A commodity-trading legend
Dennis discusses learning, rules and the uncomfortable role of losing trades. The chapter’s broad lesson is to inspect a sequence of decisions rather than demand that each trade succeed. If a method relies on occasional large favourable moves, abandoning it after a short losing sequence can change its structure. But blindly continuing a broken method is a different mistake.
Our interpretation distinguishes an ordinary adverse sequence from evidence against the method. That distinction needs criteria written in advance: implementation errors, changes in market access, costs or assumptions may justify review. “Keep going because a winner is due” is not such a criterion. Neither is “one loss means the rules are wrong”.
Worked example
Nine fictional losses of 1 unit and one gain of 12 total +3 before costs. Ten costs of 0.4 change the total to −1. The same win rate can describe opposite net results.
Limits
Historical success and training anecdotes do not establish that any copied rule will work today. Avoid overfitting and include costs and unseen data when evaluating a model.
Case connection
A mechanical rule also depends on executable markets. Black Monday asks whether your process assumptions survive a stressed environment, not merely a losing streak.
Black Monday: when exits become crowded
A risk-control rule can look sensible in isolation and become dangerous when many portfolios act on it together.
The documented sequence
On 19 October 1987, the Dow Jones Industrial Average fell 22.6% in one session. Federal Reserve History describes international selling pressure, market-structure weaknesses and portfolio-insurance strategies that amplified selling as prices fell. The next day, the Federal Reserve affirmed its readiness to provide liquidity. The episode also encouraged the development of trading pauses. These are selected facts, not a claim that one mechanism explains the entire crash. [1]
Interpretation: protection is a process
A dynamic hedge changes exposure as market conditions change. Unlike a contractual payout from a solvent counterparty, it may depend on completing transactions. If a rule requires selling more after a decline, the protection depends on finding buyers at usable prices. At portfolio level, reducing exposure appears defensive. At system level, simultaneous defensive selling can consume the very liquidity that the defence assumes. That distinction is a mechanism to examine, not a reason to reject every hedge or every systematic strategy.
Why a stop is not a price guarantee
The lesson for the interviews is about the gap between intention and execution. An exit level expresses a decision. It does not create a buyer, remove a queue or guarantee the next available price. In a hypothetical market with a last trade at 100 and the next executable bid at 85, a sell instruction triggered below 95 cannot manufacture a fill at 95. A limit order changes the problem by constraining price, but may not execute. Neither order type abolishes the trade-off.
A decision before the event
Suppose a fictional learner has a model that behaves well on ordinary daily data. Before accepting its reassuring results, the learner asks whether several positions need the same exit, whether the model allows trading interruptions, and whether execution costs increase during stress. These questions do not require predicting a specific crash date. They test dependencies. A strategy that cannot tolerate the model being wrong about liquidity has a different risk profile from one with spare capacity and smaller exposure.
Connect the interviews without flattening them
Schwartz emphasises accepting a mistake instead of defending pride. Dennis highlights the danger of judging a process by a few trades. Jones discusses changing exposure and revising a view when expected action fails to appear. Seykota draws attention to the relationship between rules and the person following them. Black Monday lets us ask each a different question: can you act, will you follow the plan, when should the plan be reviewed, and what market conditions does the plan quietly require?
Uncertainty and the wrong takeaway
A common misunderstanding is that one famous crash proves all computerised trading is harmful, or that a discretionary trader would necessarily have escaped. The cited history identifies several contributors; it does not supply a controlled experiment that ranks every possible strategy. Equally, a crisis response by a central bank is not a commitment that a future position will be rescued. Learning from an event means extracting a testable vulnerability without turning the ending into a universal law.
The memorable lesson
Write an execution assumption beside every risk rule. “Exit at the threshold” should prompt “under what trading conditions, at what possible cost, and with what remaining exposure?” This does not eliminate uncertainty. It makes it visible early enough to shape the scale of a hypothetical experiment. The best review examines both the intended decision and the practical ability to carry it out.
Consider
What would remain exposed if your preferred exit could not execute immediately?
Analysis guide
List the position, shared liquidity dependencies and delay. Consider a smaller starting exposure or pre-funded buffer in a hypothetical plan. Do not assume a different order type guarantees both execution and price.
Reflection
What evidence would make you review a method without chasing recent outcomes?