Oneiros Vision — CRITICAL (Bunker) Allocation Sensitivity
Period: 2010-01-04 to 2026-09-30 (16.7 years) Signal: production MDI regime labels, identical across all scenarios 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 Bunker exposure: 4.3% of sessions (181 of 4,211) across 14 episodes (median 11 sessions)
Only the CRITICAL (Bunker) sleeve is varied. CALM, CHOP and WARNING targets stay at
production values, so every difference below is attributable to bunker composition alone.
The reference weight map is not modified: each scenario runs against a private copy of it.
Headline Finding
Re-weighting the bunker cannot materially change capital-preservation outcomes, because the bunker is not where the strategy's losses occur. Over the full sample the bunker was in force on 4.3% of sessions and accounted for 0.4% of all loss accumulation. Moving from 85/5/10 to 50/30/20 is worth +0.26 pp of CAGR and +0.67 pp of maximum drawdown — real, statistically positive, but second-order. The reference strategy keeps scenario A: its weights are frozen, and a difference this small is not a reason to change them.
Where the Losses Actually Come From
Loss share is each regime's contribution to the sum of all negative daily returns.
| Regime | Sessions | Time | Ann. Return | Ann. Vol | Share of all losses |
|---|---|---|---|---|---|
| CALM (Risk-On) | 1,049 | 24.9% | 14.14% | 13.90% | 34.4% |
| CHOP (Neutral) | 1,833 | 43.5% | 20.09% | 13.18% | 53.9% |
| WARNING (Hedged) | 1,146 | 27.2% | 5.99% | 4.37% | 11.3% |
| CRITICAL (Bunker) | 181 | 4.3% | 1.92% | 0.97% | 0.4% |
The full-period maximum drawdown (-13.61%) ran 2014-11-28 to 2015-08-25. Bunker exposure inside that window was 11.3%, and the bunker gained +0.35% while held. No bunker weighting could have prevented that drawdown.
Scenario Definitions
The volatility circuit breaker halves every non-BIL weight and absorbs the remainder into BIL whenever 21-day realized S&P volatility reaches 25%. It was active on 47.5% of bunker sessions, so the right-hand column is what is actually held for roughly half the time the bunker is in force — a 50/30/20 target is a 75/15/10 position whenever the cap binds.
| Scenario | Target sleeve | Effective sleeve when the vol cap binds |
|---|---|---|
| A. Production (frozen) | BIL 85% / GLD 5% / UUP 10% | BIL 92% / UUP 5% / GLD 2% |
| B. Cash 70 | BIL 70% / GLD 20% / UUP 10% | BIL 85% / GLD 10% / UUP 5% |
| C. Cash 60 | BIL 60% / GLD 25% / UUP 15% | BIL 80% / GLD 12% / UUP 8% |
| D. Cash 50 | BIL 50% / GLD 30% / UUP 20% | BIL 75% / GLD 15% / UUP 10% |
| E. Cash 40 | BIL 40% / GLD 35% / UUP 25% | BIL 70% / GLD 18% / UUP 12% |
| F. Cash 50, gold-max | BIL 50% / GLD 40% / UUP 10% | BIL 75% / GLD 20% / UUP 5% |
| G. Cash 50, dollar-tilted | BIL 50% / GLD 20% / UUP 30% | BIL 75% / UUP 15% / GLD 10% |
| H. Balanced-3 | BIL 65% / GLD 15% / UUP 10% / IEF 10% | BIL 82% / GLD 8% / UUP 5% / IEF 5% |
| I. Diversified hedge | BIL 50% / GLD 20% / UUP 15% / IEF 15% | BIL 75% / GLD 10% / UUP 8% / IEF 8% |
| J. Duration-first | BIL 60% / GLD 10% / UUP 10% / IEF 20% | BIL 80% / IEF 10% / GLD 5% / UUP 5% |
Full-Period Performance
| Scenario | Cumulative | CAGR | Volatility | Max DD | Sharpe | Sortino | Calmar |
|---|---|---|---|---|---|---|---|
| A. Production (frozen) | 767.40% | 13.80% | 11.36% | -13.61% | 1.06 | 1.50 | 1.01 |
| B. Cash 70 | 786.21% | 13.95% | 11.38% | -13.35% | 1.07 | 1.51 | 1.05 |
| C. Cash 60 | 793.31% | 14.00% | 11.39% | -13.14% | 1.08 | 1.52 | 1.07 |
| D. Cash 50 | 800.41% | 14.06% | 11.40% | -12.94% | 1.08 | 1.52 | 1.09 |
| E. Cash 40 | 807.52% | 14.11% | 11.41% | -12.73% | 1.08 | 1.53 | 1.11 |
