Oneiros Vision — Implementation Sensitivity
Period: 2010-01-04 to 2026-09-30 (16.7 years) Signal: production MDI regime labels, identical in every row below Execution: signal at the close of T, traded at the close of T+1; 2 bps per-leg transaction cost Data: frozen month-end dataset, SHA-256 verified, cutoff 2026-09-30 Currency: US dollars; every portfolio holds US-listed ETFs Control: every implementation is scored against a static portfolio holding that implementation's own average weights, charged zero transaction cost
The whitepaper positions OV as a risk overlay rather than a portfolio: the MDI supplies the regime reading and the allocator selects the instruments. This report tests that positioning. If the reference ETF map is genuinely illustrative, the signal must add value across a wide family of implementations, and must survive being traded at the speed a real allocator can trade. The reference weight map is not modified.
Summary
- The overlay's contribution is mainly drawdown reduction. Against matched static portfolios the drawdown benefit is +7.29 to +12.37 pp in every ladder tested; the CAGR delta is +0.86 to +1.51 pp, a modest return contribution secondary to the drawdown effect.
- That benefit does not depend on the reference weights. Muting the overlay to RIA-scale tilts retains 102% of it.
- The signal has a shelf life of one to two weeks: a week of extra delay keeps 79% of the benefit, two weeks 21%, and a month-stale position is actively harmful (-52%).
- For an allocator who cannot trade daily, a partial-implementation glide rule kept the benefit where a calendar rebalance did not. Calendar rebalancing is a coin flip. Caps of 10.0 pp/day and faster keep at least 94% of the drawdown benefit in every ladder tested, and 10.0 pp/day itself at least 98%; at 7.5 pp/day the weakest ladder keeps 80%, and slower caps erode it quickly.
Why the Reference Chassis Overstates the Trading Burden
Sleeve travel between adjacent regimes, as one-way turnover on the production map:
| Transition | One-way trade | Occurrences |
|---|---|---|
| CALM (Risk-On) → WARNING (Hedged) | 100% | 1 |
| CHOP (Neutral) → WARNING (Hedged) | 95% | 27 |
| WARNING (Hedged) → CHOP (Neutral) | 95% | 28 |
| WARNING (Hedged) → CRITICAL (Bunker) | 65% | 14 |
| CRITICAL (Bunker) → WARNING (Hedged) | 65% | 14 |
| CALM (Risk-On) → CHOP (Neutral) | 40% | 43 |
The CHOP ↔ WARNING pair alone accounts for 55 near-total portfolio rotations, and WARNING → CRITICAL is a 65% one-way trade. At 13.5x annual turnover this is not executable at signal speed by a manager with IPS ranges, tax lots and client sign-off.
Note where that turnover comes from: 12.7x of the 13.5x is regime switching and only 0.7x is the volatility circuit breaker. The trading burden is the frequency of regime changes, not the circuit breaker, and muting the sleeve percentages barely reduces it (11.7x). Turnover has to be solved in the execution rule, which the next-to-last section does.
Sleeve Ladders: Does the Signal Survive Muting?
Each ladder sets risk-sleeve exposure for CALM / CHOP / WARNING / CRITICAL. The production ladder reproduces the shipped weight map exactly (asserted in code). Δ columns are versus that ladder's own matched static portfolio — the overlay's contribution, stripped of asset allocation.
| Ladder | CALM/CHOP/WARN/CRIT | CAGR | Vol | Max DD | Static CAGR | Static Max DD | Δ CAGR | Δ Max DD | Turnover |
|---|---|---|---|---|---|---|---|---|---|
| Production reference | 100% / 95% / 20% / 0% | 13.80% | 11.36% | -13.61% | 12.52% | -22.52% | +1.28 pp | +8.91 pp | 13.5x |
| Muted overlay | 85% / 75% / 40% / 20% | 12.20% | 9.64% | -11.25% | 11.10% | -20.33% | +1.10 pp | +9.08 pp | 11.7x |
| Gentle overlay | 80% / 70% / 50% / 35% | 11.92% | 9.41% | -11.82% | 10.82% | -20.31% | +1.10 pp | +8.48 pp | 11.6x |
| Minimal overlay | 75% / 70% / 60% / 50% | 12.08% | 9.62% | -13.65% | 10.87% | -20.94% | +1.21 pp | +7.29 pp | 11.3x |
| Defensive-tilted | 90% / 70% / 25% / 5% | 11.48% | 9.28% | -10.49% | 10.63% | -18.80% | +0.86 pp | +8.31 pp | 12.1x |
| Risk-on-flattened | 90% / 90% / 20% / 0% | 13.07% | 10.59% | -12.90% | 11.80% | -21.16% | +1.28 pp | +8.26 pp | 12.5x |
| Equity-heavy | 100% / 100% / 50% / 20% | 15.00% | 12.14% | -13.12% | 13.49% | -25.49% | +1.51 pp | +12.37 pp | 13.6x |
Every ladder improves drawdown against its own static control, and the return delta runs +0.86 to +1.51 pp. The drawdown benefit is the larger and more consistent effect: the MDI is first a risk-governance signal, and its value is robust to how the sleeves are defined.
