Sharpe Decomposition: Why the Winners Win

Date: 2026-02-26 Author: OLTA Research Abstract: The high-Sharpe baskets in the OLTA library each have a single dominant driver. The Semiconductor basket reflects the AI buildout. The Mag 7 basket reflects mega-cap tech dominance. The AI Infrastructure basket reflects the picks-and-shovels layer beneath the chip designers. The Sharpe-Max Diversified basket is the explicit hand-pick of those drivers into a single cross-asset basket. The BTC-Decorrelated Diversified basket is the cross-asset rebalance applied to the same drivers. This paper traces the Sharpe to the underlying themes at the framework level.

1. Methodology

For each high-Sharpe basket, we identify:

• The single asset or sub-basket that contributed the largest share of return.
• The structural reason that asset performed.
• The correlation structure that prevented the basket from inheriting a single-asset drawdown.

Numbers come from the consolidated backtest results. Per-constituent weights for each named basket are published in the methodology brief on request to institutional counterparties and grant reviewers.

2. OSEM7 (Semis 7) Sharpe in the [2.2, 2.4] band

Window: 980 days (multi-year intersection of usable equity history and live basket history) Total return: roughly +350 percent Volatility: around 33 percent Max drawdown: around -35 percent

Driver: the AI buildout. Specifically, the dominant GPU designer.

The largest constituent returned roughly an order of magnitude over the basket window and is the principal contributor to the basket return. The second-tier names (custom-silicon designer, leading foundry) contributed materially. The remaining constituents filled the diversification slots.

The reason: every dollar of capex spent on the global AI build-out passed through at least one of these companies. The chip designers, the foundry that fabricates almost everything, the lithography supplier, the architecture licensee, and the edge-compute anchor are the names that captured the supercycle.

Why this is a high Sharpe, not just a high return.

The basket-level volatility is the key. The dominant constituent alone has materially higher annualised volatility. A single-stock basket would deliver roughly the same return at much higher volatility, producing a lower Sharpe. The constituent cap cuts basket volatility while sacrificing modest return. The other constituents fill the diversification slots.

The Sharpe is the consequence of:

1. A genuinely strong driver (the AI semiconductor supercycle).
2. A capping rule that prevents the basket from being a single-stock proxy.
3. A correlation structure where the seven names are correlated to each other (all participate in AI capex) but uncorrelated to the broader market in 2020-2022.

Drawdown reading. The drawdown reflects the late-2021 to mid-2022 selloff in tech generally and semiconductors specifically. The recovery happened by mid-2023 and the basket then ran to new highs.

Forward expectation. Out-of-sample Sharpe should be lower. The current headline reflects a specific supercycle. Forward Sharpe is probably in the [1.0, 1.5] band if AI capex continues; lower if it plateaus.

3. OMAG7 (Mag 7) Sharpe in the [1.3, 1.4] band

Window: roughly five years Total return: roughly +315 percent Volatility: around 25 percent Max drawdown: around -45 percent

Driver: mega-cap tech dominance.

The Magnificent 7 returned roughly 4x cumulatively as a market-cap basket over the five-year window. The dominant constituent contributed the largest share at the constituent cap. The other names contributed the bulk of the rest. Two constituents had violent paths but ended materially higher.

The reason: roughly half of US equity-market returns since 2023 came from these seven names. The phrase "Magnificent 7" was coined in 2023 because every major equity benchmark was being driven by the same handful of stocks. The trend continued as AI revenue began landing at the largest cap-weighted earners.

Why this is a lower Sharpe than Semis.

The basket has broader coverage than Semis: it includes consumer-internet, an OS-and-hardware giant, a cloud-and-software giant, and a vehicle / energy concern. The diversification across business models is wider than Semis, which lowers volatility, but also lowers concentration in the strongest single driver. The basket is more conservative and less peaked.

The drawdown is the 2022 bear market hitting tech generally. The recovery happened by mid-2024. The path from the 2022 trough to the 2026 peak is materially positive on the basket.

Forward expectation. Out-of-sample Sharpe should be in the [0.8, 1.3] band. The basket benefits from being mega-cap (multiple decades of operating history, durable cash flows) but suffers from the same valuation re-rating risk as any tech basket.

