2018issue C0726-31
Worst-day caps for clustered system books
Risk is set first on a multi-system book. Each live system is scaled so a repeat of its worst-day-reference stays inside a 1.25% daily-loss-cap, then contract size is adjusted when same-day win-and-loss co-movement would stack those days.
- Loss containment is decided before attention shifts to profits, and a monthly ranking assigns contract counts before the next live window.
- Each live system is scaled so a repeat of its worst-day-reference, taken from historical daily backtest results, stays inside a 1.25% daily-loss-cap of account balance.
- About 50 systems can be live at once, so joint-loss-clustering is read from daily strategy profit and loss and used to cut size beyond a simple count of markets or names.
- Contract-feasibility keeps leftover equity for the next session, while scenario-risk and dirty-volatility give reasons to stand aside when losses can arrive larger than expected.
Set risk before the live window
Risk is set as the first constraint on the book. Loss containment is decided before attention shifts to profits.
A monthly ranking also assigns the contract count for each system selected to run in the coming month, so size is decided before the next live window.
Cap each system with a worst-day-reference
Each live system's contract size is scaled so a repeat of its worst historical day would be limited to 1.25% of account balance. That figure is the daily-loss-cap: a pre-set fraction of account equity that a single system's worst historical day is not allowed to exceed if it recurs.
The worst-day-reference is the largest one-day historical result of a system. It is taken from historical daily results produced by backtests of each system and used as the unit for scaling contracts against the chosen equity cap.
Cut size again when losses cluster
About 50 systems can be live at once. If every system printed its 1.25% worst day together, the book would lose more than half of equity, so same-day win-and-loss co-movement is measured and contract size is adjusted.
Joint-loss-clustering is the tendency of separate systems to post losses on the same calendar days. It is used to cut size beyond a simple count of markets or names.
For short-horizon books, clustering is read from daily strategy profit and loss rather than from market-to-market correlation, because a book can be long one system and short another on the same underlying.
Leave capital for the next session
Contract counts are kept feasible enough that capital remains after things go wrong, because the next session cannot be traded without leftover equity. Contract-feasibility is the practice of keeping live size small enough that a bad cluster of days still leaves capital for the next session.
Shared exposure and listed adverse cases
Position-concentration is how much of the account is exposed when many names or systems respond to the same shock. A stock-only book can stay concentrated even across many names because those names often rise and fall together. Mixing futures sectors is used to reduce that shared exposure.
Scenario-risk is an advance review of adverse what-if cases and how much account damage each case could inflict. Adverse cases are listed in advance and scored by how much they could damage the account, because losses can arrive larger than expected.
Dirty-volatility is price movement without orderly structure. Event windows that produce unstructured volatility can justify switching systems off, because models built on ordinary behavior may not bound risk then.
All readings on this track · 15 readings
- 1991Keep one risk ceiling after equity highs
- 1999Pre-trade size filters for expectancy and ruin
- 2000Budget the loss before entry with stops and position limits
- 2002Retest the entry before you exit
- 2002Option income after volatility and capital gates
- 2002Cap, scale, and delever when research buys late
- 2003Staying power starts as a planned-loser sizing rule
- 2006Leverage caps and concentration limits
- 2007A first-trade survival sequence for leveraged futures
- 2010Bound trend risk before the trade is placed
- 2011Event shocks, size caps, and second-order reweights
- 2014A copied scale-in ladder is not a personal risk plan
- 2018Worst-day caps for clustered system books
- 2019Pair-spread layer limits and event shocks
- 2020Pretrade quantity filters that bound account risk