| F. Cash 50, gold-max | 810.76% | 14.13% | 11.43% | -13.07% | 1.08 | 1.53 | 1.08 |
| G. Cash 50, dollar-tilted | 789.95% | 13.98% | 11.38% | -12.87% | 1.07 | 1.52 | 1.09 |
| H. Balanced-3 | 778.24% | 13.89% | 11.37% | -13.41% | 1.07 | 1.51 | 1.04 |
| I. Diversified hedge | 784.34% | 13.93% | 11.38% | -13.20% | 1.07 | 1.51 | 1.06 |
| J. Duration-first | 770.27% | 13.82% | 11.37% | -13.48% | 1.06 | 1.50 | 1.03 |
Change versus production
| Scenario | Δ CAGR | Δ Max DD | Δ Sharpe | Δ Worst Episode |
|---|---|---|---|---|
| B. Cash 70 | +0.15 pp | +0.26 pp | +0.01 | -0.35 pp |
| C. Cash 60 | +0.20 pp | +0.47 pp | +0.02 | -0.41 pp |
| D. Cash 50 | +0.26 pp | +0.67 pp | +0.02 | -0.47 pp |
| E. Cash 40 | +0.31 pp | +0.88 pp | +0.02 | -0.54 pp |
| F. Cash 50, gold-max | +0.33 pp | +0.54 pp | +0.02 | -1.01 pp |
| G. Cash 50, dollar-tilted | +0.18 pp | +0.74 pp | +0.01 | -0.14 pp |
| H. Balanced-3 | +0.09 pp | +0.20 pp | +0.01 | -0.23 pp |
| I. Diversified hedge | +0.13 pp | +0.41 pp | +0.01 | -0.36 pp |
| J. Duration-first | +0.02 pp | +0.13 pp | +0.00 | -0.33 pp |
Bunker-Conditional Performance
Computed over the 181 sessions on which bunker weights were held — the only window in which the scenarios differ at all.
| Scenario | Ann. Return | Volatility | Max DD | Worst Day | Worst Episode | Episode Hit Rate |
|---|---|---|---|---|---|---|
| A. Production (frozen) | 1.92% | 0.97% | -0.61% | -0.25% | -0.09% | 78.6% |
| B. Cash 70 | 4.98% | 3.05% | -2.09% | -0.69% | -0.44% | 71.4% |
| C. Cash 60 | 6.15% | 3.82% | -2.62% | -0.89% | -0.50% | 71.4% |
| D. Cash 50 | 7.33% | 4.60% | -3.14% | -1.09% | -0.57% | 71.4% |
| E. Cash 40 | 8.52% | 5.38% | -3.66% | -1.29% | -0.63% | 78.6% |
| F. Cash 50, gold-max | 9.06% | 6.11% | -4.18% | -1.29% | -1.10% | 71.4% |
| G. Cash 50, dollar-tilted | 5.60% | 3.43% | -2.18% | -0.90% | -0.23% | 64.3% |
| H. Balanced-3 | 3.60% | 2.66% | -1.97% | -0.61% | -0.33% | 64.3% |
| I. Diversified hedge | 4.58% | 3.62% | -2.71% | -0.84% | -0.45% | 64.3% |
| J. Duration-first | 2.24% | 2.50% | -1.98% | -0.52% | -0.42% | 71.4% |
Episode-Level Attribution
Every bunker episode in the sample, with the raw hedge returns that drove it. Base is
A. Production (frozen), D is D. Cash 50.
| # | Start | Sessions | Vol cap | GLD | UUP | IEF | Base | D | Edge |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 2011-08-11 | 16 | 100% | +1.88% | -0.39% | +0.36% | +0.02% | +0.27% | +0.25 pp |
| 2 | 2014-10-16 | 7 | 0% | -0.54% | +0.85% | -1.04% | +0.03% | -0.01% | -0.05 pp |
| 3 | 2015-01-08 | 21 | 0% | +1.90% | +2.75% | +0.14% | +0.35% | +1.13% | +0.78 pp |
| 4 | 2018-12-31 | 6 | 100% | +0.39% | -0.28% | +0.03% | +0.02% | +0.05% | +0.03 pp |
| 5 | 2019-08-20 | 13 | 0% | +0.57% | +0.22% | +0.46% | +0.11% | +0.26% | +0.15 pp |
| 6 | 2020-03-04 | 19 | 100% | -0.63% | +2.16% | +2.42% | +0.22% | +0.29% | +0.07 pp |
| 7 | 2022-01-31 | 9 | 0% | +2.07% | -1.69% | -2.19% | -0.09% | +0.25% | +0.35 pp |
| 8 | 2022-02-17 | 31 | 16% | +2.65% | +2.76% | -3.09% | +0.35% | +1.26% | +0.91 pp |
| 9 | 2022-04-29 | 5 | 40% | -1.02% | -0.04% | -1.52% | -0.08% | -0.33% | -0.25 pp |
| 10 | 2022-05-11 | 18 | 100% | +0.23% | -1.50% | -0.26% | -0.05% | -0.09% | -0.04 pp |
| 11 | 2022-06-17 | 17 | 88% | -6.42% | +4.19% | +2.95% | +0.05% | -0.57% | -0.62 pp |
| 12 | 2024-08-08 | 7 | 0% | +5.19% | -0.49% | +0.53% | +0.31% | +1.49% | +1.18 pp |
| 13 | 2025-04-10 | 5 | 100% | +7.74% | -3.58% | +0.34% | +0.08% | +0.83% | +0.75 pp |
| 14 | 2026-04-01 | 7 | 0% | +1.59% | -1.22% | +0.16% | +0.04% | +0.28% | +0.24 pp |
Gold was positive in 10 of 14 episodes (median +1.08%, worst -6.42%). The dollar was positive in 6 of 14 (median -0.16%).
