Rebalance Volume Optimisation
1. The Over-Clearing Problem
When Phase 2 fires, the default behaviour is to clear the entire Reserve Pool deviation back to the target balance. But clearing to exactly zero is often suboptimal:
If flow continues in the same direction after Phase 2 executes, the Reserve Pool immediately starts accumulating again, potentially triggering another Phase 2 within hours. Each Phase 2 execution carries external cost. Clearing to exactly zero maximises the number of Phase 2 triggers over time.
2. Target Residual Band
Instead of clearing to zero, Phase 2 clears down to a Target Residual — a small buffer left intentionally in the Reserve Pool. This buffer absorbs the next wave of same-direction flow without immediately re-triggering Phase 2.
Rebalance_Amount = Current_Deviation − Target_Residual
Target_Residual = Soft_Threshold × Residual_Factor
Where Residual_Factor is a configurable parameter (Phase 1 default: 0.3). This means the system rebalances down to 30% of the Soft Threshold, not to zero.
3. Worked Example: Reserve USDT Pool, USD-IDR
Parameters: Soft Threshold = $50K, Residual Factor = 0.3, Current Deviation = $80K (cooldown expired).
Target_Residual = $50K × 0.3 = $15K
Rebalance_Amount = $80K − $15K = $65K
Phase 2 RFQ is submitted for $65K, not $80K. The remaining $15K stays in the Reserve Pool as a buffer.
If the next morning's remittance flow adds $35K, the Reserve Pool goes to $50K — right at the Soft Threshold, starting a new cooldown cycle. Without the buffer, the Reserve Pool would hit $35K from zero, potentially already triggering Phase 2 again.
4. Residual Factor Guidelines
| Residual Factor | Behaviour | When to Use |
|---|---|---|
| 0.0 | Clear to zero (current behaviour) | Use only for corridors with no bilateral flow pattern or during market stress |
| 0.2 – 0.3 | Leave small buffer (recommended) | Standard setting for bilateral flow corridors; balances re-trigger frequency vs risk |
| 0.4 – 0.5 | Leave larger buffer | Use for highly cyclical corridors where same-direction flow resumes predictably after each cycle |