4. OAIE6 (AI Infrastructure Stack) Sharpe in the [1.4, 1.5] band

Window: roughly five years Total return: roughly +680 percent Volatility: around 35 percent Max drawdown: around -43 percent

Driver: one layer beneath the GPU designers. The picks-and-shovels.

The basket pairs the equipment makers that let foundries produce advanced chips with hyperscaler-grade networking, custom-silicon partners for ASIC-bound workloads, server builders that physically roll into racks, and the memory supplier that bottlenecks AI training.

A single high-growth server constituent dominated the return contribution. The semiconductor equipment names each returned roughly 3x and contributed meaningful percentage points.

The reason: the AI-infrastructure thesis is broader than any single chip designer. Every hyperscaler needs networking, memory, servers, and the equipment to manufacture custom silicon. The basket is the second-derivative of the AI thesis, capturing companies that benefit from the buildout without taking concentrated exposure to any one chip designer.

Why this is in the same Sharpe band as OMAG7.

Higher return than OMAG7 but higher volatility too. The constituent business models are narrower (all six are infrastructure-tier industrial equipment makers), so they correlate more tightly with each other than the Mag 7 do. When the AI buildout pauses, the entire basket would pause together.

The Sharpe is roughly the same as OMAG7 because the higher return and the higher volatility cancel out. The Sortino is modestly higher than OMAG7's.

Drawdown reading. The drawdown is also the 2022 bear market. The recovery was sharper than OMAG7 because the AI-infrastructure thesis caught a stronger bid in 2023.

Forward expectation. This basket has the most thematic concentration of the high-Sharpe Equities baskets. Out-of-sample Sharpe depends materially on whether AI infrastructure capex continues. Bull case in the [1.4, 1.6] band; bear case in the [0.4, 0.6] band.

5. OSHARP6 (Sharpe Max) Sharpe in the [1.5, 1.6] band

Window: 729 days Total return: in the high-double-digit range Volatility: around 23 percent Max drawdown: around -25 percent

Driver: explicit hand-pick of the highest-Sharpe single-asset positions in the OLTA dataset.

This is the Sharpe-maximised construction. Each constituent was selected because, on its own over the relevant historical window, it carried a Sharpe well above the OLTA universe median. The basket combines a small set of high-Sharpe equity anchors, a gold leg with a multi-year Sharpe near unity, and a capped BTC and ETH sleeve.

Combining several positive-Sharpe assets that average roughly 0.85 into a basket should deliver a basket Sharpe close to that. The realised basket Sharpe is materially higher than the constituent average. Why?

The cross-asset uncorrelation does the work.

The basket is constructed for low pairwise correlation across legs. The within-equity correlations are typical for an equity basket. The cross-asset correlations (equity to gold, equity to crypto, gold to crypto) are near zero. Combining low-correlation positive-Sharpe assets cuts basket volatility roughly in half compared to a single-asset position at the same return level.

In short: the construction takes positive-Sharpe assets that are weakly correlated and combines them into a basket whose Sharpe exceeds any single one's. This is a textbook diversification result applied to a 2026 cross-asset menu.

Why not 100 percent of the dominant constituent?

A single-stock position over the same window would return more in absolute terms but at much higher volatility. The diversified construction beats it because the volatility reduction is larger than the return reduction.

The Sharpe-optimised basket is the closest the OLTA library gets to a tangent-portfolio expression of the single-asset opportunity set.

Forward expectation. Out-of-sample Sharpe should be in the [0.8, 1.2] band in a normal regime. The headline is partly in-sample because the construction explicitly selected the best in-sample assets. The decay from in-sample to out-of-sample is typically 30 to 50 percent in Sharpe terms.

6. OCRLOW7 (BTC-Decorrelated) Sharpe in the [1.3, 1.4] band

Window: 729 days Total return: in the mid-fifties percent Volatility: just under 19 percent, the lowest of the family Max drawdown: around -23 percent

Driver: same family of equity, gold, and crypto themes as OSHARP6, with the equity / gold weight raised and the crypto weight cut.

The basket replaces the OSHARP6 equity composition with a broader equity set including mega-cap tech names and a broad-market sleeve. The gold weight is raised modestly and the crypto weight is cut. The result: lower return at materially lower volatility.

The Sharpe is the second-highest of the family. The construction explicitly targeted low BTC correlation, which it achieved by raising the structurally-uncorrelated equity and gold legs.

