# Morrowmix preset research

Enumerated 43,758 long-only 10% grid allocations over nine assets. Required equity shares: 20/40/60/80/100%. At most six holdings, BRK.B capped at 10%, SMH capped at 20%, other satellite equities capped at 30%. VOO core, Treasury minimums and explicit growth-tilt constraints vary by preset. Selection maximizes the worse of base/technology-weaker illustrative expected utility using training-period covariance shrunk 25% toward its diagonal. No historical mean returns enter the objective. Not global or personalized optimality.

**These are constrained research candidates, not universally best portfolios. No personal financial information is included. VTI is excluded.**

## Selected candidates

| Preset | Allocation | Feasible candidates | Full-window CAGR | Full-window max drawdown | Later-period CAGR |
|---|---|---:|---:|---:|---:|
| Shelter | VOO 10% / SCHD 10% / SGOV 80% | 21 | 5.81% | 3.56% | 6.94% |
| Steady | VOO 20% / SCHD 20% / VGIT 20% / SGOV 40% | 63 | 7.59% | 10.16% | 8.97% |
| Balance | VOO 30% / SCHD 20% / BRK.B 10% / VGIT 20% / SGOV 20% | 201 | 10.33% | 14.07% | 11.44% |
| Stride | VOO 30% / QQQM 10% / SCHD 30% / BRK.B 10% / VGIT 10% / SGOV 10% | 424 | 13.58% | 17.90% | 14.63% |
| Venture | VOO 40% / QQQM 10% / VGT 10% / SCHD 30% / BRK.B 10% | 610 | 17.44% | 22.62% | 19.12% |

Selection used covariance estimated only from November 2020 through December 2023. Later-period results start at the December 29, 2023 close and end September 30, 2026. This retrospective split does not constitute prospective out-of-sample validation. Common actual-fund data starts October 30, 2020. No pre-inception prices or proxies were used. Neither 2000–02 nor 2008 appears in this sample.

## Explicit objectives and constraints

Selection expected annual arithmetic nominal return inputs: equities 7.5%, VGIT 3.5%, SGOV 2.5%. The weaker-technology scenario reduces QQQM, VGT and SMH by two percentage points. These are illustrative assumptions, not historical estimates or predictions. The score is min(base expected return, weaker-tech expected return) − (1 + risk penalty) × annual variance / 2. Risk penalties are 16/8/4/2/0.5 across the five profiles. Covariance is shrunk 25% to its diagonal.

Required equity exposure is exactly 20/40/60/80/100%; at most six positions; BRK.B <=10%; SMH <=20%; other non-core stock sleeves <=30%. VOO minimums are 10/20/30/30/30%; growth satellite minimums are 0/0/0/10/20%. SGOV minimums are 60/20/0/0/0%; VGIT minimums are 0/20/20/10/0%. The growth tilts are preference constraints, not findings that technology must outperform. The 10% grid is coarse, and only this finite candidate set is searched.

SMH and AVUV were considered but did not win this particular conservative expected-return/variance objective. This is not a finding that those funds are inferior. SMH can outperform dramatically when a semiconductor premium materializes; it also increases sector concentration. VGT and QQQM compete with each other and overlap the VOO core. BRK.B is one company.

## Model separation

Daily adjusted closes, distributions reinvested, annual rebalance before first new-year return. Nominal returns; no contributions, taxes or trading costs. Fund operating expenses are embedded in price history. Daily-close maximum drawdown; monthly-return volatility annualized. The future calculator uses separate editable assumptions.

The future calculator defaults use editable illustrative geometric nominal returns of 7% for equity funds, 3.5% for VGIT/BND and 2.5% for SGOV/CASH. They differ from the arithmetic utility inputs used for selection. Future probabilities come from 2000 paths and are conditional scenarios, not forecast probabilities. The empirical residuals do not contain every possible crisis.

The separate simulator supports technology-underperformance, globally stressed correlations, and a one-time 30% whole-portfolio shock via the existing stress control. The deterministic panel retains monthly rebalancing; the uncertainty simulator uses annual rebalancing and states that difference.

