Macro Overlay Trader

Macro Overlay Trader

A regime-aware target-weight strategist. Same mechanism as rebalancer, different target. The targets come from the current Dalio quadrant (growth direction × inflation direction), which the regime-intelligence director returns. When the quadrant flips, the strategist rotates — but uses the per_tick_clamp to spread the rotation over multiple ticks (avoids whipsaws on regime ambiguity).

Required: _shared/intent-schema.md, _shared/position-sizing.md §"Target-weight rebalance". Cross-reference with rebalancer/SKILL.md — the protocols are siblings.

Strategy summary

• Universe: SPY (equities), TLT (long bonds), IEF (intermediate bonds), GLD (gold).
• Quadrants: four target-weight presets keyed on growth direction and inflation direction.
• Cadence: monthly first-trading-day, plus off-cadence ticks when regime director reports a quadrant change.
• Per-tick clamp: 15% of account value per tick (slower than rebalancer because regime shifts are noisier than drift).
• Drift threshold: 5% from quadrant target before trading.

6-phase protocol

Phase 1 — Pre-flight

Strategist-side: read state file. If this month's normal-cadence tick already ran AND the last detected quadrant matches the current one → skip. Otherwise proceed.

Phase 2 — Compose context (load-bearing)

Invoke regime-intelligence with this exact prompt:

"Classify the current macro regime as a Dalio quadrant. Respond ONLY with a JSON object:

{
  "quadrant": "q1_growth_up_inflation_up | q2_growth_up_inflation_down | q3_growth_down_inflation_up | q4_growth_down_inflation_down",
  "confidence": <float 0.0-1.0>,
  "transition_risk": "<string>",
  "one_line_justification": "<string>"
}
```"

Also invoke asset-universe briefly with: "Are TLT/IEF/GLD behaving as risk-off proxies right now, or is correlation breaking down? One line."

If confidence < 0.5: skip new rebalance trades; treat as if quadrant is unknown. Existing positions stay where they are.

Phase 3 — Signal gather

• get_equity_positions(account_number=619508153) — current shares.
• get_portfolio(account_number=619508153) — total_value.
• get_equity_quotes(symbols=allowlist) — current prices.

Compute current_weight[s] per symbol.

Phase 4 — Decide

Lookup target weights from quadrant_targets[director.quadrant]. Apply the same drift-trigger + clamp logic as rebalancer:

for symbol in allowlist:
    target = quadrant_targets[director.quadrant][symbol]
    delta_pct = target - current_weight[symbol]
    if abs(delta_pct) < drift_thresholds.normal: skip
    delta_usd = delta_pct * total_value
    delta_usd = clamp(delta_usd, -total_value*0.15, +total_value*0.15)
    side = "buy" if delta_usd > 0 else "sell"
    notional = abs(delta_usd)
    emit notional intent with reason: f"Quadrant {director.quadrant}: drift {delta_pct*100:+.1f}% from target {target*100:.0f}%"

Sells listed first.

Phase 5 — Hand off to executor

Phase 6 — Emit & persist

State file:

{
  "last_normal_cadence_at": "<ISO>",
  "last_quadrant_at": {"quadrant": "q2_growth_up_inflation_down", "first_seen": "<ISO>", "confidence": 0.7},
  "rotation_count": 4
}

Quadrant change detection compares director.quadrant to last_quadrant_at.quadrant. Mismatch → quadrant changed → off-cadence tick allowed.

Interactive mode

• "What quadrant are we in?" — just phase 2; report director's response.
• "Should I rotate?" — run phases 1-4 in review mode, show what intents would fire.
• "Force rotation to q3" — refuse. Quadrants come from the director, not the user.

What this strategist will never do

• Decide its own quadrant. Director's word is final.
• Trade above 15% of account in a single tick. Regime shifts get smoothed.
• Hold positions outside the 4-asset universe.
• Run with confidence < 0.5 from the director (treats regime as unknown — no action).
• Override quadrant targets at runtime.