CHAPTER 6 · The Fat Men and Their Marvelous Money Machine
Slicing cash flows does not erase risk
Explain prepayment, tranching and the difference between rearranging risk and removing it.
Michael Lewis, Liar’s Poker · Chapter 6. Original paraphrase and analysis; not the book text.
The idea beneath the story
The chapter follows the expansion of mortgage trading during 1981–1986. Lewis describes distressed sellers, the growth of Salomon’s expertise and increasingly specialised ways to trade mortgage cash flows. A central transformation is that loans once associated with local lending become material for a much broader securities business. The economic achievement is real, but the phrase money machine should prompt a question: which risks and constraints make the machine profitable, and who eventually holds them?
Start with the borrower’s option. A fixed-rate mortgage may allow early repayment. If market rates fall sufficiently, a borrower may refinance and repay the old loan. That can deprive an investor of a desirable stream of higher interest payments. If rates rise, borrowers may keep their old low-rate loans for longer, leaving the investor with cash flows that last longer just when their market value is under pressure. Credit performance can remain sound while the timing changes against the investor.
A simple hypothetical makes the asymmetry concrete. Imagine £100,000 of interest-only mortgage principal paying 6% annually, with principal returned at the end of year five. If it remains outstanding, the investor receives £6,000 each year and £100,000 at maturity. If the borrower repays after year one, the investor receives only one year of interest and must reinvest the principal. If the new reinvestment rate is 3%, the original five-year interest stream is gone. This ignores amortisation, costs and taxes to isolate timing risk; it is not a typical mortgage schedule.
Securities can redistribute cash-flow timing among different classes. Some arrangements prioritise one class for principal, while others separate interest from principal. These structures can match different investor needs. They cannot make the same underlying pool produce unlimited cash. A claim that benefits from faster repayment may have an offsetting claim that suffers. A full analysis asks what happens across the whole structure, including servicing costs and contractual rules, rather than admiring one attractive slice.
Do not confuse every mortgage-backed security with a synthetic CDO. A cash mortgage security is backed by cash flows from loans or other securities; a synthetic structure uses derivatives to reference credit outcomes. The ABACUS case is a later comparison about disclosure and selection incentives in a synthetic product. It is not the same instrument as the early mortgage pools in Lewis’s account, and it is not evidence that mortgage finance was destined to end in that particular controversy.
For a stock investor, the connection is to financial firms’ reported growth. Revenue from arranging and distributing securities may grow before the ultimate performance of those securities is known. A company may also retain inventory, guarantees or other exposures. The question is not merely how much was sold. Ask what remained on the balance sheet, how it was valued, what risks were transferred and what contingent obligations could return when conditions deteriorated.
Hypothetical risk allocation: a pool has £100 of principal, with a £10 junior layer absorbing initial credit losses and a £90 senior layer. With £8 of pool losses, the junior layer loses £8 and the senior layer none under this simplified waterfall. With £15 of losses, the junior layer is exhausted and the senior loses £5. The senior layer is protected by a buffer, not by a disappearance of loss. Correlated borrower defaults can consume a buffer faster than a model based on independent defaults would suggest.
The key reading discipline is to keep three questions separate: will borrowers pay, when will they pay, and who absorbs deviations from the expected path? Credit ratings, yields and reassuring product names do not answer all three. The learning lab lets you alter simplified assumptions and see the arithmetic, but it cannot price an actual mortgage security. Its purpose is to make the missing assumptions visible before the attractive return becomes the only number you notice.
ABACUS: selection is part of the information
ABACUS is used here to examine who selected exposures and what investors were told. It was a later synthetic structure, not a replay of the book’s early mortgage business. This case is shared with related chapters.
Documented sequence. ABACUS 2007-AC1 was a synthetic CDO linked to mortgage-security performance. In July 2010, the SEC announced a $550 million settlement with Goldman Sachs concerning misleading disclosure. Goldman acknowledged incomplete marketing information about Paulson & Co.’s role in selecting the reference portfolio and its adverse economic interest. It otherwise settled without admitting or denying the allegations. The announcement was subject to court approval. The case concerns this specific transaction and disclosure, not all mortgage securities.
