults are currently low is precisely the psychology Minsky says causes the next crisis: low defaults breed the confidence that breeds the leverage. Experts watch spreads, lending standards, and credit growth most closely when they look best.

9. Importing the US credit-cycle lens wholesale into India. India's cycle has historically been more about the monsoon, oil and the current account, and FPI flows than about an endogenous private-credit boom, so an analyst who reads India through a pure Minsky/US lens misses the oil-CAD-rupee shock channel and the global-financial-cycle (taper-tantrum) channel that actually drive much of India's cyclicality. (Though the IL&FS/NBFC episode shows India is developing its own credit cycle, so the lens must evolve, not be discarded.)

How experts think differently, in one line: the amateur asks "is the economy good or bad right now?" (a coincident, level, backward question); the professional asks "where are we in the cycle, how much excess is the credit system carrying, which growth/inflation regime is coming, and how does all that compare to what's priced in?", and then sizes risk to survive being wrong about the timing, because the timing is the one thing no one can get right.

Worked examples

Worked Example 1: Compute and read a PMI from its sub-indices (India)

Operationalizes §4.4: turn survey responses into a diffusion index and a headline PMI, and read the forward signal.

Setup. For a manufacturing survey month, the five ISM-style sub-indices come in as diffusion indexes: New Orders 56, Production 54, Employment 51, Supplier Deliveries 48, Inventories 46. (a) Confirm how the New Orders index of 56 could arise from raw responses. (b) Compute the headline PMI. (c) Is the sector expanding or contracting, and what does the composition say about the near future?

(a) The diffusion-index construction. Suppose, of the firms answering the new-orders question, 46% say "higher," 20% say "same," 34% say "lower." Then:

Diffusion index = %higher + 0.5 × %same = 46 + 0.5 × 20 = 46 + 10 = 56. ✓

The "same" responses split evenly, which is what puts the neutral point at 50 (equal higher and lower ⇒ 50). A reading of 56 means materially more firms are seeing rising orders than falling ones.

(b) The headline PMI (equal 20% weights on the five sub-indices):

PMI = 0.20 × (56 + 54 + 51 + 48 + 46) = 0.20 × 255 = 51.0.

(c) Interpretation. The headline 51.0 is above 50, so manufacturing is expanding, but only mildly (51 is a tepid, not a strong, expansion; distance-from-50 is the pace). The composition is the real signal, and it's a good-news story about the near future: New Orders (56) is well above the headline and above Production (54), so orders are running ahead of output and firms will have to lift production to keep up, which is bullish for next month. And note Supplier Deliveries at 48: because that component reads backwards (slower deliveries push it up), a low reading of 48 means deliveries are fast, with no supply bottlenecks, so the demand signal is "clean" (real demand, not a 2021-style shortage artefact). The forward read: a mild but improving expansion with rising orders, ample supply, and production set to accelerate, early/mid-cycle in flavour. (For India specifically, remember 51 sits low in India's own historical PMI range, which rarely dips below 50 except in shocks, so against India's benchmark this is a soft-ish month, not a strong one; §4.4's "no number without a benchmark.")

Worked Example 2: Build a cycle read from a mixed indicator panel (US)

Operationalizes §4.3: sort a panel into leading/coincident/lagging and name the stage, the core skill the app drills.

Setup. You are handed this US panel. Place the economy in the cycle.

Step 1, sort by family. Leading: yield curve (inverted), building permits (falling), ISM new orders (45, contracting), credit spreads (widening): all four leading indicators have rolled over, decisively. Coincident: payrolls (still positive), industrial production (flat/soft): still okay but softening. Lagging: unemployment rate (low but turning up), core CPI (elevated but easing): still "good" on the surface.

Step 2, read the pattern. This is the signature late-cycle / approaching-peak configuration, and it is the most misread moment in the whole cycle. The numbers the crowd and the headlines watch (jobs +150k, a 3.9% unemployment rate, still-positive activity) look fine, while every forward-looking indicator has already turned down (inverted curve, collapsing permits, sub-50 orders, widening spreads). The lagging indicators (unemployment ticking up from its low, inflation easing from its peak) are the tell that the peak has passed even though the levels still look benign.

Step 3, the verdict and the "so what." Late cycle, peak likely passing, recession risk elevated over the next 6–18 months (the classic post-inversion lead time). The stance implication (via §4.8): begin rotating toward defensives and quality (staples, health care, utilities) and duration (the widening spreads and inverted curve say rate cuts and a slowdown are coming), reduce cyclical and high-yield exposure, and raise the bar for adding risk, without declaring "the recession starts next month" (you can't date it; §4.10). The single most important discipline here: don't be fooled by the strong coincident/lagging numbers. That's exactly the trap late-cycle sets, and reading the leading indicators against the still-good headline data is where cycle analysis earns its keep.

Worked Example 3: Classify three financing situations as hedge, speculative, or Ponzi (India + US)

Operationalizes §4.5: apply the cash-flow-coverage test that is the most examined idea here.

The test (fix it firmly): compare the unit's expected operating cash flow to its debt obligations. Hedge = cash flow covers both interest and principal. Speculative = covers interest only (must roll the principal). Ponzi = covers neither: must borrow more or sell assets even to pay the interest.

Unit A, an Indian FMCG company (illustrative). Annual operating cash flow ~₹800 crore. It has a ₹1,000 crore term loan at 9%: interest = ₹90 crore/year, and scheduled principal repayment = ₹200 crore/year (5-year amortization). Total annual debt service = 90 + 200 = ₹290 crore, against ₹800 crore of cash flow, covered ~2.8× from income alone. Classification: HEDGE. It services and retires its debt from earnings; it needs neither refinancing nor asset sales, so it is robust to a rate spike or a frozen credit market.

Unit B, a US commercial-real-estate owner (illustrative). A building generating $10m/year net operating income, financed with a $150m interest-only loan at 6%: annual interest = $9m (covered 1.1× by the $10m NOI), but the $150m principal is due as a bullet at maturity and cannot be paid from income; it must be refinanced (a new loan taken to repay the old one). Classification: SPECULATIVE. Solvent on interest, but entirely dependent on being able to refinance the principal. The danger: if, at maturity, rates have risen (so a new loan at 8% costs $12m against $10m NOI, now uncovered) or credit has tightened (no new loan available), the unit fails despite having covered its interest all along. This is the exact structure that turns commercial real estate into a slow-motion crisis when rates rise into a refinancing wall.

Unit C, a subprime homebuyer, 2006 (illustrative), and its NBFC analogue. A borrower with $40k income buys a $500k house on a negative-amortization / teaser mortgage: the "affordable" early payment is less than the interest accruing, so the loan balance grows over time, and the borrower can only "afford" the house by refinancing against its rising price. Cash flow covers neither principal nor even the full interest. Classification: PONZI. It is alive only while house prices rise (so it can refinance): the instant prices stop rising, it is insolvent (§4.7.5). The Indian analogue is DHFL/IL&FS (§4.9): funding a 20-year infrastructure loan or mortgage with 90-day commercial paper it must roll continuously, and (for the weakest) borrowing new paper to pay interest on old. Ponzi-adjacent, alive only while the paper keeps rolling, dead the moment it doesn't (September 2018).

The teaching point: the same interest rate and the same asset can support a hedge, speculative, or Ponzi structure depending entirely on the financing. And Minsky's whole point is that, over a long calm, the mix shifts from A toward C invisibly, because C keeps paying on schedule right up until the refinancing stops. An analyst who classifies the financing structure (not just the profitability) sees the fragility that the income statement hides.

Worked Example 4: Surprises, not levels: a rate decision and a regime placement (US)

Operationalizes §4.8: why the surprise dominates the level, computed on a bond, plus placing the moment in the 2×2.

Part 1, the bond reacts to the surprise, not the hike. The market has fully priced a 50-bp hike at today's meeting. The central bank delivers only 25 bp and signals it is nearly done. A 10-year bond with modified duration 8 was priced for the 50-bp move already embedded in the yield. What happens to its price?

The key: the bond's price already reflects the expected 50-bp tightening path. The actual outcome, 25 bp and a dovish "nearly done", is a 25-bp dovish surprise relative to what was priced (the future path is now ~25 bp lower than embedded). Long yields fall by roughly that surprise. Approximate price change:

%ΔPrice ≈ −ModDuration × ΔYield = −8 × (−0.25%) = +2.0%.

