Deciding when to stop, before starting
The thresholds that govern a bad year were fixed before the first day was valued, and they cannot be moved while the account is falling.
Every systematic program has rules for entering and leaving a position. Fewer have a rule for leaving the program itself. That decision is usually made during the worst month, by the person with the most to lose from it.
We set ours in advance, published it as the Risk Limits, and timestamped it on Bitcoin before the first day of the record was valued. This note sets out what it contains and what it prohibits.
Three levels, none of them a stop
The Risk Limits set thresholds on four measures: how far the account is below its high, its return over the last 30 days, its return over the last 65 days, and how long it has been under water. Each has three levels. The first calls for a review, promptly and in any case within two business days. The second allows new positions to be opened at half size while the drawdown lasts. The third is a full re-validation of the program, with three possible outcomes, announced in the Monthly Report: continuation unchanged, continuation at reduced size, or retirement, which closes the record.
None of the levels stops trading on its own. A stop set on profit and loss alone cannot tell an ordinary losing period from a strategy that has stopped working, and once triggered it forgoes any recovery that follows. New entries stop automatically only on evidence that the system is not operating as designed: a stale data feed, a disagreement between its record of positions and the exchange's, or a component that fails to load. Each of these signals is specific and has a defined remedy.
The levels were derived from a backtest of the program before the record began. They describe what it is expected to withstand; they are not a forecast.
The rule that matters most
The thresholds cannot be amended while the account is in drawdown, and the current episode cannot be declared an exception.
Without that rule a published threshold would mean little. At a drawdown of 30% there is always a case for moving it: the conditions are unusual, the strategy was not built for them, the line was conservative to begin with. The rule settles that question in advance, when nothing is at stake.
The Risk Limits also prohibit the steps often taken instead of stopping: raising maximum exposure while the account is falling, closing positions at discretion, changing the deployed version in response to a loss, and resetting a drawdown or a day count when the version changes. Each would make a record look better without making the program better, and a verifier can check that none has been taken.
What we commit to
We do not claim the thresholds are right. They are a judgment made in advance with incomplete information, and they may prove too tight or too loose.
The commitment is narrower. The levels were set before the first day and are the same today. If the account crosses one, the record page shows it against the same scale, the review takes place within two business days, and its findings appear in the Monthly Report. If the program is retired, its record is closed and left in place.
A reader can hold us to this from the first month, well before the returns say much.