Oneiros Vision

Performance Tear Sheet

1. Executive Summary

Oneiros Vision is a regime and risk-allocation overlay. The Master Danger Index (MDI) determines when market structure is supportive, transitional, fractured, or critical. Each regime has been followed by measurably different volatility and drawdown in the S&P 500, and this document measures how different, and what a fixed strategy following the regime would have earned.

This document is structured in two layers: Layer 1 evaluates the raw MDI signal independent of any portfolio implementation. Layer 2 demonstrates the signal applied to a fixed, illustrative ETF chassis ("Option A"). Sections 10 and 11 of the full tear sheet extend this audit with sensitivity studies on the CRITICAL sleeve composition and implementation robustness.

2. Layer 1: The Master Danger Index (MDI) Signal

What matters about a regime reading is what followed it. Forward volatility and drawdown are measured over the 21 sessions a position taken on each regime signal actually holds.

Forward Volatility & Drawdown by Regime

Key Finding: The MDI is monotonic on both forward volatility and forward drawdown. The defensive regimes (WARNING and CRITICAL) account for only 31.5% of all trading days, yet the regime actually held on the S&P 500's worst 1% of sessions was defensive on 76.7% of them.

3. Key Performance Indicators (KPIs)

All reported strategy returns are net of a 2 bps per-leg transaction-cost assumption.

Against a standard 60/40 capital preservation mandate, the reference implementation generates +4.1pp of annualized return while reducing maximum drawdown by 35%. Against the S&P 500 it trails on return (13.80% against 14.18%) at two-thirds of the volatility and two-fifths of the drawdown.

Methodology note: Sharpe and Sortino are computed from the annualized arithmetic mean of daily excess returns over volatility (resp. downside deviation), the standard convention for both ratios. Calmar uses CAGR / Maximum Drawdown.

3.1a Confidence Intervals on the Headline KPIs

Every figure above is a point estimate from one realised path of one strategy over one sixteen-year window. An analyst conducting due diligence will discount a point estimate and reconstruct its range privately. Publishing the range is the stronger position, not the weaker one.

Intervals are 95%, from a stationary block bootstrap with a mean block length of twenty sessions. The block preserves volatility clustering, which an ordinary bootstrap destroys and on which drawdown depends almost entirely.

Read the widths, not just the bounds. A Sharpe of 1.06 whose interval runs from 0.66 to 1.51 is a strategy whose central estimate is respectable and whose evidence does not exclude a materially weaker one. Sixteen years of daily data is a smaller sample than it appears, because what the strategy is measured on is a handful of regime cycles rather than four thousand independent days.

The intervals are computed on the same data as every other figure on this page. The point estimates are the strategy's own, never the bootstrap mean — resampling states how far a figure would move on another draw of the same process, and must not restate the figure itself. Win Rate is deliberately absent: it counts profitable calendar years, and a resampled series has no calendar.

3.1b The Same Track in Francs, Euros and Pounds

The strategy trades US-listed instruments and is measured in the currency they price in. A client who reports in another currency did not earn the figures above. They earned them compounded with what the dollar did against their own currency, and over this window that difference is larger than most of what separates the strategy from its benchmarks.

The franc drawdown is not a strategy drawdown. Its worst run was 11 February to 9 August 2011, while the franc ran towards parity with the euro before the Swiss National Bank set its floor that September. USD/CHF fell from 0.9731 to 0.7209 — the dollar lost 25.9% — and the strategy itself fell 7.60% over those same sessions. Compounded, the two produce the -30.11% above.

Sterling is the column where the currency helped. A pound-based client earned 15.18% a year against the dollar track’s 13.80%, because the dollar rose against the pound over this window. That +1.38 pp is not skill and is not repeatable; it is the same effect that cost a franc-based client 1.75 pp a year, running the other way. A currency that flattered the past is the one most likely to embarrass the future.

A currency-hedged implementation removes most of that component, at a carry cost equal to the interest-rate differential between the two currencies. This document does not model a hedged share class, and every figure in the table is unhedged. Restated from the same frozen dataset as every other figure here; USDCHF and EURUSD are engine inputs, and GBPUSD is read from a rate frozen and digested alongside it.

3.2 Capture Ratios & Worst-Month Behaviour

Over the 16.7-year period, the reference implementation achieved an Up-Capture of 72.4% and a Down-Capture of 49.4% relative to the S&P 500 (monthly returns, geometric-mean convention).

Showing 5 of the 10 worst SPY months; the average covers all ten. The strategy finished ahead of 60/40 in all ten. Full table available in the complete tear sheet documentation.

4. Complete Annual Returns Profile

†Partial period: January 1 – September 30, 2026. A capital preservation overlay has a low structural beta (0.42). During a 12.4% risk-on rally, trailing the index is the mathematical cost of maintaining a defensive posture.

5. Maximum Drawdown Analysis

Drawdowns are peak-to-trough within each stated window, computed on identical date ranges for all three series. 2008 predates the track and is reported as not measured rather than as no loss: a strategy that did not exist did not survive anything.

