Exploratory 3x S&P diverse-indicator ensemble under a 25% drawdown cap

Objective

Issue #1310 asked for the highest returns from a 3x S&P leveraged ETF using the repository's best conceptually diverse indicators, while limiting maximum drawdown to 25% and allowing exposure scaling.

Approach

Ran a research campaign (leveraged-sp500-indicator-ensemble) that:

1. Mapped existing trend, fast-risk, breadth, and lagged-macro signal families.
2. Built a causal UPRO/cash ensemble with one-bar execution lag and optional volatility scaling.
3. Extended pre-2009 3x history with the repository synthetic levered path and labeled proxies.
4. Ranked 13,203 deterministic trials and kept only variants with max drawdown ≥ -25%.
5. Stress-tested the winner on timing, costs, ablations, regimes, and crisis windows.
6. Published an interactive Other / Research page plus an executed notebook analysis.

Files changed

• research/leveraged_sp500_ensemble.py, research/leveraged_sp500_diagnostics.py
• research/campaigns/leveraged-sp500-indicator-ensemble/campaign.json
• research/experiments/1310.ipynb, research/campaigns/leveraged-sp500-indicator-ensemble/diagnostics.json
• docs/research/leveraged-sp500-indicator-ensemble.html, docs/site-data/research/leveraged-sp500-indicator-ensemble.json
• docs/experiments/1310-analysis.html
• tests/test_leveraged_sp500_ensemble.py

Results

• Winner l3x.c5_vol_scaled_ensemble.90969248f431: 8.32% CAGR, -24.78% max drawdown, average UPRO weight ~0.38 on 1999-03-05 to 2026-07-24.
• Actual-UPRO overlap era: 12.75% CAGR, -24.26% drawdown.
• Feasible set: 80 of 13,203 trials.
• All four signal families matter in ablations; same-close lookahead is non-deployable; 10 bps costs and two-bar lag breach the drawdown floor.

Conclusions

The campaign found a historically feasible exploratory UPRO/cash rule that meets the hard drawdown and vehicle constraints. It does not claim an unqualified “extremely high” CAGR or catalog promotion. Thin operational margins, synthetic-prefix weakness, and in-sample selection keep this in the Research Library.

Owner trusted feedback after the exploratory close raised the return bar to CAGR north of 30% and authorized aggressive in-sample/hindsight fitting plus drawdown-episode indicator search. This run lacks remaining planning-visit budget for a full re-loop, so that work continues on a child issue rather than being claimed complete here.

Links

• Research page: docs/research/leveraged-sp500-indicator-ensemble.html
• Notebook analysis: docs/experiments/1310-analysis.html
• Issue: #1310