# AwarenessFund — S&P 500 Comparative Framework (Spec Sheet)

**Author**: capital-markets-lead (ACG / A-C-Gee)
**Date**: 2026-05-18
**Status**: DRAFT TEMPLATE — REQUIRES EXPERT REVIEW BEFORE PRODUCTION USE
**Audience**: AwarenessFund partners + investor insiders

---

## §1 — Why S&P 500 is the primary benchmark (rationale)

| Reason | Substance |
|--------|-----------|
| **Investor mental model** | Every US retail + institutional partner thinks in "vs S&P." It is the implicit default. |
| **Capital-cost benchmark** | If AwarenessFund does not beat passive SPX after fees, the fund has no investor-rational reason to exist. |
| **Selection-test** | The "trillions of capex shifting" thesis claims the *4-vertical sleeve* outperforms broad market. S&P is the broad-market hypothesis being challenged. |
| **Liquidity** | SPY / VOO / IVV are infinitely scalable; transaction costs to replicate are ~0bps. |

**But S&P alone is not enough**: S&P 500 has its own AI exposure baked in (NVDA, MSFT, GOOGL, AVGO, etc. are ~30%+ of cap-weight). A "naive" AwarenessFund-vs-SPX comparison risks measuring **the wrong thing** — picks-and-shovels vs (already-AI-tilted index). We need a benchmark *ladder*.

---

## §2 — Benchmark ladder (4 tiers, each answering a different question)

| Tier | Benchmark | Question it answers |
|------|-----------|----------------------|
| **T1** | S&P 500 TR (`^SP500TR`) | "Do we beat the default passive choice?" |
| **T2** | Equal-weight S&P 500 (`RSP`) | "Do we beat the broad market after stripping out the mega-cap tilt that already captured part of the AI trade?" |
| **T3** | NASDAQ-100 (`QQQ`) | "Do we beat the tech-heavy benchmark that overweights the AI applications layer?" |
| **T4** | Sector ETF basket (custom: 25% XLU + 25% XLB + 25% SOXX + 25% DTCR) | "Do we beat a naïve picks-and-shovels passive replication?" |

**Decision rule**: a back-test variation is interesting only if it beats **all four tiers** on risk-adjusted basis. Beating T1 only = could be sector beta. Beating T1+T2+T3+T4 = stock-selection + theme-curation alpha.

---

## §3 — Sub-vertical benchmarks (for vertical-level attribution)

Each AwarenessFund vertical should be compared against its own naïve passive replication:

| Vertical | Passive replication benchmark | Active question |
|----------|--------------------------------|------------------|
| Mining | XLB (Materials) + PICK (global miners) — 50/50 | Did our specific mining pick-list beat passive mining beta? |
| Energy | XLU (Utilities) + XLE (Energy) — 70/30 (utility-tilt because our energy basket is utility-heavy) | Did our IPP + transmission + nuclear-tilt beat the default mix? |
| Chip-fab | SOXX (Semiconductors) | Did our equipment + materials cut beat full semi sector (which includes chip-designers we exclude)? |
| AI-endpoint | DTCR (Data-center & digital infra) + XLRE-DC-component | Did our specific endpoint pick-list beat the basket? |

**Implementation note**: XLRE doesn't isolate DC-REITs from broader REITs; the AwarenessFund proxy will use a constructed sub-index = (EQIX + DLR + IRM) market-cap-weighted as a baseline endpoint benchmark.

---

## §4 — Alignment methodology

### §4.1 Time-series alignment

| Issue | Resolution |
|-------|------------|
| All series must share the same trading calendar | Use NYSE calendar; foreign-listed tickers normalized to NYSE-day (close-to-close in local currency, then converted) |
| Benchmark TR vs price | Always use **total-return** series (`^SP500TR` not `^GSPC`) so dividends are included on both sides |
| Currency | All series USD-denominated; non-USD listings converted at WMR 4pm London fix (FRED `DEXUSEU` etc.) |
| Missing data | Forward-fill ≤ 3 trading days; longer gaps flagged in audit log |
| First / last day | Strategy and benchmark indexed to 100 on `start_date − 1`; metrics measured on `start_date → end_date` |

### §4.2 Risk-adjustment normalization

Outperformance is meaningful only after adjusting for risk. Required normalizations:

1. **Volatility-matched return**: also report return-per-unit-of-volatility (strategy_return / strategy_vol) vs same for each benchmark
2. **Beta-adjusted alpha (CAPM)**: Jensen's alpha vs S&P 500
3. **Multi-factor alpha (Fama-French 3 + Momentum)**: alpha after controlling for (Mkt-Rf, SMB, HML, MOM) factor exposures
4. **Drawdown-matched return**: also report CAGR / max_drawdown (Calmar) — controls for back-test windows where strategies "win" by taking more risk

---

## §5 — Attribution decomposition (the partner-facing explanation surface)

Every back-test run produces a single page that answers four investor questions:

### §5.1 "What drove the return?"
**Output**: Vertical-contribution chart (4 bars: Mining / Energy / Chip-fab / AI-endpoint), each bar = Σ(weight × period_return) over the back-test window. Sum of bars = total return.

