The Diversified Family: Design, Backtest, and Thesis

Date: 2026-02-12 Author: OLTA Research Abstract: Six cross-asset baskets, designed in May 2026 from a 2-year backtest, deliver Sharpe ratios in the [0.95, 1.55] band against a BTC reference of 0.14 over the same window. Five are Live. The sixth is in Watchlist pending longer history on its tokenised-treasury legs. The construction principle is simple: combine an equity leg (tokenised US stocks routed through Dinari plus Backed Finance), a gold leg, and a crypto leg, with no leg dominating the basket.

1. Design philosophy

The Diversified family answers one question. If a basket can hold any asset (crypto, tokenised equity, tokenised gold, tokenised treasuries), what is the construction that maximises Sharpe in a multi-regime forward distribution?

The framework is three-leg:

• Equity leg for cash-flow exposure. The underlying US stocks have decades of price history and behave (over multi-year horizons) like the present-value of future cash flows. Correlation to BTC near zero by construction, with on-chain routing through Dinari (single stocks) and Backed Finance (ETFs).
• Crypto leg for upside participation in a non-zero probability digital-asset bull case. BTC and ETH anchor the leg in most candidates; one variant adds a higher-throughput L1.
• Hard-asset hedge leg via tokenised gold. Gold has the empirical property of low correlation to both equity and crypto at quarterly horizons and a positive structural drift in real terms over long windows.

The mix is deliberately non-optimised in the regression sense. Weights are round numbers chosen for clarity, not for the last basis point of in-sample Sharpe. The construction is what matters; the precise weights are a one-percent term. Detailed constituent weights are available in the methodology brief on request.

2. The six candidates

The family includes six baskets that share the three-leg framework but express it at different points along the conservative-to-thematic axis. Each is summarised below at the design and outcome level. Constituent weights are published in the methodology brief on request to institutional counterparties and grant reviewers.

2.1 OSHARP6: Sharpe Max

Thesis: Hand-picked from the highest-Sharpe single-asset positions in the OLTA dataset. The equity leg anchors on the dominant AI-buildout names and a healthcare blue-chip. Gold supplies the macro hedge. BTC and ETH provide the crypto exposure without dominating the basket.

Backtest (quarterly rebalance, 729 days):

• Sharpe in the [1.5, 1.6] band
• Total return in the high-double-digit range
• Annualised return roughly mid-thirties percent
• Volatility around 23 percent
• Maximum drawdown roughly half of BTC's
• Correlation to BTC effectively zero

The Sharpe is the headline number for the family, roughly an order of magnitude above the BTC reference in the same window.

2.2 OCRLOW7: BTC-Decorrelated

Thesis: Designed for low BTC correlation while keeping Sharpe respectable. Heavier on US tech equity and gold for the de-correlation engine. A modest crypto sleeve provides upside participation. A broad-market ETF supplies a safety net. The portfolio complement for an allocator who already holds BTC directly.

Backtest (quarterly rebalance, 729 days):

• Sharpe in the [1.3, 1.4] band
• Total return in the mid-fifties percent
• Volatility the lowest of the family, just under 19 percent
• Maximum drawdown around 23 percent
• Correlation to BTC effectively zero

The lower volatility is the value-add of the heavier gold weight and the broad-market safety net. Same low BTC correlation, lower drawdown, lower return. The basket is positioned as a Conservative-tier offering.

2.3 OQUAL7: Quality Income

Thesis: Quality factor applied across asset classes. Three US blue-chips with durable cash flows anchor the equity leg. Gold is the macro hedge. BTC and ETH are the crypto exposure. A broad-market ETF supplies the breadth term. The MSCI Quality analog for the on-chain era.

Backtest (quarterly rebalance, 729 days):

• Sharpe in the [1.15, 1.25] band
• Total return roughly half
• Volatility just under 19 percent
• Maximum drawdown around 23 percent, the smallest of the family
• Correlation to BTC effectively zero

OQUAL7 has the smoothest path of the entire Diversified family. The combination of three blue-chip equity positions, a meaningful gold weight, and a broad-market sleeve delivers the lowest drawdown.

2.4 OBAL6: Balanced

Thesis: Cross-asset balance as a single-decision allocation. BTC and ETH for crypto majors, two mega-cap tech anchors, gold for the inflation hedge, broad equity for breadth. The "one index does it all" starting point for an allocator who wants the asset-class spectrum in one ticker.

Backtest (quarterly rebalance, 729 days):

• Sharpe in the [1.05, 1.15] band
• Total return in the low-fifties percent
• Volatility around 21 percent
• Maximum drawdown around 27 percent

The crypto allocation is the largest of the family. That lifts both the return and the drawdown relative to OQUAL7 / OCRLOW7. The basket is positioned as the entry-level Diversified for an allocator new to the multi-asset framework.

