LessonPost-mortemRiskCrises and market historyInvesting
Beware of geeks bearing formulas: a model built on calm history cannot see the storm
Warren Buffett · Berkshire Hathaway's 2008 letter to shareholders · 2009
The idea
In the 2008 letter Buffett blames much of the mortgage losses on models built from years of moderate house price rises. The models were not wrong about their data; they were wrong about whether the data covered the world. A history of modest moves produces a distribution with no room for a large one, and a product priced off that distribution looks safe until the move arrives. The same letter records the five years and more than four hundred million dollars it took to unwind a derivatives book in benign markets, and doubts that transparency alone can fix instruments that complex. The warning generalises to any model whose inputs are the recent past.
Ask what period a model was fitted to, and whether that period contained the event you are worried about.
When it works
Any risk model, stress test or structured product that reports a probability of loss.
When it may fail
No model can include an event that has not happened, and Buffett's own businesses use models. The remedy is a margin for the unseen, not the abandonment of the arithmetic.
How to use it
For a model's key distribution, find the start date of its history and list the stress events after that date. If the list is empty, the model has never seen a stress, and its tail is an assumption.
Where it came from
Letter Berkshire Hathaway's 2008 letter to shareholders · Warren Buffett
What the source says. Buffett blames much of the mortgage-security losses on history-based models built on years of moderate house price rises, and recounts that unwinding General Re's 23,218 derivatives contracts took five years and over $400 million in benign markets.
L3VLUP’s reading. The letter is as much about the limits of transparency as about models: Buffett doubts disclosure can cure a book that nobody can value. That is an argument against complexity itself, not only against bad inputs.
Connected ideas
Supports
A rating is a model output, and the model rests on a correlation assumptionAsk what a rating assumes about defaults happening together before trusting what it says about the pool.
Says something similar
Every bubble runs displacement, boom, euphoria, distress, revulsionAsk where a market is in the sequence rather than whether it is a bubble; the stages are more recognisable than the verdict.
Takes it further
When investing, pessimism is your friend and euphoria the enemyThe price of risk tracks the mood, so the useful time to buy is when the mood is worst, and the dangerous time is when it is best.
Says something similar
Responsibility for a company's actions sits with its management, not its regulatorsA firm that did what the rules allowed and lost has not been excused by the rules; management chose the position.
Concepts
Put it to work on L3VLUP
More like this
- TechniqueCreditFilm
A capital stack is a queue for losses, and the order is the whole product
Seniority is a queue for losses; the thickness of the slices beneath you is your protection.
From The Big Short · 2015
- FrameworkBankingFilm
Illiquid and insolvent are different problems with different rescues
Lending cures illiquidity; only capital cures insolvency. Decide which problem you have before choosing the remedy.
From Too Big to Fail · 2011
- Case studyCrises and market historyLetter
In a crash, long-duration cash flows move far less than the price, if you stay solvent
State the liquidity first, because the argument that prices overshoot value is only available to those who can wait.
From Bruce Flatt's March 2020 update to Brookfield shareholders · Bruce Flatt
- FrameworkInvestingBook
Low risk and high uncertainty are not the same thing
Look for situations where the range of outcomes is wide but the chance of permanent loss is small, because the market prices both as risk.
From The Dhandho Investor · Mohnish Pabrai