Statistical strength
Scenario D's total episode-level edge is +3.75 pp over 16.7 years (+0.224 pp/yr). Resampling the 14 episodes 10,000 times gives a 90% interval of [+0.88, +6.68] pp ([+0.053, +0.400] pp/yr), with a 1.6% probability the edge is non-positive. The three best episodes supply 77% of it.
The edge is positive but rests on 14 episodes totalling 181 sessions. Treat it as a weak prior, not an established result.
The Hedge Selection Problem
Behaviour on the 50 worst SPY sessions since 2010 (SPY average -4.16%), alongside each hedge's standalone risk:
| Asset | Avg. return on worst equity days | Positive | Ann. Vol | Max DD |
|---|---|---|---|---|
| GLD | -0.06% | 50% | 16.79% | -45.56% |
| UUP | +0.15% | 62% | 7.32% | -19.00% |
| IEF | +0.47% | 80% | 6.58% | -23.92% |
| BIL | +0.00% | 38% | 0.29% | -0.43% |
Gold is not a day-to-day tail hedge: it was positive on only 50% of the worst equity sessions and averaged -0.06%. It carries 16.8% annualized volatility and a -45.6% maximum drawdown of its own. Gold hedges inflation and real-rate crises; Treasuries and the dollar hedge liquidity crises. The 2022 episodes and the 2020 episode in the table above stress opposite hedges, which is why no single concentrated hedge dominates.
Crisis Windows
Total return / maximum drawdown within each window. Bunker exposure inside the window is identical across scenarios (2011 downgrade 19%, 2018 Q4 2%, 2020 COVID 58%, 2022 inflation 40%).
| Scenario | 2011 downgrade | 2018 Q4 | 2020 COVID | 2022 inflation |
|---|---|---|---|---|
| A. Production (frozen) | -2.89% / -7.10% | -1.16% / -2.76% | -6.53% / -8.86% | -8.93% / -9.13% |
| B. Cash 70 | -2.73% / -7.10% | -1.14% / -2.76% | -6.87% / -8.86% | -8.82% / -9.57% |
| C. Cash 60 | -2.69% / -7.10% | -1.14% / -2.76% | -6.84% / -8.86% | -8.72% / -10.05% |
| D. Cash 50 | -2.65% / -7.10% | -1.15% / -2.76% | -6.81% / -8.86% | -8.62% / -10.53% |
| E. Cash 40 | -2.61% / -7.10% | -1.15% / -2.76% | -6.78% / -8.86% | -8.53% / -11.01% |
| F. Cash 50, gold-max | -2.53% / -7.10% | -1.13% / -2.76% | -7.34% / -8.97% | -8.70% / -11.46% |
| G. Cash 50, dollar-tilted | -2.77% / -7.10% | -1.16% / -2.76% | -6.29% / -8.86% | -8.55% / -9.60% |
| H. Balanced-3 | -2.76% / -7.10% | -1.13% / -2.76% | -6.68% / -8.86% | -9.22% / -9.41% |
| I. Diversified hedge | -2.71% / -7.10% | -1.12% / -2.76% | -6.60% / -8.86% | -9.30% / -10.04% |
| J. Duration-first | -2.79% / -7.10% | -1.11% / -2.76% | -6.48% / -8.86% | -9.62% / -9.82% |
Note the asymmetry in 2020: gold-heavy variants finish the bunker episode ahead but travel a deeper path through the window, because gold fell -12.35% during the March 2020 liquidity scramble while the dollar gained +7.36%.
Methodology and Limitations
Signals are generated once and reused, so the regime path, turnover timing and circuit-breaker state are identical across scenarios. Returns use adjusted-close ETF total returns. Bunker-conditional statistics attribute each session's return to the regime whose weights earned it: the signal detected at the close of T and traded at the close of T+1.
Data constraint. UUP and IEF have no usable history in this database before 2010-01-06, which is why the backtest starts in 2010. No scenario here has been tested through 2008 — the one episode in modern history where gold fell roughly 30% into a systemic equity crash. The sample contains one genuine liquidity crisis (March 2020).
Instrument risk. GLD, UUP and IEF carry expense ratios, tracking error and real price risk that a T-bill sleeve does not. A sleeve labelled capital preservation that holds 30% gold can lose money in a crisis; the current 85% cash sleeve structurally cannot lose more than a few tens of basis points.
These are modelled results on historical data, not a forecast or a guarantee.