The bond rises ~2% on a day the central bank hiked rates. To the beginner reading the level ("rates went up, bonds should fall"), this is baffling; to the professional reading the surprise ("they hiked less than priced and signalled the end — a dovish surprise"), it is obvious. The level was a hike; the surprise was dovish; the surprise won. This is the §4.8 principle made arithmetic, and the direct generalization of M7.02's "read the surprise, not the print."

Part 2, place the regime in the 2×2 and pick the winners. Suppose alongside this, incoming data show growth decelerating below expectations (payrolls and ISM softening past what was priced) while inflation is falling faster than expected (core CPI surprising down). Place the regime and name the likely asset winners.

• Growth: falling vs expected. Inflation: falling vs expected. → bottom-left quadrant: deflationary slowdown / recession-ward.
• Winners: long-duration government bonds (rates falling as growth and inflation both undershoot, consistent with Part 1's bond rally), quality-defensive equities (staples, health care, utilities), and cash. Losers: cyclicals, small-caps, high-yield credit, and commodities.
• The stance: this pairs exactly with the late-cycle → recession read of Worked Example 2: leading indicators rolled over, now the data are confirming the slowdown, inflation is receding, and the central bank is pivoting dovish. The regime says own duration and defensives, and critically, the bond rally is being driven by the downside growth and inflation surprises, i.e., by the world coming in worse (for growth) and cooler (for inflation) than was priced. Had the data merely matched the (already gloomy) consensus, the bonds would not have rallied. The move comes from the gap between realized and expected, never from the level everyone already owned.

Worked Example 5: Leverage amplifies the loss, and the drawdown-to-recovery trap (India)

*Operationalizes §4.7.1 and §4.10: why the leverage inside a position, not the size of the market fall, decides whether a correction wipes you out, and why the climb back is crueller than the fall.*

Setup (synthetic). Rhea (synthetic) puts up ₹5,00,000 of her own money and gears it 4× to hold a ₹20,00,000 equity book, funded by ₹15,00,000 of margin debt (the order of leverage an aggressive margin-trading/derivatives book can carry; 1929's stock buyers ran a far more extreme 10×, §4.7.1). The borrowed ₹15,00,000 is a fixed claim — owed in full whatever the market does.

Step 1, a mild market fall lands as a brutal equity loss. The index falls 15%. The book is now worth ₹20,00,000 × (1 − 0.15) = ₹17,00,000; the debt is still ₹15,00,000, so Rhea's equity is 17,00,000 − 15,00,000 = ₹2,00,000. Her loss on her own capital:

(5,00,000 − 2,00,000) ÷ 5,00,000 = 60%.

A 15% market fall became a 60% equity loss, amplified by exactly the leverage factor (60% ÷ 15% = 4×). Leverage does not change what the asset does; it multiplies what happens to your slice of it.

Step 2, the wipeout threshold. Equity hits zero when the book falls to the debt level (₹15,00,000), a fall of (20,00,000 − 15,00,000) ÷ 20,00,000 = 25%, which is precisely 1 ÷ leverage. A 4× book is wiped out by a 25% decline; 1929's 10× book was wiped out by just a 10% decline (1 ÷ 10), which is why a routine correction detonated into a margin-call cascade (§4.5). The leverage sets the size of the fall you cannot survive.

Step 3, the drawdown-to-recovery asymmetry. Rhea's equity is down 60% (₹5,00,000 → ₹2,00,000). To climb back to ₹5,00,000 she needs a gain of 3,00,000 ÷ 2,00,000 = 150%, not 60%. The general rule is recovery = d ÷ (1 − d) for a drawdown d (here 0.60 ÷ 0.40 = 1.5). The hole is always deeper to climb out of than to fall into, and the relationship is convex, getting disproportionately worse the deeper you go (a −80% drawdown needs +400% back; Practice P14). (Leverage's one mercy: because it cuts both ways, the underlying book need only rise 20,00,000 ÷ 17,00,000 − 1 = 17.6% to restore her equity, but that is a demanding, uncertain wait while carrying margin debt.)

Interpretation. This is §4.7.1's first lesson as arithmetic, leverage turns a correction into a collapse, fused with §4.10's discipline. The reason to reduce risk as euphoria and excess build is not that you can time the top (you can't); it is that the drawdown math is merciless and leverage makes it lethal. An unlevered ₹5,00,000 that fell 15% would be down ₹75,000 and need only ~17.6% to recover; the identical market move, geared 4×, costs 60% and demands 150% back. The margin of safety you surrender in the boom is exactly the survival you need in the bust.

Worked Example 6: What a P/E of 150 was actually pricing in (US)

Operationalizes §4.7.4: turn the dot-com valuation lesson into arithmetic; separate "the technology is real" from "the price is sane" by reverse-engineering the growth a multiple demands.

Setup. At the March 2000 peak, Cisco traded near a P/E of ~150 and was briefly the world's most valuable company at ~$550bn (public; check: these are the §4.7.4 figures). The internet displacement was entirely real. The question that would have saved you is not "is the technology real?" but "what does a P/E of 150 require to be true?" Work it on NovaSwitch (synthetic), a US networking name at the same P/E of 150, against a mature quality-compounder peer at a P/E of 25.

Step 1, the earnings yield says the price is almost pure growth. A P/E of 150 is an earnings yield of 1 ÷ 150 = 0.67%, against a US equity required return of ~8% (M0.01 / Phase 3). Today's earnings justify essentially none of the price; virtually 100% of the value is a bet on future growth that has not yet appeared. That is the definition of a valuation resting on narrative rather than fundamentals (§4.6, euphoria).

Step 2, how much growth, for how long? For the price merely to stand still while the stock is re-rated down to a normal P/E of 25, earnings must rise by 150 ÷ 25 = 6.0×. How many years of high growth is that?

(1 + g)^N = 6 → N = ln 6 ÷ ln(1 + g).

• At 25%/yr: N = 1.7918 ÷ 0.2231 = 8.0 years.
• At a scorching 35%/yr: N = 1.7918 ÷ 0.3001 = 6.0 years.
• At a merely-excellent 15%/yr: N = 1.7918 ÷ 0.1398 = 12.8 years.

So NovaSwitch must compound earnings at 25% for eight straight years just to make today's price look ordinary, with zero price appreciation to the buyer over that time. Any stumble, and the re-rating runs the other way, violently.

Step 3, the drawdown that followed (and the asymmetry). Cisco fell ~80% from the 2000 peak and, per §4.7.4, never durably reclaimed it two decades on. A −80% drawdown requires 0.80 ÷ 0.20 = +400% just to break even (the d ÷ (1 − d) rule of Worked Example 5); Amazon's >90% fall required +900%. Amazon the company went on to greatness. But Amazon the March-2000 stock took roughly a decade to break even.

Interpretation. This is the core investing lesson made numeric: a true, world-changing displacement can still be a ruinous investment, because valuation always matters (§4.7.4). Being right that the internet would change the world (correct) and buying it at a P/E of 150 (catastrophic) were two entirely different decisions, and only the second one touched your money. The reverse-DCF discipline (M3.05) is exactly Step 2 run in earnest: what future is the price demanding, and is it remotely achievable? When the honest answer is "eight years of 25% growth just to tread water," you are looking at stage-3 euphoria, and the right response is to reduce exposure, without any claim to know the month it ends (§4.10).

Practice set

Problems 1–4 are guided (framework applied with scaffolding), 5–9 independent, 10–12 exam-condition (12 minutes each, closed book). Write your full answer before reading the solution. Round sensibly.

P1 (guided, NBER dating). In year X, a country's real GDP falls in Q1 and Q2, but over the same six months nonfarm employment rises by 1.5 million, real incomes rise, and industrial production rises. (a) By the "two consecutive quarters" folk rule, is this a recession? (b) By the NBER's actual method, is it? (c) Name the three D's the NBER weighs and which one this episode fails. (d) Which real US episode does this describe?

Solution. (a) Yes by the folk rule: two negative GDP quarters. (b) No by the NBER method, because a recession is a significant, broad-based decline "spread across the economy," and here employment, income, and production all rose, so there is no diffusion of decline. (c) Depth, diffusion, duration. This episode fails diffusion (and arguably depth): the decline was confined to GDP accounting (trade/inventory quirks), not spread across jobs, income, and output. (d) The US in 2022 (negative real GDP in Q1 and Q2 2022, but millions of jobs added and no NBER recession), the canonical proof that the two-quarters rule is wrong.

