α
AlphaSteve
← Brain index

Drawdown Protocol — portfolio-level kill criteria

Per-thesis kill criteria police individual positions. They are silent when ten positions move against the portfolio simultaneously because the regime turned, the value factor de-rated, or several theses were correlated through a risk dimension that wasn't surfaced at the time of writing. This file is the kill-criteria layer above the position level — a written protocol that triggers automatically on NAV drawdown depth and on time-in-drawdown duration, before emotional discipline becomes the last line of defense.

The protocol is advisory, not enforced. At each band, the daily portfolio check raises the flag and proposes the recommended action with a one-line rationale. The user decides whether to execute, override, or modify. Writing the bands down in advance is the discipline; following them is the operator's call when the moment arrives. Without the written record, the moment becomes ad-hoc rationalization at exactly the point where rationalization is hardest to resist.

What gets measured

Two parallel signals, each tracked daily in Performance:

  1. Absolute drawdown — peak-to-trough on portfolio NAV, calculated from the close. The peak is the highest closing NAV recorded since inception or since the last full recovery to a new peak (whichever is later).
  2. Relative drawdown vs. benchmark — rolling 6-month return of the portfolio minus rolling 6-month return of SPY, and minus rolling 6-month return of RPV. A portfolio that is −10% in a market that is +5% is in a worse epistemic state than one that is −15% in a market that is −20%. The relative track captures this.

The protocol band that fires is whichever is worse at any given check. A portfolio at −7% absolute but −12% relative-to-SPY is in the −10% band, not the −5% band.

The bands

Each band inherits the actions of the bands above it. Documentation is appended to the day's portfolio note in 12-Portfolio/Daily-Notes/ and cross-linked from the per-thesis calibration trackers of any positions that contributed materially.

Band 1 — −5% Acknowledge

The day's portfolio note carries a "drawdown active" header. No procedural change otherwise. The point is to make the kit see the state explicitly so it does not normalize to it. New positions allowed at the normal cadence; existing positions managed against their existing kill criteria.

Band 2 — −10% Re-read

Portfolio-wide forced re-read within the next 5 trading days. Each active position gets a one-paragraph status update in its calibration tracker: kill-criteria proximity, central value vs. current price, anything new since the verdict date. No new positions added without an explicit written override of the form "this opportunity is materially more attractive than the highest-conviction position currently held." Override reasoning logged in Transactions.

The point is to force opportunity-cost arithmetic before adding to a drawdown. Most additions in a drawdown are emotionally motivated; the override requirement makes the emotion visible.

Band 3 — −15% De-risk

Trim Core 1 sizing back to Core 2 sizing across the book, unless the thesis refresh from the −10% re-read explicitly re-confirmed Core 1 conviction with a written re-confirmation paragraph dated within the prior 10 trading days. Cash floor raised to 30% of NAV.

The point is to acknowledge that the kit's confidence calibration may be off when many positions move against it simultaneously. De-risking is not a verdict that the theses are wrong; it is a verdict that confidence should be smaller until evidence resolves.

Band 4 — −20% Forced de-grossing

Cash floor raised to 50% of NAV. Every active position gets a full thesis rebook within 10 trading days — not a refresh, a re-read against current price and current information as if the name were a fresh initiation. A kit-level debrief is filed in 10-Calibration/ analyzing whether the drawdown is:

  • Regime-attributable — broad market down, value factor down, sector down, with our positions tracking peers proportionately. The discipline worked; the regime is hostile. Response: patience and continued sizing discipline.
  • Methodology-attributable — positions falling on stock-specific news the thesis didn't anticipate, kill criteria firing more frequently than base rate would predict, central values systematically too high. The discipline isn't working as designed. Response: methodology calibration review.
  • Mixed — both. Response: separate the components and act on each.

The distinction matters for what comes next. A regime drawdown calls for waiting; a methodology drawdown calls for revision.

Band 5 — −25% Freeze and review

No new buys. Only sells allowed (sell triggers, kill criteria, opportunity-cost exits). The user is briefed via the standard channel with the kit-level debrief from Band 4 attached. The 06-falsification protocol is consulted — does the drawdown intersect with any of the conditions that would call the broader lens into question?

