The breaker that could never reset

A drawdown breaker that flattens the portfolio needs a rule for coming back. Ours waited for a recovery that a portfolio sitting in cash can never produce, and four years of testing never noticed.

The momentum book has two controls that act on the whole portfolio rather than on individual stocks. Both make the same decision, get out, for different reasons.

• The trend switch. Every day the benchmark index is compared with its 150-day average. Above it, the book is invested. Below it, the book sells everything and holds cash, immediately, without waiting for the next monthly rebalance.
• The drawdown breaker. If the book ever falls 20% from its highest value, it is flattened, whatever the trend switch says.

Acting on the whole book is deliberate. A momentum book exits stocks when they drop out of the ranking, not when they hit a price stop, so there is no per-stock stop to fall back on. One decision about the market, applied to everything at once, is simpler to reason about and simpler to test. Over twenty years of history, switching both controls off roughly doubled the worst drawdown while adding only a little return.

The rule that looked sensible

The breaker's original rule read naturally: pause at a 20% drawdown, resume at 10%. Trip when things get bad, come back once they are half as bad. It sat in the configuration, was documented and reviewed, and looked like a control.

It could not work. Tripping the breaker flattens the book. A book holding only cash does not rise or fall with the market, so its drawdown cannot shrink from 20% to 10%. The condition for coming back depended on a recovery that the breaker itself had made impossible.

Why nobody noticed

The main four-year test never reached the breaker. Its worst drawdown stayed well short of 20%, so the trip never happened and the broken release was never exercised. A control that never fires in testing looks exactly like a control that works.

It took a twenty-year test on real daily prices to find it:

• 22 Jan 2008The breaker tripped as the global financial crisis took hold, and the book went to cash.
• 2009 – 2011The market recovered strongly. The book stayed in cash throughout.
• 25 Oct 2011The breaker finally released, and not because of the market. A small notional interest rate on idle cash had slowly nudged the book's value back up to the threshold.

With that interest rate set to zero, as it should be for a delivery account, the breaker never released at all. The book would have sat in cash for the rest of the test.

The fix: come back on a market condition

Re-entry now depends on something a flattened book cannot freeze: the market itself. After a trip, the book waits at least 21 trading sessions and returns only once the benchmark is back above its trend average. Over the same twenty years the breaker now trips twice, in January 2008 and September 2018, and holds the book flat for 79 of 4,925 sessions, instead of three years.

Testing controls, not just strategies

The broader change was in how controls are tested. A safety mechanism has to be exercised on a window long and bad enough to make it fire, and then checked for what happens after it fires. A backtest that never trips the breaker says nothing about the breaker, however good the rest of the numbers look.