P2 (guided, sort the indicators). Classify each as leading (L), coincident (C), or lagging (G): (a) the 10y–3m yield-curve spread; (b) the unemployment rate; (c) nonfarm payroll employment; (d) building permits; (e) core CPI inflation; (f) industrial production; (g) high-yield credit spreads; (h) unit labour costs.

Solution. (a) L: the yield curve (inversion leads recessions 6–18 months). (b) G: the unemployment rate lags (peaks after the recession ends). (c) C: payrolls are the marquee coincident series. (d) L: permits lead (housing is rate-sensitive and turns first). (e) G: inflation lags the cycle (peaks after the economy peaks). (f) C: industrial production is coincident. (g) L: credit spreads widen before defaults arrive. (h) G: unit labour costs lag (wages are sticky and negotiated late).

P3 (guided, compute a PMI). A manufacturing survey returns sub-indices: New Orders 58, Production 55, Employment 52, Supplier Deliveries 50, Inventories 45. (a) Compute the headline PMI. (b) Is the sector expanding or contracting? (c) For the New Orders question specifically, if 30% of firms said "same" and the rest split between higher and lower to produce the 58 reading, what percent said "higher"?

Solution. (a) PMI = 0.20 × (58 + 55 + 52 + 50 + 45) = 0.20 × 260 = 52.0. (b) 52.0 > 50 → expanding (mildly; distance from 50 is modest). (c) Diffusion index = %higher + 0.5 × %same, so 58 = %higher + 0.5 × 30 = %higher + 15, giving %higher = 43% (and %lower = 100 − 43 − 30 = 27%). More firms saw rising orders (43%) than falling (27%), consistent with the sub-50-pivot logic.

P4 (guided, classify the financing). For each, state hedge, speculative, or Ponzi, and the one-line reason: (a) a profitable company whose operating cash flow covers interest 4× and also funds scheduled principal repayments; (b) a bank funding 10-year loans with overnight deposits it rolls continuously; (c) a property speculator whose rental income doesn't even cover the loan interest, who plans to sell the flat at a higher price to repay; (d) a firm covering its interest comfortably but with a bullet principal it must refinance at maturity.

Solution. (a) Hedge: cash flow covers both interest and principal from income; no dependence on refinancing. (b) Speculative: covers interest, but the principal (deposits) must be continually rolled; solvent but liquidity-dependent (the structural condition of all banks). (c) Ponzi: covers neither interest nor principal from income; survives only on rising asset prices (sell higher to repay). (d) Speculative: interest covered, but the principal cannot be repaid from income and must be refinanced; fine until rates rise or credit freezes at the refinancing wall.

P5 (independent, the Minsky moment). Explain, in your own words and in four sentences, (a) why Minsky says "stability is destabilizing," (b) what specifically happens at the Minsky moment, and (c) why the crash then feeds on itself (name the mechanism).

Solution. (a) In a long calm, defaults are rare, so lenders and borrowers grow confident and accept ever more leverage and thinner margins of safety, and the stability rewards boldness and punishes caution, so the financing mix migrates from hedge toward speculative and Ponzi, building fragility invisibly. (b) At the Minsky moment, a trigger (a rate rise, a price that stops rising, a prominent default) means the Ponzi units, who needed rising prices or easy refinancing, can no longer refinance and are forced into distress asset sales. (c) The crash feeds on itself through debt-deflation (Fisher): forced selling lowers asset prices, which pushes the next leveraged holder underwater and forces their sale, while the fixed nominal debt gets heavier in real terms as prices fall ("the more the debtors pay, the more they owe"): a self-reinforcing downward spiral.

P6 (independent, map a case to the anatomy). Map the 2008 GFC onto the five Kindleberger stages, one crisp line per stage (displacement, boom, euphoria, distress, revulsion). Then state the single most important structural lesson.

Solution. Displacement: low post-dot-com rates + a global savings glut + the belief that US house prices never fall nationally + the innovation of securitization. Boom: the credit chain, from originate-to-distribute mortgages → MBS → CDOs → AAA ratings → sold globally, funded short-term in shadow banking (repo/ABCP). Euphoria: subprime and Ponzi "liar loans," ~30–40× investment-bank leverage, "house prices only go up," valuation replaced by the AAA stamp. Distress: house prices peak in 2006, subprime defaults rise, CDO marks fall, Bear Stearns rescued (Mar 2008). Revulsion/panic: Lehman fails (Sept 2008), the shadow-banking funding runs, AIG bailed out, credit freezes globally, stopped only by a massive lender-of-last-resort response (TARP, QE, swap lines). Structural lesson: securitization did not eliminate risk, it hid, concentrated, and lengthened it; "AAA" is worthless if the underlying is Ponzi; and shadow banking is banking without the safety net, so a bank-run can happen in wholesale funding markets.

P7 (independent, the growth/inflation 2×2). For each regime, name the quadrant label and the asset class(es) that tend to win: (a) growth rising, inflation falling; (b) growth falling, inflation rising; (c) growth falling, inflation falling; (d) growth rising, inflation rising. (e) Which quadrant destroys a classic 60/40 stock-bond portfolio, and why?

Solution. (a) Goldilocks: equities (especially growth/tech) and corporate credit win. (b) Stagflation: commodities, gold, and TIPS win; both equities and nominal bonds lose. (c) Deflationary slowdown / recession: long-duration government bonds win; cash and quality-defensive equities hold up. (d) Reflation / overheating: commodities, real assets, EM equities, cyclical/value, and TIPS win; nominal bonds struggle. (e) Stagflation (b) destroys 60/40 because both legs fall at once: stocks fall on the growth hit and rising discount rate, and nominal bonds fall on the inflation/rate surprise, so the bonds provide no offset to the stocks (exactly the 2022 experience, the worst 60/40 year in a century).

P8 (independent, surprise vs level). The market has priced a 25-bp cut. The central bank instead holds (no change) and signals it is worried about inflation. (a) Is this a hawkish or dovish surprise? (b) Which way do bond yields move, and why? (c) A 7-year bond has modified duration 6; if the surprise pushes its yield up 20 bp, estimate the price change. (d) State the general principle in one line.

Solution. (a) Hawkish surprise: the market expected easing (a cut) and got no easing plus inflation-worried language; relative to what was priced, that is a hawkish shift. (b) Yields rise: the expected future policy path is now higher than embedded, so bond prices fall / yields rise, even though the central bank did nothing to the level (a hold). (c) %ΔPrice ≈ −ModDuration × ΔYield = −6 × (+0.20%) = −1.2%, so the bond falls ~1.2%. (d) Assets are priced off the surprise relative to expectations, not the level: a "no change" can move markets sharply if a change was priced, and the direction is set by actual minus expected.

P9 (independent, India's cycle). (a) Name three forces that have historically driven India's business cycle more than an endogenous private-credit boom does in the US. (b) Explain why an oil-price spike is a particularly Indian macro shock. (c) Identify India's clearest "Minsky moment" and the specific financing structure that caused it.

Solution. (a) Any three of: the monsoon/agriculture (rural demand and food inflation), oil and the current-account/rupee channel (India imports ~85% of its crude), global capital flows / FPI (the global financial cycle; taper-tantrum vulnerability), and the government investment/capex cycle (and PSU-bank-driven credit). (b) India imports ~85% of its oil, so an oil spike is a terms-of-trade shock that simultaneously widens the current-account deficit, weakens the rupee, and raises inflation, forcing the RBI tighter even into a slowdown, a stagflation-flavoured dynamic with no US analogue (the US is roughly oil-neutral). (c) The 2018 IL&FS default (and DHFL in 2019): NBFCs funding long-dated infrastructure loans and mortgages with short-term commercial paper they rolled continuously (a maturity mismatch, speculative-to-Ponzi); when IL&FS defaulted, the market refused to roll anyone's paper, a domestic sudden stop that froze NBFC credit.

P10 (exam-condition, the full cycle read). You are handed this panel: yield curve inverted 10 months; building permits down 15% y/y; ISM new orders 44; high-yield spreads widening; payrolls +90k and slowing; unemployment 4.1%, up from 3.7%; core inflation 3.8%, easing from 5%; the equity market making new highs. (a) Sort into leading/coincident/lagging. (b) Name the cycle stage. (c) Identify the trap in this panel and how a professional avoids it. (d) State one portfolio tilt and one thing you would not claim.