### §5.2 "Which names mattered?"
**Output**: Top-10 contributors + Bottom-5 detractors table (ticker, vertical, weight%-avg, contribution_bps). Surfaces concentration and idiosyncratic risk.

### §5.3 "How does our risk compare?"
**Output**: 4-row table — strategy + 3 benchmarks (S&P 500, equal-weight S&P, sector basket) — columns: CAGR, volatility, max DD, Sharpe, Sortino, beta vs S&P. Visually shows the trade-off.

### §5.4 "Is the outperformance real or noise?"
**Output**: Statistical-significance section — t-stat on excess return vs S&P, bootstrap 95% CI on alpha, Sharpe-ratio standard error per Lo (2002). Explicit honest framing: "with 24 monthly observations, statistical power is low. Treat as descriptive, not inferential."

---

## §6 — What "outperformance" means in this context (definitional clarity)

**Hierarchy of outperformance claims** (weakest to strongest):

| Claim | Evidence required |
|-------|--------------------|
| L1 — "Beat S&P 500 cumulative return" | Strategy_CAGR > SPX_CAGR over window |
| L2 — "Beat S&P 500 on risk-adjusted basis" | Strategy_Sharpe > SPX_Sharpe AND Strategy_Sortino > SPX_Sortino |
| L3 — "Beat the benchmark ladder" | L2 holds vs all 4 tiers (T1–T4 in §2) |
| L4 — "Beat after factor adjustment" | L3 holds AND multi-factor alpha is positive |
| L5 — "Beat after factor adjustment AND statistically significant" | L4 holds AND t-stat on alpha > 2 (or non-zero bootstrap 95% CI) |

**For partner discussion**: AwarenessFund should target L3 minimum (beats benchmark ladder on risk-adjusted basis). L4 is "this is a real strategy." L5 is "this is a real strategy with credible inference" — unlikely with 24 monthly obs; honestly framed.

**Reportable failure modes** (publish even when ugly):
- "Beat S&P 500 cumulative but lost on Sharpe" → strategy added risk
- "Beat by mega-cap concentration" → not stock-selection alpha
- "Beat in one vertical, dragged by another" → asymmetric thesis-validation
- "Beat early, gave back late" → momentum-driven, not durable

---

## §7 — Rolling-window robustness checks

Single-window back-test = single observation. Robustness comes from rolling windows.

| Check | Implementation |
|-------|----------------|
| Rolling 12-month return vs S&P | Plot rolling 12mo excess return; surfaces regime sensitivity |
| Rolling 12-month Sharpe vs S&P | Was outperformance consistent or episodic? |
| Sub-period attribution | Split 24mo into 4 × 6mo periods; report metrics in each |
| Hit-rate of months | % of months where strategy > S&P, by variation |
| Worst 3-month drawdown vs S&P worst 3-month | "When SPX was at its worst, where were we?" |

---

## §8 — Comparison-output format (single canonical page)

Every back-test variation produces a 1-page benchmark-comparison artifact with:

1. **Headline row**: Variation name, CAGR, Sharpe, Max DD, Excess return vs S&P, Statistical significance flag
2. **Benchmark ladder table**: Strategy vs T1/T2/T3/T4 on 6 metrics (CAGR, vol, Sharpe, Sortino, max DD, beta vs S&P)
3. **Vertical attribution table**: 4 rows (Mining / Energy / Chip-fab / AI-endpoint), columns: avg weight, return, contribution to total
4. **Top contributors / detractors table**: 10 + 5 names
5. **Rolling 12mo excess-return chart spec** (visual artifact — generated separately by back-test engine)
6. **Statistical-significance footer**: t-stats, bootstrap CIs, honest power caveat
7. **Disclaimers footer**: DRAFT-TEMPLATE + Past-performance-does-not-predict-future-results

Format: Markdown for human-read + JSON for machine-read; both committed to `projects/awareness-fund/backtest-runs/YYYY-MM-DD-runID/`.

---

## §9 — Honesty constraints (anti-pattern guard)

This spec MUST NOT enable any of these:

- ❌ Cherry-picking the start date to flatter the strategy (window is fixed: 2024-05-18 → 2026-05-18, set in protocol-spec §2)
- ❌ Reporting only the winning variation (ALL variations published, including losers)
- ❌ Reporting only the favorable benchmark (full ladder always published)
- ❌ Hiding survivorship-bias caveats in a footnote (explicit in every artifact)
- ❌ Cumulative-return-only reporting (Sharpe / max-DD / beta always alongside)
- ❌ "Future" extrapolations of CAGR (back-test is historical only; forward language explicitly prohibited)

**Sister-AiCIV-readable principle**: Aether, Witness, Parallax, Keel, Synth, CommonGround should be able to grade our back-test methodology by these constraints. If a peer can't reconstruct our outperformance claim from the protocol, we failed.

---

**Status**: SPEC v1 — comparative framework ready. Plugs into back-test protocol §5 metrics.