SEC · Goldman Sachs ABACUS settlement announcement, July 2010 · Checked 10 October 2026
Original analysis: the sample was chosen. A buyer can understand the broad asset category yet lack important information about how the particular exposure was selected. Two portfolios with the same label can have different characteristics. If a participant helping choose the reference assets benefits when those assets perform poorly, the buyer has a reason to scrutinise selection. The issue is not that opposing views exist; opposing views are ordinary in markets. It is whether the process and relevant interests were represented accurately.
A synthetic structure also requires careful language. It references credit outcomes through derivatives rather than simply passing through payments from a collection of newly funded household loans. That distinction matters when discussing who is financing what and how exposures are created. The early mortgage-market development in Lewis’s book and this later structure belong to different periods and arrangements. Treating them as identical would hide the very contractual detail the lesson asks readers to examine.
Hypothetical worked example. Suppose a buyer is offered exposure to twenty companies described as a broad sample of an industry. A participant who profits from defaults helped choose all twenty, focusing on firms with near-term refinancing needs. The buyer might still choose to invest at an appropriate price, but the selection rule is relevant evidence. An industry-wide average default rate would not necessarily describe this selected group. This example is invented and does not assert the actual portfolio composition of ABACUS.
The stock-market connection is curated evidence. A fund presentation may select winning trades; a management deck may select favourable competitors; a research service may select companies likely to support its thesis. Selection need not be improper to affect interpretation. The reader should ask what the full population was, who selected the sample and whether the selection rule was known before the outcome. A correct calculation on a biased sample remains a biased basis for a broader claim.
A common misunderstanding is that a sophisticated investor should never need disclosure. Expertise helps interpret information; it does not reveal an undisclosed process automatically. Another is that a seller having inventory or an opposing view proves a violation. It does not. The specific representations, material information and legal context matter. Our educational comparison concerns the information problem and does not offer a legal conclusion about other transactions.
For a shareholder evaluating a financial firm, distribution revenue should therefore be considered alongside the quality of disclosure and client relationships. A profitable transaction may create later remediation costs or reputational damage if its presentation is defective. Conversely, clear disclosure can allow parties with different objectives to transact knowingly. The goal is not a market in which everyone agrees, but one in which disagreement is not disguised by an inaccurate account of the product.
Takeaway. Ask who built the basket before asking whether the basket’s headline return is attractive. Obtain the selection rule, the relevant economic interests and the limits of the comparison data. Then decide what remains uncertain. Complexity should lead to more precise questions, not a suspension of them.
Workshop: follow the cash, the option and the loss
Draw the cash-flow chain before comparing yields. A borrower pays a servicer; fees and expenses may be deducted; remaining payments pass through contractual rules to investors. Each arrow represents a process that can introduce timing, cost or operational risk. A security name compresses that chain into a label, but the label cannot replace the rules. Ask which cash flows the investor owns, which are conditional and which are senior to other claims. Two instruments referencing similar loans can create quite different experiences for their holders.
Pooling can reduce the importance of a single borrower, but only to the extent that the underlying risks are not perfectly shared. A thousand borrowers in one employment market may be exposed to the same local shock. Geographic spread can help while still leaving a common sensitivity to interest rates or house prices. Diversification is therefore a question about dependence, not simply the number of items. A pool with more loans can be easier to model than an individual loan in some respects, yet remain vulnerable to a factor that affects all of them together.
Distinguish scheduled amortisation from voluntary prepayment. Scheduled amortisation is the contractual reduction of principal over time. Prepayment changes that schedule because the borrower exercises a permitted option or otherwise pays earlier. For the investor, faster principal return can reduce credit exposure but also remove future interest and create reinvestment risk. Slower repayment can keep funds tied up. There is no universal statement that faster is always better or slower always worse; the purchase price, claim type and reinvestment opportunity determine the effect.
A hypothetical premium example helps. An investor pays £105 for a claim that returns £100 of principal plus interest. If the borrower repays almost immediately under the assumed contract, there may be little interest to offset the £5 premium. Another investor buys a principal-only claim at £70 and benefits from receiving its £100 sooner, all else equal. These are simplified examples that omit contract-specific compensation and costs. They show why two holders exposed to the same repayment event can prefer different outcomes. The word mortgage does not reveal which preference applies.