Solution. (a) Leading: inverted curve, permits −15%, ISM new orders 44, spreads widening, equity market (leading but noisy, here diverging from the other four). Coincident: payrolls +90k (positive but slowing), and activity implied soft. Lagging: unemployment (4.1%, turning up from its low), core inflation (3.8%, easing from the peak). (b) Late cycle / peak passing, elevated recession risk over the next 6–18 months (post-inversion window). (c) The trap: the equity market at new highs and the still-positive payrolls/low-4% unemployment make it feel like a healthy expansion, while four of five leading indicators have decisively rolled over. This is the classic late-cycle configuration where the watched (coincident/lagging) numbers look fine and the forward numbers have already turned. The pro weights the leading indicators over the reassuring coincident/lagging data and notes that the equity high is a lone, noisy leading signal contradicted by permits, orders, spreads, and the curve. (d) Tilt: rotate toward defensives (staples/health care/utilities) and duration, trim cyclicals/high-yield, raise the bar for adding risk. Would NOT claim: that "the recession begins in month X". The turn is undatable in real time; you tilt risk, you don't call the day (§4.10).

P11 (exam-condition, diagnose the crisis type and the right response). Three shocks: (i) an economy where inflation has run at 12% for years, wage-price spiral entrenched, expectations un-anchored; (ii) a leveraged property-and-stock bubble bursts and the corporate sector, insolvent on assets but solvent on cash flow, refuses to borrow at any interest rate; (iii) a pandemic shuts the economy exogenously. For each, name the historical archetype and the correct policy response, and explain why the wrong response would be to treat it like the others.

Solution. (i) 1970s stagflation: the cure is a deliberate, credibility-restoring tightening/recession (Volcker: raise rates hard, break inflation expectations, accept the recession). Treating it like a demand slump (easing) would entrench the inflation, as Burns did. (ii) Japan's balance-sheet recession: the private sector deleverages regardless of rate cuts ("pushing on a string"), so monetary policy is nearly impotent and fiscal policy must fill the demand hole, while forcing loss recognition to avoid zombies. Treating it like an inflation problem (tightening) or expecting rate cuts to revive borrowing would deepen and prolong the slump. (iii) 2020 COVID exogenous shock: a fast, overwhelming monetary + fiscal response to bridge the shutdown; because the cause is exogenous (not a credit excess), activity rebounds once the economy reopens. Treating it as a normal credit-cycle recession (a slow, cautious response) would risk turning a temporary shutdown into a lasting cascade. But over-responding has a cost (the 2021–23 inflation), so the response must be sized to bridge, not to over-stimulate a supply-constrained economy. The meta-lesson: diagnose the nature of the shock (inflation regime vs balance-sheet/deflation vs exogenous) before choosing the response, because the 1970s and 2008 required opposite medicines, and giving one the other's cure is catastrophic.

P12 (exam-condition, the honesty question). A junior analyst says: "The yield curve is inverted, valuations are euphoric, and Minsky says we're at the top — so I've moved the whole portfolio to cash and shorted the index." Critique this in four sentences, invoking the core discipline above.

Solution. The analyst has correctly recognized the regime and the excess (inverted curve, euphoria, late-cycle fragility) but has then committed the cardinal error: treating a regime read as a market-timing call. Euphoria and inversions can persist for many months or years ("the market can stay irrational longer than you can stay solvent"), the exact trigger and turn are unforecastable and only datable in hindsight, and being early, especially with an outright short, is financially indistinguishable from being wrong until you are ruined. The correct action is to tilt risk down in proportion to the excess (raise cash somewhat, rotate toward quality/defensives, reduce leverage and the most-stretched positions), not to go all-cash-and-short on a date the frameworks explicitly cannot supply. Marks's discipline governs: you can't predict, you can prepare, so prepare by reducing risk as excess builds, without pretending to know the day it ends.

P13 (guided, leverage amplifies losses). In 1929, stocks could be bought on 10% margin. (a) What leverage is that, and what size of market fall wipes out the buyer's entire equity? (b) A speculator puts $1,000 of own money into a 10×-levered position ($10,000 of stock, $9,000 borrowed); compute the loss on their equity after market falls of 8% and 12%. (c) What does the 12% case reveal that a simple "you can only lose 100%" intuition misses?

Solution. (a) Leverage = 1 ÷ 0.10 = 10×; equity is wiped out by a fall of 1 ÷ 10 = 10% (the position drops to the $9,000 owed). (b) 8% fall: position = 10,000 × 0.92 = $9,200; equity = 9,200 − 9,000 = $200; loss = (1,000 − 200) ÷ 1,000 = 80%, an 8% market move, 10×'d. 12% fall: position = $8,800; equity = 8,800 − 9,000 = −$200; loss = (1,000 − (−200)) ÷ 1,000 = 120%. (c) At −12% the buyer has lost their entire $1,000 and owes a further $200: negative equity, which triggers a margin call for more cash; unable to pay, they are sold out, and that forced sale pushes the price onto the next 10× buyer (§4.5's cascade). Leverage means you can lose more than 100%, and that the losses of the over-levered become everyone else's trigger.

P14 (guided, the drawdown-to-recovery math). Using recovery gain = d ÷ (1 − d), compute the percentage gain needed to break even from each peak-to-trough drawdown (all public): (a) the Nasdaq's −78% (dot-com, 2000–02); (b) the S&P 500's −57% (2007–09 GFC); (c) the S&P's −34% (COVID, 2020); (d) the Nikkei's −80% (post-1989). (e) A synthetic Indian small-cap falls −75%. What does it need? (f) What does comparing (c) with (d) tell you about why avoiding deep drawdowns matters more than squeezing the last of a rally?

Solution. (a) 0.78 ÷ 0.22 = +354.5%. (b) 0.57 ÷ 0.43 = +132.6%. (c) 0.34 ÷ 0.66 = +51.5%. (d) 0.80 ÷ 0.20 = +400%. (e) 0.75 ÷ 0.25 = +300%. (f) The relationship is convex: a −34% fall needs only +52% back, but a −80% fall needs +400%, more than seven times the climb for roughly 2.4× the fall. Losses compound against you disproportionately, so staying out of the deep hole (by reducing risk as excess builds, §4.10) does far more for long-run compounding than chasing the final gains of a euphoric top. This is the arithmetic beneath Marks's "you can't predict, you can prepare."

P15 (independent, reverse the multiple). Bharat Digital (synthetic), an Indian consumer-tech name, trades at a P/E of 80; a mature quality peer trades at a P/E of 20; the Indian equity hurdle is ~12% (M0.01). (a) Compute Bharat Digital's earnings yield and say what it implies about how much of the price is "today" versus "the future." (b) For the price to hold while the stock re-rates to a normal P/E of 20, by what multiple must earnings grow? (c) How many years is that at 30%/yr and at 20%/yr? (d) State, in one sentence, the risk this valuation creates.

Solution. (a) Earnings yield = 1 ÷ 80 = 1.25%, barely a ninth of the 12% hurdle, so almost the entire price is a claim on future growth, not current earnings. (b) 80 ÷ 20 = 4.0×. (c) N = ln 4 ÷ ln(1 + g): at 30%, 1.3863 ÷ 0.2624 = 5.3 years; at 20%, 1.3863 ÷ 0.1823 = 7.6 years. (d) Because the value is almost all future growth priced years out, even a small disappointment in the growth rate re-rates the stock violently downward: the multiple that expanded on narrative contracts on the first miss (the dot-com and K-10 experience, §4.7.4); being enthusiastic about the sector is not the same as the price being survivable.

P16 (independent, the refinancing wall). Harbor Tower (synthetic) is a US office building with net operating income (NOI) of $10m/yr, flat, financed by a $150m interest-only loan whose principal falls due as a bullet at maturity. (a) At the original rate of 4.0%, compute the interest, the interest-coverage ratio (ICR = NOI ÷ interest), and classify the financing (hedge / speculative / Ponzi). (b) At maturity it must refinance, and rates have risen to 7.5%. Recompute interest and ICR, and reclassify. (c) By what percent did annual debt-service grow, and what actually breaks the building, its profitability or its financing structure?