This is the band where the framework-level question becomes live. Most drawdowns stop here or unwind without escalating; that's expected. The point is that the question is on the table, in writing, when it should be.

Band 6 — −30% Operational pause

The kit stops opening new theses and stops producing buy verdicts pending a strategic decision. Existing positions continue to be monitored against kill criteria. The optimization run pauses Tier 2 changes (Tier 1 still runs, since fixing typos is uncontroversial regardless of NAV state). The user makes the call: continue with the strategy unchanged, modify the strategy via the falsification response menu, or pause operations entirely.

A −30% drawdown is the band at which every value PM in history has either been validated by subsequent recovery or has been the proximate failure point of the fund. The pause acknowledges the weight of the moment.

Time-in-drawdown — the parallel signal

Drawdown depth is one signal; duration is another. They carry different information.

Any drawdown lasting more than 180 trading days triggers a "regime persistence" entry in 10-Calibration/, independent of the depth band. Long drawdowns suggest methodology or regime mis-fit rather than acute event risk. The calibration entry asks: are we drawing down because we're early, because we're wrong, or because the regime is hostile to our lens? Each answer implies a different response, and the response menu is the same as the Band 4 debrief (regime / methodology / mixed).

A short, deep drawdown is an event. A long, shallow drawdown is a verdict. The protocol treats them differently.

What is not a drawdown trigger

To prevent false alarms:

  • Intraday or single-day moves below a band threshold that recover by close — the protocol fires on the close, not the tick
  • Mark-to-market movement on a position whose underlying business is unchanged and whose thesis kill criteria have not fired (this is volatility, not risk realization — see permanent-capital-loss)
  • Drawdown that coincides with a broad-market decline of similar magnitude where every value-shop comparable is in the same drawdown (this is regime, not methodology, and is what cash buffers are for)

The bands measure the portfolio's state. They do not interpret the state. The interpretation lives in the Band 4 debrief, which the user produces with the kit's help.

How the bands integrate with existing discipline

The drawdown protocol does not replace per-thesis kill criteria or per-position sell triggers. It runs in addition to them. Per-thesis criteria continue to fire on stock-specific events; the drawdown protocol fires on portfolio-level state. Both can fire in the same week, and they answer different questions.

The protocol also does not replace the 06-falsification framework. Falsification asks: is the lens itself wrong? Drawdown asks: is the portfolio in a state that warrants caution? A drawdown can fire without falsification firing (regime, not lens). Falsification can fire without a drawdown (a multi-year process-correct alpha gap with no acute decline). They are independent.

Operating mechanics

The daily portfolio check (9:00 AM, weekdays) reads this file as part of its existing workflow. The check:

  1. Computes current absolute drawdown and current relative drawdown (vs. SPY and vs. RPV)
  2. Identifies the worst band that applies
  3. If the band has crossed since the prior check, writes the band-crossing event to the day's portfolio note with the recommended actions
  4. If a band's mandated action has not been completed within its required window (e.g., a Band 2 re-read after 5 trading days), surfaces the pending action in the day's note as outstanding

The protocol's compliance is reviewed at each per-thesis calibration checkpoint and at the six-month-test evaluation. A protocol the operator ignored is worse than no protocol, because it produces a false sense of discipline; the calibration record exposes whether the bands actually shaped behavior.

Linked

  • Portfolio — current positions and the drawdown-state header that this protocol updates
  • Performance — daily ledger including the absolute and relative drawdown columns
  • permanent-capital-loss — the deeper risk frame this protocol operationalizes
  • position-sizing-kelly — the sizing framework whose limits the protocol enforces under stress
  • margin-of-safety-pricing — why cash buffers exist; the protocol's cash floors are this discipline made explicit
  • 06-falsification — the framework-level question the protocol routes toward at Band 5
  • six-month-test — the calibration checkpoint that scores protocol compliance
  • Transactions — where overrides and de-grossing actions are logged