## Rejected alternatives and sensitivity

### Shelter
Next-scoring candidates: [{"weights": {"VOO": 10, "SCHD": 10, "VGIT": 10, "SGOV": 70}, "score": 0.02581609641484213}, {"weights": {"VOO": 10, "BRK-B": 10, "SGOV": 80}, "score": 0.025012073766732184}].
Using full-sample covariance instead would select: {"VOO": 10, "SCHD": 10, "SGOV": 80}. Later-period returns never enter the selection score.

### Steady
Next-scoring candidates: [{"weights": {"VOO": 20, "SCHD": 10, "BRK-B": 10, "VGIT": 20, "SGOV": 40}, "score": 0.025641195686554714}, {"weights": {"VOO": 20, "SCHD": 10, "BRK-B": 10, "VGIT": 30, "SGOV": 30}, "score": 0.024655986664217328}].
Using full-sample covariance instead would select: {"VOO": 20, "SCHD": 10, "BRK-B": 10, "VGIT": 20, "SGOV": 40}. Later-period returns never enter the selection score.

### Balance
Next-scoring candidates: [{"weights": {"VOO": 30, "SCHD": 30, "VGIT": 20, "SGOV": 20}, "score": 0.03177782156782463}, {"weights": {"VOO": 30, "SCHD": 20, "BRK-B": 10, "VGIT": 30, "SGOV": 10}, "score": 0.03164580714294643}].
Using full-sample covariance instead would select: {"VOO": 30, "SCHD": 20, "BRK-B": 10, "VGIT": 20, "SGOV": 20}. Later-period returns never enter the selection score.

### Stride
Next-scoring candidates: [{"weights": {"VOO": 30, "QQQM": 10, "SCHD": 30, "BRK-B": 10, "VGIT": 20}, "score": 0.0392914033790227}, {"weights": {"VOO": 30, "VGT": 10, "SCHD": 30, "BRK-B": 10, "VGIT": 10, "SGOV": 10}, "score": 0.039040234427196945}].
Using full-sample covariance instead would select: {"VOO": 30, "QQQM": 10, "SCHD": 30, "BRK-B": 10, "VGIT": 20}. Later-period returns never enter the selection score.

### Venture
Next-scoring candidates: [{"weights": {"VOO": 40, "QQQM": 20, "SCHD": 30, "BRK-B": 10}, "score": 0.05175712562914002}, {"weights": {"VOO": 30, "QQQM": 10, "VGT": 10, "SCHD": 30, "BRK-B": 10, "AVUV": 10}, "score": 0.05145443900962181}].
Using full-sample covariance instead would select: {"VOO": 40, "QQQM": 20, "SCHD": 30, "BRK-B": 10}. Later-period returns never enter the selection score.

## Overlap and data checks

The UI reports measured monthly return correlations and shared issuer-listed holdings, not a fabricated weighted holdings-overlap percentage. A full same-date holdings-level overlap measure is not available in this dataset. The distinction is explicit.

Historical prices are third-party Yahoo Finance split/distribution-adjusted closes. Issuer crosschecks for VOO, SCHD, SMH and QQQM were performed in the separate validation report. VGT, VGIT and SGOV also passed separate issuer spot checks. All eleven spot checks were within 0.15 percentage points of the referenced issuer returns; exact differences and sources are in issuer-checks.json. Independently computed year-segment endpoint identities passed for all five selected portfolios. Monthly calibration data and path parameters are shipped so the simulation is reproducible.

Raw public responses and their SHA-256 hashes remain in the local data/ directory; they contain no personal data.

## Sources

- [Vanguard VGT / VGIT product information](https://advisors.vanguard.com/investments/products/vgt/vanguard-information-technology-etf)
- [iShares SGOV product information](https://www.ishares.com/us/products/314116/ishares-0-3-month-treasury-bond-etf-sgov)
- [Yahoo Finance adjusted-price source](https://finance.yahoo.com/quote/VOO/history/)