Solution. (a) Interest = 150 × 0.040 = $6.0m; ICR = 10 ÷ 6 = 1.67×. Speculative: it covers interest comfortably but cannot repay the $150m principal from income; the bullet must be refinanced (§4.5). (b) Interest = 150 × 0.075 = $11.25m; ICR = 10 ÷ 11.25 = 0.89× (< 1). Ponzi: NOI no longer even covers the interest, so it survives only by borrowing more or selling the asset. (c) Debt-service grew (11.25 − 6.0) ÷ 6.0 = +87.5% while NOI was flat. The building was never unprofitable; its financing structure killed it: a solvent, interest-covering asset became unfinanceable purely because it depended on refinancing into a higher-rate, tighter-credit market. That is the exact mechanism that turns commercial real estate into a slow-motion crisis when rates rise into a refi wall (Worked Example 3's Unit B).

P17 (independent, an NBFC crosses the Minsky line). Setu Finance (synthetic), an Indian NBFC, holds a ₹1,000 crore book of 12-year infrastructure loans, funded entirely by 90-day commercial paper it rolls continuously. Its operating cash flow before funding costs is ₹95 crore/yr. (a) At a CP rate of 8%, compute its interest bill and interest coverage, and classify it. (b) After an IL&FS-type shock it can only roll the paper at 12%. Recompute and reclassify. (c) Beyond the coverage ratio, what is the deeper danger in this structure, and which real 2018 episode is it?

Solution. (a) Interest = 1,000 × 0.08 = ₹80 cr; ICR = 95 ÷ 80 = 1.19×. Speculative: it covers interest but must perpetually roll the entire ₹1,000 cr of principal (a pure maturity mismatch: 12-year assets, 90-day funding). (b) Interest = 1,000 × 0.12 = ₹120 cr; ICR = 95 ÷ 120 = 0.79× (< 1). Ponzi: cash flow no longer covers even the interest. (c) The coverage ratio is not even the real killer: the structure depends on refinancing ₹1,000 cr of principal every 90 days, so if the market simply refuses to roll the paper at any rate (a domestic sudden stop), the NBFC defaults instantly: liquidity death, not slow insolvency. This is the IL&FS default of September 2018 (with DHFL following in 2019), India's clearest Minsky moment (§4.9): a decade of calm bred the mismatch, and one default froze the rollover the whole sector depended on.

P18 (timed, 12 minutes, all six parts). (i) A position bought on 20% margin carries what leverage, and what market fall wipes out the equity? (ii) A ₹-denominated position geared 5× falls 8%. What is the loss on the investor's equity? (iii) A stock is down 90% from its peak. What gain is needed to break even? (iv) A US tech stock trades at a P/E of 100 while a mature peer trades at 25; by what multiple must earnings grow for the price to be "normal," and roughly how many years is that at 30%/yr? (v) A firm covers its loan interest 1.5× but cannot repay the principal from income (it must refinance). Minsky class? (vi) Rates then rise so its interest bill doubles and coverage becomes 0.75×. New Minsky class?

Solution. (i) Leverage = 1 ÷ 0.20 = 5×; wiped out by a 20% fall (1 ÷ 5). (ii) 5 × 8% = 40% equity loss (exactly, while equity stays positive). (iii) 0.90 ÷ 0.10 = +900%. (iv) 100 ÷ 25 = 4.0×; N = ln 4 ÷ ln 1.30 = 5.3 years. (v) Speculative: interest covered, principal must be rolled. (vi) 1.5 ÷ 2 = 0.75× < 1 → Ponzi: income no longer covers even interest, so it survives only on new borrowing or asset sales. Scoring: 6/6 in time = fluent with the leverage, drawdown, valuation, and financing arithmetic; ≤4 → redo Worked Examples 5–6 and P13–P17 before the mastery check.

P19 (independent, place the prescription). Each statement below is one school's answer to a recession. Name the school (neoclassical or real business cycle, Keynesian, monetarist, Austrian, or Minsky and the credit cycle), give the causal claim it rests on, and state the strongest objection to it. (a) "The central bank let nominal spending collapse. Had it kept the money supply growing on a stated rule, this would have been an ordinary quarter." (b) "The boom misdirected capital into projects that were never viable at an undistorted interest rate. Let them fail; propping them up only defers the reckoning." (c) "Private demand has gone and prices are not clearing it. The government should borrow and spend until it comes back." (d) "The slump is the efficient response to a genuine technology and energy shock. There is nothing to stabilise; work on the supply side." (e) "Nothing shocked the system from outside. Across a decade of calm the financing mix drifted out of hedge structures and into speculative and then Ponzi ones, and a single repricing exposed all of it at once."

Solution. (a) Monetarist, resting on the claim that a downturn is central bank error rather than an inevitable event; the objection is that the money-to-spending relationship broke down after the 1980s, so the rule lost its target. (b) Austrian, resting on malinvestment caused by credit expanded below the natural rate; the objection is the concentrated human cost, plus Fisher's debt-deflation showing that liquidation can overshoot and destroy solvent firms as well (§4.5). (c) Keynesian, resting on demand collapse plus sticky wages and prices; the objection is timing and political economy, since stimulus lands late and is rarely withdrawn. (d) Neoclassical or real business cycle, resting on real shocks and an optimal response; the objection is involuntary unemployment, which an optimal response should not produce. (e) Minsky and the credit cycle, resting on endogenous fragility; the objection is that it dates crises poorly in advance and calls tops that do not arrive. Then the meta-point, which is the item's real target: (a), (c) and (e) all support acting in a crisis and disagree about who acts and when, while (b) and (d) question whether acting helps at all. Reading a policy debate is mostly a matter of hearing which of those five claims a speaker has already assumed.

Applied mini-project: place this moment in the cycle and write a one-page regime note

Goal. Produce the deliverable everything above exists to enable: a dated, evidence-based regime read of the economy right now (do one for the US and one for India), written the way a professional would, from primary data, read as rate-of-change versus expectations, with an explicit stance and an explicit statement of what you are not claiming. This is a dry run of the M7.07 dashboard capstone and a permanent journal habit.

Data to pull (all free; snapshot the values and the date).

• US leading: the 10y–3m and 10y–2y yield-curve spreads (FRED: T10Y3M, T10Y2Y); building permits (FRED: PERMIT); the ISM Manufacturing PMI and its New Orders sub-index (ismworld.org / news reports); high-yield credit spreads (FRED: BAMLH0A0HYM2); the S&P 500 trend; and the Conference Board LEI (conference-board.org).
• US coincident/lagging: nonfarm payrolls and the unemployment rate (FRED: PAYEMS, UNRATE); industrial production (INDPRO); core PCE / core CPI (PCEPILFE, CPILFESL).
• India: the S&P Global India Manufacturing and Services PMI; IIP and GST collections (mospi.gov.in; gst.gov.in); CPI (mospi.gov.in); bank and NBFC credit growth, the current-account balance, FPI flows, and FX reserves (RBI DBIE, rbi.org.in); the 10-year G-sec yield and USD/INR.

Your tasks.

1. Sort and read (each economy). Put every indicator in its leading/coincident/lagging bucket, note each as rising/falling and better/worse than expected, and write the cycle stage (early/mid/late expansion; peak; contraction; trough) with the pattern-evidence for it (mirror Worked Example 2 / P10).
2. Place the growth/inflation regime. State which of the four 2×2 quadrants each economy is in (growth rising/falling, inflation rising/falling, versus expectations), and name the asset classes the regime favours and disfavours (§4.8).
3. Read the credit cycle / excess. Are spreads tight or wide, standards loose or tight, credit growth fast or slow? Is there any visible euphoria or Ponzi-finance build-up (a hot sector, stretched valuations, a leverage story)? Where on the Kindleberger anatomy, if anywhere, does the current environment sit? (For India, look specifically at NBFC/retail credit growth and the oil/CAD position.)
4. State a stance and a disclaimer. One paragraph: given the regime and excess, how would you tilt risk (sector/asset tilts, cash level, leverage), and one explicit sentence on what you are not claiming (you are not calling a turn or a date).
5. File it dated in your decision journal, so you can re-grade the call against what actually happened — the only way calibration ever improves.

Deliverable. Two one-page regime notes (US and India), dated, each ending in a stance and a disclaimer. Keep them; the M7.07 capstone turns this into a standing dashboard.

Scoring rubric (0/1/2 per criterion; 16 = excellent, 12 = pass).