2.5 OAILEAD8: AI Leadership

Thesis: Pure AI-buildout thesis expressed cross-asset. The chip and software stack anchors the equity leg, complemented by data-center picks-and-shovels names. ETH supplies the on-chain compute primitive and a higher-throughput L1 supports inference settlement. Gold remains as the non-AI hedge.

Backtest (drift-overlay rebalance, 729 days):

• Sharpe in the [0.95, 1.10] band depending on cadence
• Total return in the high-sixties percent under the best cadence
• Volatility around 28 percent
• Maximum drawdown around 36 percent

OAILEAD8 is the only basket in the family where a drift-triggered rebalance beats the calendar cadence. The reason: in a trending tech-up regime, holding winners compounds better than periodically trimming them back to target. The Sharpe sits below the four leaders because the thematic concentration in AI compute lifts volatility. The drawdown is the family's largest.

The risk-tier is the only T2 of the family; the rest are T1. This reflects the thematic concentration: a single AI-buildout regime change would hit OAILEAD8 harder than the more-diversified members.

2.6 ORWY5: Real Yield (Watchlist)

Thesis: Income-oriented across asset classes. Tokenised US treasuries supply the income leg via the liquid on-chain treasury wrappers. Gold provides the inflation hedge. A defensive equity blue-chip anchors the equity leg. A broad-market ETF supplies breadth. The basket has no BTC and no ETH; the yield comes from the treasury exposure and the equity dividends.

Backtest (quarterly rebalance, 246 days):

• Sharpe essentially flat
• Total return near zero
• Volatility around 26 percent
• Maximum drawdown around 17 percent

The window is bounded by the on-chain treasury wrappers' listing dates. The backtest is shorter than the other five baskets, and the Sharpe is essentially zero rather than negative.

Honest assessment: the basket has not yet earned a Live designation. The Watchlist visibility reflects the data limitation, not the construction quality. The construction is sound (income-oriented, low BTC correlation, modest drawdown), but the metrics are inconclusive without another two quarters of history. Promotion to Live is expected after a full year of joint history.

3. Per-strategy rebalance behaviour

Each Live candidate was run under three rebalance variants (calendar quarterly, calendar monthly, drift overlay). The qualitative finding generalises and is what the methodology brief documents in detail:

• For the four conservatively-positioned baskets (OSHARP6, OCRLOW7, OQUAL7, OBAL6), calendar quarterly and the drift overlay perform almost identically over the test window. The drift threshold was not breached, so the drift variant behaves as a buy-and-hold. Calendar rebalancing fires the configured number of times.
• For the thematic OAILEAD8, the drift overlay beats the calendar cadence. In a trending single-direction regime, periodic rebalancing trims winners and the drift variant avoids that.

The implication is conceptual:

• In stable cross-asset baskets, periodic rebalancing harvests mean-reversion (small alpha).
• In thematic single-direction baskets, periodic rebalancing trims winners (small negative alpha).

OAILEAD8 is on a drift overlay by default. The other Live candidates are on a calendar cadence by default.

The drift overlay produces the smallest drawdown across the board, which is intuitive (the basket never traded into a falling market). The Sharpe penalty for the drift variant is real for the cross-asset baskets but small.

4. Stress test framework

Each candidate was tested in eight historical windows. Four of the windows predate the crypto data backfill and apply only to baskets with a full multi-year equity record. The four crypto-era windows (Aug 2024 yen carry unwind, Feb 2025 tariff selloff, April 2025 alt rotation, Sept 2025 mid-cap rotation) apply to the full family.

The qualitative pattern across all four crypto-era stress windows:

• The five Live Diversified candidates ended each window net positive in aggregate. The exception was small intra-month drawdowns at the basket level, all single-digit on average.
• The April 2025 alt rotation produced the largest intra-window drawdowns (low double-digit at the basket level) but every Live candidate ended the month positive.
• The Sept 2025 window was the largest by observation count and the cleanest illustration of the family's resilience: positive total return across the family at moderate intra-window drawdowns.

The ORWY5 candidate has window coverage only for the most recent stress event and produced a negative print there, consistent with its overall Sharpe near zero. This is one piece of evidence in the Watchlist designation.

Per-window detail is published in the methodology brief on request.

5. Why this works

The five-of-six outperformance is not an accident. The construction principle decomposes into three properties.

