risk-governor

name	risk-governor
description	Compute volatility-targeted position sizing using GARCH conditional volatility. Applies continuous inverse-vol scaling (w = vol_target/σ_t) plus regime and quant score overrides. Use when user asks about position size, how much to hold, risk management, or when GARCH vol is elevated.

Skill: Risk Governor

Computes recommended position weight using GARCH annualised vol + anomaly regime + composite quant score.

Trigger

Use this skill when the user asks:

• "How much GOLDBEES should I hold?"
• "What position size given current vol?"
• "Is it safe to add at this volatility?"
• "Run the risk governor"
• "What does the risk model say?"

Recommended Command (use the MCP pipeline tool)

run_pipeline        ← calls Risk Governor + Kelly blend in one step
get_latest_signal   ← reads last stored RG weight from DB

Manual Command (standalone, no Kelly blend)

python -c "
import sys; sys.path.insert(0,'.')
from src.tools.risk_governor import compute_position_weight, explain_decision, vol_target_for
from src.db.pool import get_pool
import pandas as pd, warnings
warnings.filterwarnings('ignore')

try:
    # Use GOLDBEES ETF prices — NOT COMEX GOLD (different asset class)
    price_df = get_pool().query_df('''
        SELECT trade_date,
               toFloat64(argMax(open,   imported_at)) AS open,
               toFloat64(argMax(high,   imported_at)) AS high,
               toFloat64(argMax(low,    imported_at)) AS low,
               toFloat64(argMax(close,  imported_at)) AS close,
               toFloat64(argMax(volume, imported_at)) AS volume
        FROM market_data.daily_prices
        WHERE symbol='GOLDBEES' AND category='etfs'
        GROUP BY trade_date ORDER BY trade_date DESC LIMIT 300
    ''')
    price_df['trade_date'] = pd.to_datetime(price_df['trade_date'])
    price_df = price_df.sort_values('trade_date').reset_index(drop=True)
    from src.ml.anomaly import run_composite_anomaly
    df_r, _, _ = run_composite_anomaly(price_df)
    garch_vol = float(df_r['garch_vol'].dropna().iloc[-1])
    regime    = str(df_r['regime'].iloc[-1])
    latest    = float(df_r['close'].iloc[-1])
    ema50     = float(price_df['close'].ewm(span=50, adjust=False).mean().iloc[-1])
    below_ema = latest < ema50
except Exception as e:
    garch_vol = 16.5; regime = '✅ Normal'; below_ema = False
    print(f'Warning: using defaults ({e})')

vol_target = vol_target_for('GOLDBEES')
d = compute_position_weight(
    garch_annual_vol_pct=garch_vol,
    regime=regime,
    vol_target_pct=vol_target,
    price_below_ema50=below_ema,
)
print(explain_decision(d))
"

Formula

w(t) = min(1.0,  vol_target / σ_t)  ×  regime_mult  ×  trend_mult

Vol target per asset class: gold/silver = 15%, equity = 18%, debt = 5%.

Vol	Weight	Tier
≤15%	100%	FULL
22%	68%	REDUCED
34.5%	43%	HALF
38% + Flash Crash + bear score	10%	MINIMAL

Regime Multipliers

• ⚡ Flash Crash → 0.50×
• 🔥 Volatile Breakout → 0.75×
• ⚠️ Crowded Long → 0.80×
• 🧨 Blow-off Top → 0.70×
• ✅ Normal / 📈 Strong Trend → 1.00×