Case Lab: run case XL-7.2 (2008, From the Record) in the app's Cases tab for this module. Walk the Global Financial Crisis link by link from the public record (the §4.7.5 credit chain: subprime/Ponzi mortgages → MBS → CDO → AAA stamp → repo / shadow-banking run), map each stage onto the Kindleberger anatomy (§4.6), and deliver the verdict on which structural feature was most load-bearing (the leverage, the maturity/liquidity mismatch, or the ratings fiction), answering our worksheet questions against the record.

Reading & resources

• Charles Kindleberger & Robert Aliber, Manias, Panics, and Crashes: the source of the bubble anatomy in §4.6, and the best single book on the history and pattern of financial crises. Read it as the master template. [Paid] [Intermediate]
• Hyman Minsky, Stabilizing an Unstable Economy (and the short 1992 paper "The Financial Instability Hypothesis", free online): the origin of the hedge/speculative/Ponzi taxonomy and "stability is destabilizing." The 1992 paper is short and worth reading in the original; the book is denser. [Free paper / Paid book] [Intermediate→Advanced]
• Ray Dalio, How the Economic Machine Works (free 30-min video + essay) and Principles for Navigating Big Debt Crises (free PDF from his site): the debt-cycle framing behind the whole module and the template case-studies (his "archetypal" big debt cycle maps directly onto §4.5–4.7). [Free] [Beginner→Intermediate]
• Howard Marks, Mastering the Market Cycle and his Oaktree memos (free at oaktreecapital.com). The definitive treatment of §4.10: "you can't predict, you can prepare," taking the market's temperature, and the pendulum of sentiment. The single best complement to this module's honesty section. [Paid book / Free memos] [Beginner→Intermediate]
• Richard Koo, The Holy Grail of Macroeconomics: Lessons from Japan's Great Recession: the balance-sheet-recession thesis (§4.7.3) in full; explains why monetary policy is impotent after a leveraged bubble bursts. Essential for the Japan case. [Paid] [Advanced]
• Milton Friedman & Anna Schwartz, A Monetary History of the United States (the Great Contraction chapter): the policy-failure thesis of 1929–33 (§4.7.1); read at least the standalone The Great Contraction excerpt. [Paid] [Advanced]
• Adam Tooze, Crashed: the definitive narrative history of 2008 and its decade-long aftermath, including the global and shadow-banking plumbing (§4.7.5). [Paid] [Intermediate]
• Liaquat Ahamed, Lords of Finance: the central bankers of the 1920s–30s and the road to the Depression; makes 1929 and the gold-standard policy trap vivid. [Paid] [Beginner→Intermediate]
• Carmen Reinhart & Kenneth Rogoff, This Time Is Different: Eight Centuries of Financial Folly: the data behind "cycles rhyme" (§4.10); crises are recurrent and the "this time is different" refrain is the tell. [Paid] [Intermediate]
• The NBER Business Cycle Dating Committee page (nber.org/cycles): the primary source for the US recession chronology and the actual definition; read the FAQ on why not two quarters of GDP (§4.2). [Free] [Beginner]
• The Conference Board: Leading Economic Index methodology and releases (conference-board.org). The ten LEI components and how they're combined; the reference for §4.3. [Free] [Intermediate]
• ISM (ismworld.org) and S&P Global PMI (pmi.spglobal.com): the PMI methodology documents (the diffusion-index construction of §4.4) and the monthly releases for the US and India. [Free] [Beginner→Intermediate]
• FRED (fred.stlouisfed.org). Plot the cycle yourself: T10Y3M/T10Y2Y (yield curve), PERMIT (permits), BAMLH0A0HYM2 (HY spreads), USREC (NBER recession shading), PAYEMS/UNRATE, INDPRO, PCEPILFE. Overlay the recession bars and see which indicators led. [Free] [Beginner]
• RBI DBIE (dbie.rbi.org.in) and MOSPI (mospi.gov.in) hold India's primary data: IIP, CPI, credit growth, current account, FX reserves, FPI flows. This is the India half of the mini-project and the M7.07 dashboard. [Free] [Intermediate]
• Aswath Damodaran, Narrative and Numbers (and his blog posts on bubbles): the discipline that would have saved you in 2000 (§4.7.4): every story turned into a number, every price reverse-engineered for what it implies. [Paid book / Free blog] [Intermediate]
• Podcasts: Odd Lots (Bloomberg) for live cycle/credit debates, and We Study Billionaires on Marks and cycles, for hearing practitioners reason about where we are in plain language. [Free] [Intermediate]
• Going deeper: Antti Ilmanen, Expected Returns (2011), the chapters on drawdowns, leverage and the "volatility tax". The rigorous, quant's-eye treatment of the drawdown-to-recovery asymmetry and the compounding drag that leverage magnifies, which Worked Example 5 and Practice P13–P14 compute by hand; the formal case for why surviving the deep hole beats chasing the last of the rally. [Paid] [Advanced]
• Going deeper: Aswath Damodaran, "Pricing vs Value" posts + his dot-com-era valuation archives (aswathdamodaran.blogspot.com; data at pages.stern.nyu.edu). The reverse-multiple discipline of Worked Example 6 and P15 in the master's own hand: reverse-engineer the growth a P/E of 80 or 150 silently demands, then test it against what any business can plausibly deliver, the habit that separated "the internet is real" from "Cisco at 150× is sane" in 2000. [Free] [Intermediate]

The Modern Analyst's Addendum

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

AI-Augment this skill

{
  "skill": "Dating the cycle by breadth rather than by the two-quarter rule, the leading/coincident/lagging taxonomy, PMIs and diffusion indices, Minsky's financing structures and the credit cycle, the Kindleberger five-stage anatomy, seven historical cases, the growth/inflation 2x2, India's cycle character, and the discipline of reading regime rather than timing turns",
  "use": "The module's central skill is pattern recognition across cases, which is a thing a language model is genuinely good at and a thing it is genuinely dangerous at, for the same reason: it will always find the pattern. Three uses that hold up. (1) ANATOMY MAPPING AS A CHALLENGE, not a confirmation. Give it a situation and ask it to argue BOTH that the five-stage template fits and that it does not, naming the specific evidence each way, then say which stage-marker is missing. §4.6 is explicit that the costume always changes, and a tool asked 'does this look like a bubble' will oblige; a tool asked 'what would have to be true for this NOT to be one' does work. (2) FINANCING-STRUCTURE CLASSIFICATION, which is the module's most examined idea and is mechanical: paste a company's cash flow, interest, amortisation and maturity schedule and ask for the hedge/speculative/Ponzi classification WITH the coverage ratios shown. The ratios make it checkable; the label alone does not. (3) CASE RECALL AS A STARTING POINT — 'what were the displacement, the excess, the trigger and the policy response in the 1997 Asian crisis' gets you a serviceable skeleton to verify against Kindleberger and the primary accounts. What it must never do: tell you where in the cycle we are now. That requires today's leading indicators, and every one of them is a dated series you fetch. A model asked for a cycle read will produce a fluent, confident stage placement built on whatever was in its training data, and stage placement is exactly the judgement this module says is hard.",
  "tools": ["Chat assistants — Claude, ChatGPT for adversarial anatomy mapping, financing-structure classification with ratios shown, and historical case skeletons", "Coding assistants — for the indicator panel, the diffusion-index arithmetic and the base-rate engine below", "Document/PDF AI — for the RBI Financial Stability Report, BIS Annual Economic Report early-warning tables, and IMF Article IV financial-sector assessments"],
  "prompt": "Here is a financing situation: annual operating cash flow, interest expense, scheduled amortisation, and the maturity schedule of the principal. Classify it as hedge, speculative or Ponzi under Minsky's test, and show the coverage ratios you used for each leg — interest coverage and total-debt-service coverage — so I can check them. Then state the single change (a refinancing rate, a cash-flow shortfall, a maturity date) that would move it down one category, and the exact threshold at which it does. Do not tell me anything about the current state of the credit cycle.",
  "verify": "Two checks, and the second is the one this module exists for. (1) Every classification must come with its ratios, and every ratio must be recomputable from the inputs you supplied. Worked Example 3's units are 2.76x total-service coverage, 1.11x interest coverage and a structure that covers neither — those numbers, not the labels, are the analysis. (2) Every claim that an indicator 'predicts' something must come with its base rate, its false-alarm rate, and the horizon over which it was evaluated. The quantitative lens below shows why: with a rare event, a signal can catch most crises, cry wolf only occasionally, and still be wrong the large majority of the times it fires. Any tool — or any human — that tells you an indicator 'has predicted every recession' without those three numbers has given you a headline, not a finding. And hold §4.10's discipline over the whole exercise: a correct regime read that is early is indistinguishable from a wrong one until much later, so nothing here is scored on the next quarter.",
  "diy": "The gate is unaided: you compute a diffusion index from raw survey responses and a headline PMI from its sub-indices, sort a mixed panel into leading, coincident and lagging and name the stage, classify a financing structure as hedge, speculative or Ponzi from its cash flows, map a case onto the five-stage anatomy, compute a levered equity loss and the recovery it demands, and reverse-engineer the growth a multiple requires. Nothing in this callout is admissible in that room."
}