Now examine a loss waterfall. Suppose £100 of principal supports a £10 junior claim and a £90 senior claim. The junior absorbs the first £10 of credit losses. This arrangement changes the distribution of loss across investors, but the total remains the pool’s loss under the stated assumptions. The senior claim’s apparent stability depends on the buffer being sufficient. A model that underestimates common shocks can overstate protection, because correlated losses can exhaust the junior layer. The proper question is how the protection behaves across scenarios, not whether the label sounds conservative.
Do not confuse a credit-loss waterfall with a payment-timing waterfall. A structure may prioritise principal payments for one class without using exactly the same priority for credit losses. Other rules can redirect cash when tests are breached. The details matter. A diagram with three neat boxes may be an excellent introduction and a poor substitute for the governing document. When the terms are too complex to explain, record that limitation. Complexity can be managed by specialists, but it does not become harmless because a reader cannot readily understand it.
Pricing also depends on the market’s ability to absorb the claim. A model can estimate present value under assumptions while an executable quote reflects inventory, balance-sheet capacity and the number of willing buyers. A temporary gap between model value and sale price can matter greatly to a financed holder. The model is not necessarily wrong, and the quote is not necessarily irrational. They may answer different questions: expected cash-flow value for a patient owner versus immediate liquidation value under current constraints. An analysis should state which question it is answering.
For an equity investor studying an intermediary, map the revenue chain beside the risk chain. Origination, servicing, structuring, underwriting and trading generate different fees and obligations. A firm may sell an asset while retaining a guarantee, an exposure through another instrument or a reputational reason to provide support. Do not presume that every sale transfers every risk. Equally, do not presume that nothing was transferred. Inspect the disclosed retained interests and the circumstances in which cash could be demanded. The business model lies in the combination of revenue and residual exposure.
Competition changes the value of expertise. Early participants may earn substantial margins because few others can analyse or distribute a product. As techniques spread, the same service can remain useful while commanding a lower price. The analyst should compare profit with the capital and complexity required to earn it. A firm may respond by innovating, improving efficiency or taking more difficult risks. Those responses need separate evaluation. A more complex product is not automatically better or worse; ask what customer need it serves and which new assumptions it introduces.
End with a plain-language risk statement. Describe who pays, when payment can change, who takes the first loss and how the investor could exit. Add the assumptions that remain unverified. If you cannot write those sentences, an attractive yield is premature evidence. The workshop is not a pricing engine for mortgage securities or a forecast of a future crisis. It is a way to make financial engineering inspectable. Its stock-market value is the ability to ask better questions about financial firms whose earnings depend on moving, packaging and retaining these claims.
One final distinction is between a guarantee and the price of the guaranteed claim. A contractual protection may reduce a specified loss while leaving interest-rate sensitivity, timing uncertainty or sale-price risk. The guarantor’s capacity and the exact covered event also matter. Do not use the existence of protection as a reason to stop reading its scope. In a hypothetical contract that covers unpaid principal but not a premium paid in the market, repayment of all principal does not necessarily recover the investor’s purchase cost. The example illustrates a boundary, not the terms of a particular programme. State what is protected, against which event, by whom and when payment occurs. Those four questions turn a reassuring adjective into a description that can be compared with the holder’s actual exposure.
Decision exercise (hypothetical)
A £100 pool has a £10 first-loss layer. The pool loses £15. What happens in a simple two-layer structure?
- Only £10 can ever be lost
The buffer allocates losses; it does not cap the pool’s losses.
- The senior layer loses £15
The junior layer absorbs the first £10 in this stated structure.
- The junior layer loses £10 and the senior £5
Yes. Losses are allocated, not erased.
Recall and self-check
Why can falling rates hurt a mortgage investor?
Refinancing can return principal early and remove higher-rate future interest.
Are credit risk and timing risk identical?
No. A borrower can repay in full yet change the timing in a way that hurts the holder.
Reflection
Which three questions would you ask before accepting a structured product’s headline yield?
Ask about payment, timing and loss allocation; then identify which assumptions you cannot verify.
Open interactive chapter & notes