EWMA-based volatility forecasting with prediction intervals • Hyperliquid Perps
EWMA is a volatility forecasting method that gives more weight to recent observations. The RiskMetrics approach uses λ = 0.94, meaning today's volatility estimate is 94% of yesterday's estimate plus 6% of today's squared return.
σ²ₜ = λ × σ²ₜ₋₁ + (1 - λ) × r²ₜ₋₁ Where: σ²ₜ = Variance forecast for time t λ = Decay factor (0.94 for daily, adjusted for intraday) r²ₜ₋₁ = Previous period's squared return
For forecasts beyond 1 period, we use the property that EWMA variance forecasts converge to unconditional variance over time. Short-term forecasts rely more on recent EWMA; longer-term forecasts blend toward historical average.
95% confidence bands are calculated assuming log-normal distribution of volatility. The width of bands increases with forecast horizon, reflecting increasing uncertainty.
When volatility is forecasted to increase, position sizes should decrease proportionally to maintain constant dollar risk. The dashboard provides adjusted position sizes based on forecast volatility vs historical average.
EWMA assumes volatility clustering but doesn't capture leverage effects (asymmetric response to up/down moves) or regime changes. For more sophisticated forecasting, consider GARCH models. Always use forecasts as one input among many, not as sole decision criteria.