Signal Shelf Life
Extra lag applied on top of the T+1 execution already assumed everywhere else, on the muted
ladder. Retention is the share of the daily-execution drawdown benefit that survives.
| Extra lag | Max DD | Δ Max DD vs static | Retention |
|---|---|---|---|
| None — daily (T+1) | -11.25% | +9.08 pp | 100% |
| +1 day | -12.11% | +8.22 pp | 91% |
| +1 week | -13.13% | +7.21 pp | 79% |
| +2 weeks | -18.42% | +1.91 pp | 21% |
| +1 month | -25.08% | -4.73 pp | -52% |
| +2 months | -25.11% | -4.75 pp | -52% |
| +1 quarter | -23.84% | -3.48 pp | -38% |
The signal decays fast and then inverts: acting on a month-old MDI reading is worse than not acting at all, because the position is reliably wrong at turning points. In this sample the benefit survived only when positions changed within roughly ten sessions of a regime change.
None of the tested delays improves the result: one extra day keeps 91% of the benefit (+8.22 pp against +9.08 pp at zero). The baseline already trades at the close after the signal, so any further delay only makes the position staler.
Calendar Rebalancing Is a Coin Flip
A single rebalance calendar confounds the policy with the luck of which days it lands on. Every possible starting offset is tested below, on the muted ladder.
| Policy | Worst offset | Median offset | Best offset | Worst Max DD | Offsets worse than static |
|---|---|---|---|---|---|
| Monthly (21 offsets) | -4.32 pp | +0.80 pp | +7.49 pp | -24.75% | 9/21 |
| Quarterly (63 offsets) | -6.32 pp | -3.43 pp | +6.38 pp | -26.14% | 51/63 |
Monthly rebalancing has a median benefit of +0.80 pp against +9.08 pp for daily execution, and 9 of 21 start dates are net harmful: it is not a dependable implementation. Quarterly is worse: 51 of 63 offsets destroy value, with the worst calendar reaching -26.14% maximum drawdown.
Glide Rules: Trading More Slowly Than Daily
A glide rule caps the largest single-position change per session and scales every other position by the same factor, so the book stays fully invested and always travels toward the current regime's weights in the reference model. It carries no calendar dependence by construction.
On the muted ladder, the same implementation as the shelf-life and calendar tests above:
| Max move per day | CAGR | Max DD | Δ Max DD vs static | Turnover |
|---|---|---|---|---|
| 0.5 pp | 10.86% | -21.60% | -0.45 pp | 3.6x |
| 1.0 pp | 10.64% | -21.62% | -1.02 pp | 5.8x |
| 2.0 pp | 10.83% | -19.37% | +1.11 pp | 8.0x |
| 3.0 pp | 10.95% | -17.25% | +3.20 pp | 9.3x |
| 5.0 pp | 11.15% | -13.61% | +6.78 pp | 10.4x |
| 7.5 pp | 11.47% | -11.50% | +8.85 pp | 11.0x |
| 10.0 pp | 11.75% | -11.35% | +8.99 pp | 11.3x |
| 15.0 pp | 11.78% | -11.53% | +8.81 pp | 11.4x |
| 25.0 pp | 11.93% | -11.42% | +8.91 pp | 11.6x |
On this ladder no glide cap beats execution at the next close. The best tested setting, 10.0 pp/day, reaches -11.35% maximum drawdown against -11.25% for full next-close execution and keeps 99% of the signal's drawdown contribution.
