Volatility Blend Weights

Standard Blend

σ_blend = 0.7 × σ_short + 0.3 × σ_long

Dove: - σ_short: EWMA con λ = 0.06061 (span 32) - σ_long: EWMA con λ molto piccolo (span >>100, or historical average)

Perche Blendare?

Short-term only (100% weight): - Cattura volatility clustering ✓ - Ma: Massime posizioni pre-crisis ✗ - Ma: Ignore mean reversion ✗

Long-term only (100% weight): - Cattura mean reversion ✓ - Ma: Miss regime changes ✗ - Ma: Poor short-term forecast ✗

Blend (70/30): - Best of both worlds ✓ - Smoother (lower costs) ✓ - Safer (no max position pre-crash) ✓

Derivazione Pesi

Weights da regression analysis:

Future σ regressed on: - Current short-term σ - Historical long-term σ

Optimal weights ≈ 0.7 / 0.3 (robusto across instruments).

Esempio Pratico

Pre-crisis scenario:

Short-term σ = 10% (currently calm) Long-term σ = 18% (historical average)

100% short-term:

σ = 10% → Position massima (danger!)

Blend:

σ = 0.7×10% + 0.3×18% = 7% + 5.4% = 12.4%

→ Position più conservativa (safer!)

Alternative Weights Tested

Short Weight Long Weight Forecast Power Safety
1.0 0.0 Best Worst
0.8 0.2 Good OK
0.7 0.3 Good Good
0.5 0.5 OK Better
0.0 1.0 Worst Best

70/30: Optimal trade-off.

Cost Reduction

Blend riduce trading costs:

100% short-term: - Noisy estimate - Frequent position changes - High turnover

70/30 blend: - Smoother estimate - ~20-30% less trading - Lower costs

Implementation

# Daily update
short_term_vol = ewma(returns, lambda=0.06061)
long_term_vol = historical_mean_vol  # or very slow EWMA

blended_vol = 0.7 * short_term_vol + 0.3 * long_term_vol

position = calculate_position(capital, target, price, blended_vol)

Long-Term Estimate Options

Option A: Historical mean

σ_long = Mean(σ over all history)

Option B: Very slow EWMA

σ_long = EWMA(λ = 0.001)  # span ~2000 giorni

Option C: Expanding window

σ_long = Std(all returns to date)

Tutti funzionano reasonably well.

Concetti Correlati

  • [[EWMA]] - metodo per short-term estimate
  • [[Volatility Clustering]] - short-term phenomenon
  • [[Mean Reversion]] - long-term phenomenon
  • [[Position Sizing]] - usa σ_blend