Modern Data Analysis

By hand first. You built a headline PMI from five sub-indices and read the composition rather than the number; you sorted a mixed panel into three families and named a late-cycle configuration from the disagreement between them; you classified three financing structures on their coverage ratios; you turned a 15% market fall into a 60% equity loss and worked out the 150% it demands back; you reverse-engineered eight years of 25% growth out of a P/E of 150. Keep every one, and the panel-sorting most of all — it is a judgement about which indicators lead and why, and no pipeline supplies that.

Today's workflow. Three changes. First, the panel is a table with a rule attached. The §4.3 taxonomy becomes a small dataframe (series id, family, direction that counts as deterioration, and the transformation applied), so a cycle read is reproducible and, more importantly, auditable against your last one: when you say the picture has changed, you can show which cell changed. The hardest discipline here is the one §4.3's own list implies and never states, which is that the transformation must be fixed in advance. An indicator read as a level this month and a rate of change next month is not an indicator, it is a mood. Second, and this is the real modern content here, any signal you propose gets scored properly before you use it. An early-warning indicator is a rare-event classifier, and rare-event classifiers have a mathematics that is unkind and completely explicit: sensitivity, specificity, base rate, precision. The BIS publishes credit-to-GDP gaps and debt-service ratios quarterly and the crisis-dating literature publishes dated chronologies, so the four numbers are computable rather than rhetorical, and the quantitative lens below shows what they do to a confident-sounding claim. Third, surprise beats level, mechanically. §4.8 and Worked Example 4 both turn on the gap between realised and expected, so the panel carries a consensus column and the read is built on the residual. Where consensus is not free, and in India it usually is not, say so rather than substituting the level.

Tools & sources (IN + US). pandas for the panel, numpy for the diffusion and base-rate arithmetic, matplotlib for the three-family chart; US, FRED by id for USREC (the NBER recession indicator, which is what makes episodes countable rather than remembered), T10Y3M and T10Y2Y (the §4.3 premier leading indicator), PERMIT (building permits), NEWORDER (new orders for nondefense capital goods excluding aircraft, the core-capex leading series), ICSA (initial jobless claims), BAMLH0A0HYM2 and BAMLC0A0CM (ICE BofA high-yield and investment-grade option-adjusted spreads, the credit-cycle indicator §4.5 calls premier), DRTSCILM (the SLOOS net percentage tightening standards), PAYEMS, INDPRO, UNRATE, CPILFESL; the BIS statistics portal for credit-to-GDP gaps, debt service ratios for the private non-financial sector and for households, and total credit series, free, quarterly, covering both India and the US on a documented, comparable methodology, and the closest thing that exists to a published early-warning dataset; the Laeven–Valencia Systemic Banking Crises Database (IMF, free) and Reinhart–Rogoff's dataset for the dated crisis chronology you need to compute a base rate at all; India, where MOSPI covers IIP and quarterly GDP, the RBI Financial Stability Report (bi-annual, free) carries the banking stability indicator, the systemic risk survey, stress-test results and GNPA projections: India's credit-cycle dashboard, already assembled. Then there are the RBI Monetary Policy Report, RBI's OBICUS survey for capacity utilisation, and RBI's Industrial Outlook Survey and Consumer Confidence Survey for sentiment, all free, all on a published calendar; FBIL and CCIL for benchmark corporate-bond yields from which an AAA-over-G-sec spread can be built. Now the paid line, and here it bites harder than anywhere else in the phase. The flagship leading indicator of §4.4 is licensed on both sides: ISM publishes the current month's PMI in a free press release and sells the history, and S&P Global does the same for the India Manufacturing and Services PMI, so the number is free and the series is not, which matters because §4.4's own instruction is to read a PMI against its own history. The Conference Board's Leading Economic Index is likewise a licensed product. And the sharpest asymmetry in the phase: India has no official monthly unemployment rate. The PLFS is quarterly for urban areas and annual overall; the monthly figure everyone quotes is CMIE's Consumer Pyramids, a paid product with no free equivalent, where the US gets UNRATE monthly for nothing. India also has no liquid high-yield bond index comparable to BAMLH0A0HYM2, because the market below AA is thin, so the single best credit-cycle indicator in the US toolkit has no Indian counterpart at all. Build the India panel around what RBI publishes free and say plainly which US indicator you are doing without.

# Rare-event signal arithmetic, leverage, and the diffusion index — recomputed in-session (R3)
def precision(base, sens, spec):                 # PPV, and false alarms per true call
    tp, fp = base * sens, (1 - base) * (1 - spec)
    return tp / (tp + fp), fp / tp
for b, s, sp in ((0.05, 0.80, 0.80), (0.05, 0.80, 0.90), (0.05, 0.80, 0.95), (0.10, 0.80, 0.80)):
    p, r = precision(b, s, sp)
    print(f"base {b:.0%} sens {s:.0%} spec {sp:.0%} -> precision {p:5.1%}, {r:.2f} false alarms per true call")
# base 5% sens 80% spec 80% -> precision 17.4%, 4.75 false alarms per true call
# base 5% sens 80% spec 90% -> precision 29.6%, 2.37 false alarms per true call
# base 5% sens 80% spec 95% -> precision 45.7%, 1.19 false alarms per true call
# base 10% sens 80% spec 80% -> precision 30.8%, 2.25 false alarms per true call
for N in (1, 3, 5):                              # the horizon multiplies the base rate
    bN = 1 - (1 - 0.05) ** N
    print(f"{N}-year window: base {bN:.1%}, precision at 80/80 = {precision(bN,.8,.8)[0]:.1%}")
lev = lambda L, x: min(L * x, 1.0)               # equity loss; wipeout at a fall of 1/L
rec = lambda d: d / (1 - d)                      # the gain needed to get back
print(f"{lev(4,.15):.0%} {rec(lev(4,.15)):.0%} {1/4:.0%} | {lev(10,.10):.0%} {1/10:.0%}")  # 60% 150% 25% | 100% 10%
print(f"{46 + 0.5*20:.0f} {0.20*(56+54+51+48+46):.1f}")                                     # 56  51.0

Verify. Prove every piece against the worked figures above. Worked Example 1: the new-orders diffusion 46 + 0.5 × 20 = 56, the headline 0.20 × 255 = 51.0, and the new-orders-minus-inventories spread +10. Worked Example 3: Unit A's debt service 90 + 200 = ₹290 crore against ₹800 crore of cash flow, coverage 2.76×; Unit B's interest 0.06 × 150 = $9m against $10m of net operating income, coverage 1.11×. Worked Example 4: −8 × (−0.25%) = +2.00%. Worked Example 5: book ₹17,00,000, equity ₹2,00,000, equity loss 60% on a 15% market fall, wipeout threshold 1/4 = 25%, recovery on equity 3,00,000 / 2,00,000 = 150%, and the book's own required rise 20/17 − 1 = 17.6%. Worked Example 6: earnings yield 1/150 = 0.667%, re-rating multiple 150/25 = 6.0×, and N = ln 6 / ln(1+g) giving 12.82, 8.03 and 5.97 years at 15%, 25% and 35% (ln 6 = 1.7918), with the drawdown rule 0.80/0.20 = 400% and 0.90/0.10 = 900%. All reproduced here. Then compute what the module leaves implicit but which sharpens Worked Example 3 considerably: Unit B's refinancing rate at which interest coverage falls to exactly 1.0 is 10 / 150 = 6.67%, so a speculative unit paying 6% is 67 basis points from being unable to cover interest at all, and that threshold, not the label, is the analysis.

