AI Power Demand Strategy Research

Research-first v1 around electricity, grid, nuclear, cooling, and power infrastructure beneficiaries from AI data-center buildout. No paper or live orders were submitted.
Best candidate
vol_targeted_power_momentum_guard
Net Sharpe
1.006
Net CAGR
+11.41%
Net MaxDD
-21.19%

Best Candidate Equity Curve

Candidate Ranking

CandidateNet SharpeNet CAGRNet MaxDDOOS SharpeMax Existing Corr
vol_targeted_power_momentum_guard1.006+11.41%-21.19%1.5530.453
simple_power_thematic_momentum0.964+22.74%-31.35%1.2690.428
defensive_utility_grid_barbell_guard0.948+13.12%-24.22%1.5500.825
power_vs_ai_infra_relative_strength0.982+26.31%-35.26%1.1040.674
vol_adjusted_power_risk_parity0.874+17.42%-39.78%1.2970.657

Latest Target Weights

SymbolWeight
SHY+44.38%
TRGP+11.20%
VRT+8.99%
PWR+7.82%
GRID+6.62%
WMB+5.97%
KMI+5.60%
NXE+4.82%
PAVE+4.59%

Data and Timestamp Assumptions

Provider: yahoo_finance_yfinance. Window: 2018-01-02 to 2026-06-05. Signals for date t use adjusted close data shifted to t-1, and positions[t] earn close-to-close returns[t]. Current-universe survivorship remains a promotion blocker.

Decision

# Decision: EXP-ai-power-demand-v1

## Recommendation

`paper_review_candidate`

ready_for_paper_review: NO

## Best Candidate

`vol_targeted_power_momentum_guard`

The selected candidate has net Sharpe 1.006, net CAGR +11.41%, net max drawdown -21.19%, annual turnover 7.351, and max existing-strategy/proxy correlation 0.45319056601740043.

## Candidate Ranking

| candidate                            |   objective |   net_sharpe |   net_cagr |   net_max_drawdown |   oos_sharpe |   max_existing_corr | bias_pass   |
|:-------------------------------------|------------:|-------------:|-----------:|-------------------:|-------------:|--------------------:|:------------|
| vol_targeted_power_momentum_guard    |    1.30494  |     1.00635  |   0.114094 |          -0.211923 |      1.55259 |            0.453191 | True        |
| simple_power_thematic_momentum       |    1.10451  |     0.964154 |   0.227367 |          -0.313498 |      1.26929 |            0.427903 | True        |
| defensive_utility_grid_barbell_guard |    1.04009  |     0.947587 |   0.131215 |          -0.24222  |      1.54956 |            0.825184 | False       |
| power_vs_ai_infra_relative_strength  |    1.02577  |     0.981515 |   0.263099 |          -0.352562 |      1.10362 |            0.673908 | True        |
| vol_adjusted_power_risk_parity       |    0.928807 |     0.874218 |   0.174201 |          -0.397758 |      1.29692 |            0.657038 | True        |

## P0/P1/P2 Status

- P0 bias gate: True
- P0 DSR/PBO: needs_input; 5 candidate families were tested and no CPCV/PBO panel was run.
- P1 OOS: computed with split 2024-01-01.
- P1 portfolio interaction: computed against prior AI/power research where available plus QQQ/SOXX/SMH/XLU proxies.
- P2 live readiness: not applicable; no broker orders, fills, queue, latency, or paper-trading stream.

## Blockers

- No first-pass P0 blocker found, but hidden holdout and paper dry-run package are still missing.

## Promotion Gaps

- hidden holdout or walk-forward/CPCV with DSR/PBO is still missing
- promotion-grade Alpaca data and liquidity/spread checks are still missing

## Next Candidate Mutation

- Replace current-universe Yahoo data with Alpaca SIP/IEX adjusted daily data from ops credentials only if a human asks for promotion-grade data.
- Run walk-forward/CPCV with a fixed candidate set and deflated Sharpe/PBO estimate.
- Add explicit liquidity filter for `SMR`, `OKLO`, and other optionality names before any paper-review package.
- Compare marginal portfolio contribution against actual live/paper strategy returns, not only proxies.