The Cap That Holds in Every Ladder
Share of each ladder's own next-close drawdown benefit that each cap keeps:
| Max move per day | Production | Muted | Gentle | Minimal | Defensive | Risk-on flat | Equity-heavy | Weakest |
|---|---|---|---|---|---|---|---|---|
| 0.5 pp | 0% | -5% | 1% | -1% | -5% | -6% | -0% | -6% |
| 1.0 pp | -39% | -11% | -2% | 0% | -6% | -43% | -23% | -43% |
| 2.0 pp | -8% | 12% | 18% | 20% | 23% | -9% | -7% | -9% |
| 3.0 pp | 26% | 35% | 38% | 40% | 51% | 25% | 12% | 12% |
| 5.0 pp | 79% | 75% | 69% | 70% | 95% | 83% | 48% | 48% |
| 7.5 pp | 111% | 98% | 92% | 90% | 99% | 111% | 80% | 80% |
| 10.0 pp | 117% | 99% | 98% | 98% | 100% | 118% | 101% | 98% |
| 15.0 pp | 115% | 97% | 98% | 111% | 100% | 112% | 96% | 96% |
| 25.0 pp | 104% | 98% | 99% | 107% | 100% | 105% | 94% | 94% |
One ladder can flatter a cap. At 7.5 pp/day the muted ladder keeps 98% of its benefit, but the equity-heavy ladder keeps 80%. 10.0 pp/day is the slowest cap that keeps at least 95% in every ladder (98% in the weakest), and no faster cap tested keeps less than 94%. It is the default speed on the terminal's EXECUTION screen, which applies it to the portfolio manager's own target.
Above 100%, the glide reduced that ladder's drawdown further than trading the whole rotation at the next close did.
What a Cap Asks of the Desk
Book traded per session, and sessions to complete, for the production map's most common regime changes:
| Max move per day | CHOP → WARNING (95% of the book) | WARNING → CRITICAL (65% of the book) | CALM → CHOP (40% of the book) | Trading days a year |
|---|---|---|---|---|
| 5.0 pp | 12% a session for 8 sessions | 6% a session for 11 sessions | 10% a session for 4 sessions | 65 |
| 7.5 pp | 18% a session for 6 sessions | 9% a session for 8 sessions | 15% a session for 3 sessions | 50 |
| 10.0 pp | 24% a session for 4 sessions | 12% a session for 6 sessions | 20% a session for 2 sessions | 37 |
| 15.0 pp | 36% a session for 3 sessions | 18% a session for 4 sessions | 30% a session for 2 sessions | 30 |
| Next close | 95% a session for 1 session | 65% a session for 1 session | 40% a session for 1 session | 12 |
The cap limits the largest single position, not the book: every other position moves by the same fraction of its own gap. On the production map a CHOP → WARNING change moves 95% of the book and its largest leg is 40 points, so at 10.0 pp/day the desk trades about 24% of the book a session for 4 sessions, and trades on about 37 days a year. Transitions are shown before the volatility circuit breaker; trading days are counted on the realised production path, as sessions that move more than 0.1% of the book.
There is a floor. On the muted ladder, at 5.0 pp/day the rule keeps 75% of the daily benefit, and at 0.5 pp/day the benefit falls to -0.45 pp — consistent with the shelf-life table above, where a position two weeks stale has lost most of its value. A glide keeps the benefit only while it is faster than the signal decays.
The mechanism is straightforward: partial implementation damps whipsaw. The CHOP ↔ WARNING pair flips 55 times, and a glide rule never completes most of those round trips.
How to Read the Tables
The reference ETF backtest answers "what would OV's own model portfolio have returned". That is not the question for a portfolio manager who keeps their own holdings. This report asks what following the MDI's regime added to a range of different portfolios, and the answer is a drawdown reduction that survived changing the sleeve percentages and realistic execution.
The Δ columns are the signal's contribution; the CAGR of any row is mostly a property of the instruments chosen. The glide tables show the cost of trading more slowly than daily: within a few percent of daily execution from 10 pp a day up, and quickly worse below it.
Limitations
Ladders share the production sleeve composition and vary only sleeve exposure; a client whose growth sleeve is concentrated, single-stock or private will behave differently. The matched-static control uses full-sample average weights, which is not information an allocator had in advance — it is a fair attribution control, not an investable alternative. The glide plateau is flat, but where it begins moves between ladders (7.5 pp/day keeps between 80% and 111% depending on the ladder), and the dependable cap is chosen on this sample as the slowest that clears the band in every ladder. Regime labels come from production signals generated once, so nothing here tests the robustness of the MDI itself. All results are modelled on historical data and are not a forecast.