Quantitative lens

The frameworks above are qualitative by design and that is right, but one claim inside them is quantitative and is made constantly, here and everywhere else: that some indicator "has predicted every recession" or "flags crises in advance." That claim has an exact arithmetic, and running it changes what you are allowed to say. Everything below is computed from stated assumptions and the module's own worked figures; the sensitivity and specificity values are illustrative inputs, chosen to be generous to the signal, not measurements of any published indicator.

Take the most flattering assumptions you can defend and the answer is still uncomfortable. Suppose an early-warning indicator catches 80% of crises (sensitivity) and cries wolf only 20% of the time in calm years (specificity 80%). Suppose a systemic banking crisis strikes a given country in a given year with probability 5%, roughly once in twenty years, which is the right order of magnitude for a developed economy. Then when the signal fires, the probability that a crisis actually follows is 17.4%, and for every true call it produces 4.75 false alarms. Tighten the false-alarm rate to 10% and precision rises to 29.6% with 2.37 false alarms each; tighten it to 5% (a signal that is quiet in nineteen calm years out of twenty) and precision reaches 45.7%, still under a coin flip. This is not a criticism of any particular indicator. It is what happens to any classifier applied to a rare event, and it is why the module's §4.10 warning against timing is a mathematical statement and not a temperamental one.

Which means the lever that matters is not the one anyone talks about. Push sensitivity from 80% all the way to a perfect 100% while holding specificity at 90%, and precision moves only from 29.6% to 34.5%: catching every single crisis buys under five points. Push specificity from 80% to 95% at a fixed 80% sensitivity and precision goes from 17.4% to 45.7%, more than doubling. The asymmetry is structural: with a 5% base rate there are nineteen calm years generating false positives for every crisis year generating a true one, so the false-positive rate is multiplied by nineteen and the true-positive rate by one. An indicator is improved by being quiet, not by being alert, and the entire folk-praise for an indicator that "has never missed one" is praise for the dimension that barely matters.

Turn the question around and the bar becomes concrete. What false-alarm rate would a signal need to be right merely half the time it fires? At a 5% base rate and 80% sensitivity: below 4.21%, meaning specificity above 95.79%. At a 4% base rate: below 3.33%. Those are demanding numbers, and they are the honest reason that credit-gap and debt-service-ratio indicators are published as gauges to be weighed rather than as triggers to be acted on, which is exactly how the BIS presents them.

And here is the move to watch for in any study you read. Evaluate the same signal over a multi-year window instead of a single year and its apparent precision transforms, because the base rate does. An annual crisis probability of 5% becomes 14.3% over three years and 22.6% over five. Run the same unchanged 80/80 signal against those and precision goes from 17.4% at one year to 40.0% at three and 53.9% at five. Nothing about the indicator changed. So when a paper reports that a gauge "signals crises within three years," it is reporting a number that is roughly two and a half times the one-year figure, and comparing it to a differently-windowed result is meaningless. Always ask what window the precision was computed over; it is usually the largest single driver of the headline.

The other quantitative claim above, that leverage sets the size of the fall you cannot survive, is exactly true and worth stating as a general rule. Equity loss is leverage × market fall, wipeout arrives at a fall of 1/leverage, and the gain needed to recover a drawdown d is d/(1−d). So a 4× book loses 60% on a 15% market fall and is wiped out at 25%; a 1929-style 10× book is wiped out at 10%, which is an ordinary correction. And the recovery function is convex in a way that compounds the danger: the marginal cost of one more point of drawdown is 1/(1−d)², so at a 50% drawdown each further point costs 4× what the first point cost, at 80% it costs 25×, and at 90% it costs 100×. That is why the argument for reducing risk into euphoria does not depend on calling the top, which is §4.10's point exactly. The asymmetry does the work: you give up some upside in exchange for not being in the region where the recovery arithmetic becomes unrecoverable.

Honest limits, and the first one governs the rest. The sensitivity and specificity numbers above are assumptions, not measurements. I have not run any indicator against any crisis chronology in this addendum, and I am not reporting anyone's published performance; the values were chosen to be generous, and the point is that even generous values produce a low precision, which is a structural result about rare events rather than a finding about credit gaps. If you want the real numbers, they are computable: the BIS publishes the credit-to-GDP gap and debt-service-ratio series free and quarterly, Laeven–Valencia publishes the dated crisis list free, and the join is an afternoon's work. Do it, and quote your own numbers with your own window stated, because that is the only version of this claim worth making. Second, the base rate itself is not a constant: it differs by country, by decade, and by how a "crisis" is defined, and the chronologies disagree at the margins, so a precision figure inherits every one of those choices. Third, treating the years as independent draws is wrong, because crises cluster, the signal is serially correlated, and a proper evaluation needs country-clustered standard errors and out-of-sample testing, all of which will make the intervals wider than the point estimates suggest. Fourth, none of this arithmetic touches the deeper problem §4.10 names, which is that even a well-calibrated signal tells you nothing about when: precision is about whether, and the whole warning here is that whether and when are different questions. And finally, nothing above is a statement about the present. Where we are in the cycle is not knowable from this page; it requires today's panel, fetched from the series named above, with the release dates attached.

Do it in code: hold the panel as a table with the transformation fixed in advance; score any proposed signal on all four numbers (base rate, sensitivity, specificity, window) before you use it, and lead with the false-alarm count rather than the hit rate; report precision with the window stated in the same sentence; and compute the wipeout threshold and recovery requirement of any levered position before you take it rather than after. See G1/QM1.01, G2/DA2.01 and G2/QD1.02.

Where this goes next: galaxy cross-links

• G1 · QM1.01 Probability, Distributions & Estimation: conditional probability and base rates done properly, which is the machinery behind every number in the lens above and the single most transferable idea in this module.
• G1 · QM1.02 Hypothesis Testing & Regression: evaluating a signal against an outcome without fooling yourself: in-sample versus out-of-sample, clustered errors, and why a track record assembled after the fact is not a test.
• G2 · DA2.01 Supervised Machine Learning: precision, recall, ROC and the class-imbalance problem in their native setting; an early-warning indicator is a rare-event classifier, and every pathology of one applies to it.
• G2 · QD1.02 Signal & Factor Construction: turning an indicator into a usable signal: fixing the transformation in advance, avoiding lookahead in a series that gets revised, and the difference between a signal and a story about one.
• G3 · CN1.02 Hypothesis-Driven Problem Solving: stating what evidence would change your regime read before you go looking, which is the antidote to a template that always fits.

Teach it back & journal

Feynman prompt. Write a one-page explainer for a smart 15-year-old titled "Why does the economy keep booming and busting — and why can't anyone see the crash coming?" You must convey, without jargon: that the economy moves in waves driven mostly by credit (borrowing lets people spend more than they earn, until they can't); that the danger builds precisely when things feel safest (because calm makes everyone braver and more borrowed, which is Minsky's idea in plain words); that every bubble follows the same rough story (a real new thing → everyone piles in with borrowed money → prices go crazy and people buy just because prices are rising → someone gets scared → everyone runs for the door at once); and, the honest part, that even people who see the bubble usually can't time when it pops, so the smart move is not to predict the day but to carry less risk when things look euphoric and more when everyone is terrified. If your page needs the words "diffusion index," "Ponzi finance," or "growth/inflation regime," you don't own it yet. Say it in kitchen-table words (borrowing, confidence, the herd, "don't bet the house on knowing the day").

Journal prompt. Three entries. (1) Do the mini-project's first step in miniature: pull today's US 10y–3m yield-curve spread, the latest ISM and India PMI, and one credit-spread reading (FRED and the PMI releases), write the date beside each, and in three lines state (i) where you think we are in the cycle, (ii) which growth/inflation quadrant, and (iii) what you are explicitly not claiming. This is the first entry of a habit you'll keep for years, and the "as of" you'll re-grade against later. (2) Take one company from your studied list and write two sentences: which point in the business cycle is its demand most sensitive to, and which of the seven case-study patterns (leverage/margin, un-anchored inflation, balance-sheet recession, narrative-over-valuation, securitized credit chain, exogenous shock, currency/maturity mismatch) would be most dangerous to it, and why? (3) Reflect honestly: before this module, did you believe you could (or should try to) time the market's tops and bottoms? Write down, in your own words, the difference between recognizing regime and excess and calling the turn, and one concrete way you will act on the first without pretending to the second. This is the temperament the module most wanted to install, and writing the distinction is how it sticks.

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