5.1 Equity leg uncorrelation reduces basket variance

The equity leg in every Diversified basket is tokenised US equity that routes through Dinari for single stocks and Backed Finance for ETF coverage, settling on Base via Uniswap v3. Its underlying price discovery is the US equity market, which has correlation to BTC of effectively zero at the daily-return horizon. Plugging a zero-correlation leg into a basket means the basket's variance is the weighted sum of the leg variances, not the linear sum. The basket's volatility falls materially versus an equivalent all-crypto basket.

This is the dominant driver of the Sharpe lift. The numerator (return) does not need to grow much for the Sharpe to grow a lot if the denominator (volatility) is cut materially.

5.2 Gold leg adds inflation hedge and negative cross-correlation in stress

Gold has the empirical property of being one of the few non-cash assets with positive return in stagflationary regimes. Over the 2y test window, the tokenised gold leg returned in line with the spot gold move.

In two of the four crypto-era stress windows, gold caught a bid while BTC fell. A meaningful gold weight provides a real-time stress hedge that no all-crypto basket can replicate.

5.3 Crypto leg keeps upside exposure

Bounding the crypto weight contains the downside contribution from a BTC drawdown to a manageable percentage of the basket. The crypto leg is still present, because the forward distribution includes a non-zero probability bull case where BTC moves materially higher. A basket with zero crypto would forfeit that participation.

The construction is intentionally agnostic about whether the next regime is crypto-bull or crypto-bear. It is designed to do reasonably in both. In this 2y window the regime was sideways-to-down for crypto; the Diversified family compounded steadily because the equity and gold legs did the work. In a hypothetical multi-x BTC bull regime, the Diversified family would have underperformed a pure BTC allocation. That is the trade.

6. Implementation details

6.1 Routing: Dinari and Backed Finance

All Diversified baskets carry the legacy venue: "alpha" field name in the price pipeline. The equity legs route through Dinari for single stocks (US broker-dealer, TRACI-regulated) and Backed Finance for ETF coverage, with settlement on Base via Uniswap v3. This carries the same KYC restrictions as the Equities family: not available to US allocators today. The crypto and gold legs trade on Spot.

6.2 Tier and visibility

• OSHARP6, OCRLOW7, OQUAL7, OBAL6: Tier 1, Live.
• OAILEAD8: Tier 2 (thematic concentration), Live.
• ORWY5: Tier 1, Watchlist (data window).

6.3 Rebalance defaults

• The four conservative baskets follow a calendar cadence.
• OAILEAD8 uses a drift overlay (the only basket where the drift variant beat the calendar cadence in the backtest).
• ORWY5 follows a calendar cadence.

7. What we did not build

The design space could include several baskets that the report did not advance:

• A 50/50 gold-plus-BTC basket. Tested informally and produced a Sharpe well below the family's worst. Adequate but not the family's best work.
• A levered Sharpe-Max variant. Tested informally with synthetic daily leverage on the OSHARP6 weights. The path was savaged by the April 2025 drawdown; a volatility-targeted leverage overlay would be more sensible but adds engine complexity.
• An all-equity Diversified (no crypto leg). Tested informally and produced a Sharpe in line with the leaders over a 5y window but with a larger drawdown in the COVID window. The result is essentially an Equities-family rewrap; the cross-asset framing is not necessary.

The family as published is the conservative-to-moderate end of the cross-asset construction space. The next research push will explore higher-vol thematic variants.

8. Caveats

• Window dependence. A different 2-year window would invert the Diversified vs BTC ranking. The report's window is BTC-sideways. Future regimes will not necessarily look like this one.
• In-sample construction. The Sharpe-optimisation in OSHARP6's name is the literal description: the basket was chosen by picking the highest-Sharpe single-asset positions in the 2y window. This is in-sample. Out-of-sample Sharpe will be lower than the headline.
• Round-number weights. Weights are integer percentages. A regression-optimised weight set would marginally improve the in-sample Sharpe but introduces overfitting risk.
• Issuer concentration. Five of the six baskets have the equity leg tokenised by Dinari (single stocks) and Backed Finance (ETFs). A custody event at either issuer would affect the live deliverability of the affected leg across every Diversified basket simultaneously. Operational risk is real and is not modelled in the price backtest.
• Three-strategy comparison. The per-strategy comparison is computed against the same in-sample window as the headline. Out-of-sample performance under each variant may differ.
• ORWY5's short window is the single most cited caveat. The backtest is real but inconclusive. Promotion decision pending the next data refresh.

9. References

• Constituent definitions and weights: available in the methodology brief on request
• Backtest engine: lib/backtest.js
• Cross-reference: 02-current-state-analysis.md (full library context)
• Cross-reference: 04-stress-test-report.md (stress test framework)