The two 2022 windows are not a duplicate. The benchmarks show the same drawdown in both because their trough came in October, so extending the window to December adds nothing to them. The strategy does not: -9.72% to October against -11.36% to December. The strategy’s worst point in 2022 came after the market’s. The 2015 Flash Crash is the counter-example — the strategy fell further than the 60/40 (-10.81% against -6.47%) and almost as far as the S&P 500 (-11.07%), and it is in the table for that reason, as are the 2016 referendum and the 2023 Credit Suisse weekend, where it also fell further than the 60/40.

6. Tail Risk Metrics

6.1 Beyond the Sample: Fitted Tail

An empirical quantile is a step function over the observations beyond it, and it can never exceed the worst day actually seen. Past the 99th percentile only a few dozen such observations exist, so the figure stops being a measurement and becomes an artefact of where the sample happened to stop.

A Generalized Pareto distribution is therefore fitted to the losses beyond the 95th percentile by peaks-over-threshold. The fitted shape parameter is +0.05 — a heavy tail. It is positive at every threshold tested, +0.03 at the 90th percentile and +0.16 at the 97.5th: the sign is robust, and the magnitude depends on where the threshold is drawn.

One caution, stated because it would otherwise mislead. A Cornish-Fisher expansion — the conventional correction of a normal quantile for skew and kurtosis — is not used for the far tail. With excess kurtosis near 5 it falls outside its usable range and overstates the far-tail loss, in the direction that looks prudent. The platform computes it, flags it invalid past the 95th percentile, and reports the fitted and empirical figures instead.

The fitted tail extrapolates; the empirical one does not. Both are shown because neither alone answers the question, and a reader is entitled to see where the measurement ends and the model begins.

7. Liquidity & Implementation Friction

The reference implementation deducts 2 bps per trade leg (round-trip) from every allocation change. Costs are charged on executed weights — traded at the close after each signal — and the trade that establishes the first position is charged like any other. Average annual total turnover was 13.46x (one-way turnover of 6.73x).

8. System Turnover & Volatility Profile

9. Regime-Attributed Return Analysis

Each session's return is attributed to the regime whose weights earned it — the signal traded at the previous close — and annualized from the compound daily returns earned while that regime was held.

10. Monthly Returns (%)

YTD is the calendar-year return, as in the annual returns table; for 2026 it runs to September 30 and is not a full year. ‡ Out-of-sample signal: since July 20, 2026 every regime call has been logged live at the time, in an append-only, hash-chained record, and the backtest's calls match it on all 52 sessions to September 30 (July counts from the 20th). The returns in those months are still those of the hypothetical reference implementation, not of a live account.

11. Sensitivity Studies

Two published studies test whether the results above depend on choices made in the reference implementation. Both hold the regime path fixed and change one thing; all figures are in US dollars.

CRITICAL (Bunker) Allocation Sensitivity varies only the CRITICAL sleeve, across ten mixes of cash, gold, the dollar and Treasuries. The CRITICAL regime was in force on 4.3% of sessions and accounted for 0.4% of the strategy's loss accumulation, so re-weighting it moves the headline results by fractions of a point; the reference strategy keeps its frozen sleeve.

Implementation Sensitivity varies the risk-sleeve ladder and the speed of trading. Across seven ladders the regime reduced maximum drawdown by 7.29 to 12.37 points against a matched static portfolio and added 0.86 to 1.51 points a year. A glide capping the largest single move at 10% a session kept at least 98% of the drawdown benefit, while calendar rebalancing was as likely to destroy it as to keep it.

Data Sources

Market data sourced from: FRED (Federal Reserve Economic Data), Yahoo Finance, CBOE, AAII. Treasury rate data from FRED series DGS3MO. ETF price data from Yahoo Finance.

Disclosures & Methodology

Metrics are calculated from actual trading sessions only. Portfolio and benchmark returns use adjusted (total-return) ETF closing prices. Two benchmarks are reported: SPY (S&P 500 total return), and 60/40 (a daily-rebalanced 60% SPY / 40% IEF proxy — no advisory fee, tax, or rebalancing-cost assumption is applied to it). Both are computed over the identical January 2010 – September 2026 window on the same trading calendar as the portfolio. The implementation trades each session's regime and circuit-breaker weights at the close of the following session and earns returns from the session after that, and deducts a 2 bps per-leg transaction cost from allocation turnover. Where history is unavailable, the model substitutes MTUM with QQQ (through 2013-04-18), QUAL with SPY (through 2013-07-18), and USMV with SPY (through 2011-10-20). VIG carries a complete daily price history across the full 2010–2026 backtest window and is never substituted.

Inputs are dated by publication: net liquidity (Federal Reserve H.4.1) and the ICE BofA high-yield spread enter the signal on the session after the date they describe, which is when they are published.

All performance results presented herein are derived exclusively from historical backtesting conducted from January 4, 2010 through September 30, 2026. Every figure and chart is generated from a frozen month-end dataset whose SHA-256 digests are recorded, so the document reproduces exactly. While no live trading track record exists for this strategy, live out-of-sample signal logging commenced on July 20, 2026. Risk-adjusted ratios use the 3-Month U.S. Treasury Yield (FRED: DGS3MO) as the risk-free rate. This document is for informational purposes only and does not constitute financial advice or an offer to sell securities. Past backtested performance is not indicative of future results.