A+B+Diversifier Sleeves — Live Trading Implementation

This folder (Live/) contains everything needed to run the A+B+Diversifier Sleeves strategy live with Alpaca. Nothing else is used for implementation; all production code lives here.

---

1. Strategy

Idea

Combine two equity dual-momentum engines (A and B) with three truly diversifying sleeves to build a robust, non-overfit, live-tradeable portfolio.

Target allocation (sleeves)

Sleeve	Weight	Description
Strategy A	20%	Phase 3 mom_corr: momentum + correlation, top 5
Strategy B	20%	Top-3 Dual-Momentum: pure momentum, top 3
Rates sleeve	20%	Bond trend: TLT / IEF / BIL
Bear sleeve	20%	SH when SPY < SMA200, otherwise BIL
CTA proxy	20%	PDBC / DBMF / KMLM in uptrend, otherwise BIL

Why this structure

• A and B are highly correlated, so switching rules between them add no OOS value.
• C (Triple EMA + Macro + Kurt) was discarded as a primary engine due to look-ahead and overfit risk.
• The three sleeves (Rates, Bear, CTA proxy) break correlation and approximately double Sharpe and triple Calmar in backtest.
• Gold is omitted because it degrades Calmar in the OOS sample.

---

2. Entry / Exit

Core engines (A and B)

• Momentum signal: RSI(14) on adaptive cumulative return over 63/126/252 days, blended by VIX percentile.
• Slow filter: price above 252-day SMA.
• Fast filter: EMA 8/21/50 + MACD 12/26/9 aligned bullish.
• Selection: every Friday, pick the top-N assets by composite score.
• Risk overlay: if VIX is above its 70th percentile (last 252 days), cut exposure by half.

Portfolio rebalancing

• Base frequency: annual (first 5 trading days of January).
• Drift threshold: if any sleeve deviates more than ±10% from its target weight, rebalance.
• Execution: sells first, then buys, to free buying power.

Costs

• 10 bps one-way in backtest (adjust with observed real slippage in paper trading).
• Alpaca ETFs are commission-free, but SH/PSQ/VIXY may have wider spreads.

---

3. Sizing

Sleeve level

A       = 20%
B       = 20%
rates   = 20%
bear    = 20%
cta     = 20%

Ticker level

• A and B: equal-weight within their selected top-N.
• Rates: 100% in the best of TLT / IEF / BIL based on SMA200.
• Bear: 100% in SH if SPY < SMA200, otherwise BIL.
• CTA proxy: equal-weight among PDBC / DBMF / KMLM in uptrend, otherwise BIL.

Account scaling

• The runner reads Alpaca account equity and allocates dollars proportionally.
• Fractional shares are supported by default; if unavailable, it rounds to whole lots.

---

4. Ensemble

The strategy is a fixed-weight ensemble at the sleeve level, not dynamic. Research showed that A/B/C switching rules added overfit and failed to beat A or B alone OOS.

• A and B are two implementations of the same dual-momentum engine with different scoring and concentration.
• Each contributes 20% of NAV.
• The diversifier sleeves contribute 60% of NAV.

---

5. File structure

Live/
├── README.md
├── pyproject.toml             # Project metadata + deps + pytest config
├── requirements.txt            # `pip install -r requirements.txt`
├── .env.example                # Template; copy to .env (gitignored)
├── live/                       # Strategy package
│   ├── __init__.py
│   ├── core_signals.py         # Point-in-time engines A and B
│   ├── data_feed.py            # Alpaca + yfinance + cache
│   ├── alpaca_executor.py      # Order placement
│   ├── portfolio.py            # Sleeve construction and targets
│   ├── risk.py                 # Safety guardrails
│   ├── state.py                # Peak equity / weights persistence
│   └── monitor.py              # Live vs backtest monitoring
├── scripts/
│   └── rebalance.py            # Daily entry point (was live_runner.py)
├── notebooks/
│   └── A_B_Diversifier_Analytics.ipynb
├── tests/
│   ├── __init__.py
│   └── test_live_pipeline.py   # Smoke tests
└── logs/                       # Runtime artifacts (gitignored): state.json, orders_*.csv, target_weights.jsonl
    └── .gitkeep

---

6. Setup

Installation

cd Live
pip install -r requirements.txt

Credentials

Create a .env file inside Live/ with:

ALPACA_API_KEY=PK...
ALPACA_API_SECRET=...
ALPACA_LIVE=false

Never commit .env to Git. It is already in .gitignore.

---

7. Usage

Dry-run (does not touch Alpaca)

cd Live
python scripts/rebalance.py --date 2025-06-27 --dry-run --prefer-yfinance

This uses yfinance, simulates a $100,000 account, and prints the orders it would send.

Paper trading

1. Make sure ALPACA_LIVE=false.
2. Remove --dry-run:

cd Live
python scripts/rebalance.py --prefer-yfinance

Real live trading

1. Validate several months in paper.
2. Set ALPACA_LIVE=true.
3. Run without --dry-run.

---

8. Scheduler

Run once per day after market close. Example cron on Linux/macOS:

35 16 * * 1-5 cd /path/to/Live && python scripts/rebalance.py

On Windows use Task Scheduler or Git Bash cron.

---

9. Risk guardrails

The runner automatically blocks execution if:

• Price data is more than 2 days stale.
• Any target ticker is missing from the price panel.
• A single position exceeds 50% of NAV.
• A+B combined exceeds 70% of NAV.
• Live drawdown exceeds -10% from peak equity.
• It is a weekend or fixed US holiday (1/1, 7/4, 12/25).

Also, the code only trades in paper unless ALPACA_LIVE=true is explicitly set.

---

10. Monitoring

cd Live
python -m live.monitor

Shows:

• Latest run date.
• Account equity.
• Number of orders and turnover.
• Sleeve-level weights.

---

11. Tests

cd Live
python -m pytest tests/ -q

Smoke tests do not call Alpaca; they use cached yfinance data.

---

12. Expected metrics (backtest)

Period	CAGR	Vol	Sharpe	Max DD	Calmar
2015-2025	16.5%	7.5%	2.20	-7.18%	2.30
OOS 2020-2025	20.6%	-	2.36	-7.18%	2.87

Historical results do not guarantee future performance.