Why this is below the Sharpe leader.

The trade-off is the same as comparing OMAG7 to OSEM7: broader diversification cuts both return and volatility. OCRLOW7 cuts volatility more than it cuts return, but only just. The Sharpe is slightly lower than OSHARP6.

The advantage of OCRLOW7 is the lower drawdown and the lower volatility footprint. For an allocator who wants Diversified-grade Sharpe but with the smoothest possible path, OCRLOW7 is the choice.

Forward expectation. Out-of-sample Sharpe in the [0.8, 1.1] band, similar to OSHARP6 but slightly more resilient to a tech-stock drawdown because the heavier gold leg is more stable.

7. Why the lower-Sharpe Diversified baskets are still high-quality

The mid-band Diversified members are below the leaders but still well above BTC's reference.

OQUAL7 trades Sharpe for breadth: a broad-market sleeve adds beta that lowers Sharpe slightly but adds resilience to a tech-specific drawdown. The total return is the smallest of the family but the path is the smoothest (the smallest maximum drawdown in the family).

OBAL6 trades Sharpe for accessibility: the basket is the "one ticker does the asset-class spectrum" expression. Cross-asset balance with no thematic concentration.

OAILEAD8 trades Sharpe for thematic concentration: the chip-stack-plus-data-center weight is a high-conviction AI-buildout expression. Volatility is the family's highest, which lifts the denominator and lowers the Sharpe to the [0.95, 1.10] band. The basket also has the family's highest total return under its preferred cadence.

8. Cross-reference: the Equities high-Sharpe leaders

The Equities family in approximate Sharpe order (for baskets with windows that exceed 365 days):

• OSEM7 in the [2.2, 2.4] band, anchored by the semiconductor supercycle.
• OINDDEF4 in the [1.4, 1.5] band, anchored by defense, data-center power, and rare earth.
• OAIE6 in the [1.4, 1.5] band, anchored by picks-and-shovels.
• OMAG7 in the [1.3, 1.4] band, anchored by mega-cap tech dominance.
• OCMD7 in the [0.9, 1.0] band, anchored by gold and silver.
• OEN5 in the [0.8, 0.9] band, anchored by energy majors.
• OFRONT3 in the [0.8, 0.9] band, three pre-revenue stories.
• OETFB5 in the [0.75, 0.85] band, broad market.
• OBIO6 in the [0.6, 0.7] band, healthcare.
• OAIS8 in the [0.55, 0.65] band, AI software.

Each high-Sharpe basket can be summarised in one sentence about what drove it. The pattern is consistent: a single dominant thematic driver anchors each basket. The diversification within each basket prevents single-stock blowup, but the cross-basket correlation to that single theme is what determines the Sharpe.

9. The implication for portfolio construction

The Diversified family does what the Equities family teaches: combine high-Sharpe themes into a single basket, weight each theme to control volatility contribution, and let the cross-thematic correlation do the work.

OSHARP6 is the most concentrated expression of this principle: the AI thesis, second-tier AI and crossover defense, a healthcare anchor, the macro gold hedge, and a capped crypto sleeve.

The basket Sharpe is the natural consequence of combining low-correlation positive-Sharpe assets into a single basket. The framework is general; the specific blend is documented in the methodology brief.

10. Caveats

• In-sample selection. OSHARP6 was selected by picking the best-Sharpe single-asset positions in the 2y window. This is in-sample by definition. Out-of-sample Sharpe is meaningfully lower.
• Driver attribution is approximate. The per-constituent return contributions discussed above are based on standalone returns times approximate weights, ignoring the rebalance effect. Exact attribution depends on the path; the engine does not produce per-constituent attribution natively.
• Correlation structure is window-specific. Cross-asset correlations shift modestly across windows. The window choice affects the basket Sharpe within a narrow band.
• Forward expectations are point estimates. The Sharpe ranges given for each basket are rough forecasts under "normal regime" assumptions. A regime change would shift them.

11. References

• Per-index full results: consolidated backtest exports
• Constituent definitions and weights: methodology brief on request
• Engine source: lib/backtest.js
• Cross-reference: 03-diversified-family-paper.md for the Diversified design philosophy
• Cross-reference: 06-outperformance-vs-btc.md for the BTC-relative ranking