Skip to content

Rolling Vintage Rebalancing

Rolling rebalancing is a time-diversified trading strategy that spreads position entries across multiple days. Instead of rebalancing the entire portfolio at once, it creates daily "vintages" - each containing a fraction of the target allocation.

Why Use Rolling Mode?

Time Diversification

In full mode, you rebalance everything on a single day. If that day happens to be a local peak (for longs) or trough (for shorts), your entire portfolio suffers. Rolling mode spreads entries across N days, reducing timing risk.

Matches Prediction Horizon

If your predictions have a 30-day forward horizon (like pred_30d), each prediction captures expected alpha over 30 days. After 30 days, that prediction has "expired" - the alpha window has closed. Rolling mode naturally aligns with this by closing positions after rolling_days.

Reduced Turnover

When the same asset appears in consecutive vintages, the system automatically nets the trades. Instead of closing and reopening, it only trades the difference.

How It Works

The Vintage Model

Day 1:  Create Vintage_1 with 1/30 of target allocation
Day 2:  Create Vintage_2 with 1/30 of target allocation
...
Day 30: Create Vintage_30 with 1/30 of target allocation
        → Portfolio now at full allocation

Day 31: Vintage_1 expires (positions close)
        Create Vintage_31 (new positions open)
        → Steady state: 1/30 expires, 1/30 opens daily

Target-Sum Calculation

The portfolio's target position is the sum of all active vintage positions:

Vintage BTC ETH SOL
Day 1 0.01 -0.5 —
Day 2 0.02 — 10.0
Day 3 — -0.3 5.0
Total 0.03 -0.8 15.0

The system then calculates: Trade = Target - Current Wallet for each asset.

Unit-Based Tracking

Vintages store asset quantities (units), not USD values. This prevents drift issues:

  • You buy 0.1 BTC at \(50k (\)5,000)
  • BTC rises to $100k (now worth $10,000)
  • When vintage expires, you sell 0.1 BTC (the original units)
  • No leftover position from price appreciation

Configuration

portfolio:
  num_long: 60
  num_short: 60
  target_leverage: 3.0
  rebalancing:
    mode: rolling           # Enable rolling mode
    rolling_days: 30        # Match to prediction horizon
    seed_full: true         # Full deployment from day 1
    at_time: "18:15"        # Daily rebalance time (UTC)

Key Parameters

Parameter Description
mode "full" (default) or "rolling"
rolling_days Vintage lifespan. Match to prediction horizon.
seed_full If true, seeds all vintages on first run using historical predictions. If false, gradually ramps up.
every_n_days Ignored in rolling mode (always daily)

Cold Start Options

Gradual Ramp-Up (seed_full: false)

  • Day 1: 1/30 of target leverage
  • Day 15: 50% of target leverage
  • Day 30: Full target leverage
  • Day 31+: Steady state

Pros: Conservative, reduces timing risk on deployment Cons: Underinvested for first N days

Immediate Full Deployment (seed_full: true)

  • Day 1: System loads last 30 days of predictions
  • Creates 30 vintages with staggered "birth dates"
  • Immediately at full target leverage
  • Day 2: Oldest vintage expires, new one opens

Pros: Fully invested from day 1, time-diversified from start Cons: Requires historical prediction data

CLI Usage

Rebalance with Rolling Mode

# Use config settings
cc-liquid rebalance

# Override mode from CLI
cc-liquid rebalance --set portfolio.rebalancing.mode=rolling

View Active Vintages

cc-liquid vintages

Output: 

┌──────────────────────────────────────────────────────────┐
│               Active Vintages (30-day rolling)           │
├────────────┬─────┬────────────┬───────────┬─────────────┤
│ Birth Date │ Age │ Expires In │ Positions │ Value       │
├────────────┼─────┼────────────┼───────────┼─────────────┤
│ 2024-11-01 │ 29d │ 1d         │ 4         │ $1,234.56   │
│ 2024-11-02 │ 28d │ 2d         │ 4         │ $1,345.67   │
│ ...        │ ... │ ...        │ ...       │ ...         │
│ 2024-11-30 │ 0d  │ 30d        │ 4         │ $1,456.78   │
└────────────┴─────┴────────────┴───────────┴─────────────┘

Total vintage value: $42,000.00
Active vintages: 30 of 30

Backtest Rolling Mode

# Use config mode
cc-liquid analyze

# Override from CLI using --set
cc-liquid analyze --set portfolio.rebalancing.mode=rolling --set portfolio.rebalancing.rolling_days=30

# Compare strategies
cc-liquid analyze --set portfolio.rebalancing.mode=full    # Then compare to:
cc-liquid analyze --set portfolio.rebalancing.mode=rolling

Position Sizing

With rolling mode, each vintage contains 1/rolling_days of the target allocation:

  • Total positions: num_long + num_short = 120
  • Rolling days: 30
  • Positions per vintage: 120 / 30 = 4 (2 long, 2 short)
  • Leverage per vintage: target_leverage / 30

Minimum Account Size

Ensure your account can support the vintage granularity:

Daily allocation = Account Value / rolling_days
Per-position size = Daily allocation / (num_positions_per_vintage)

If per-position size falls below min_trade_value ($10), trades will be skipped.

Example: With $10,000 account, 30-day rolling, 4 positions per vintage: - Daily allocation: $333 - Per position: $83 ✓ (above minimum)

With $1,000 account: - Daily allocation: $33 - Per position: $8.25 ✗ (below minimum, trades may fail)

Edge Cases

Scenario Handling
Missed rebalance day Target-sum self-corrects on next run
Stale predictions Warning if latest prediction > 1 day old
Asset delisted Skipped in target sum, position closes naturally
Manual intervention Target-diff approach handles any wallet state

Comparing Full vs Rolling

Aspect Full Mode Rolling Mode
Timing Risk Higher (single entry point) Lower (spread across N days)
Turnover Higher (full rebalance) Lower (only 1/N changes daily)
Complexity Simple Requires state tracking
Cold Start Immediate full deployment Gradual or seeded
Best For Infrequent rebalancing Daily trading with horizon-matched predictions