Har rv volatility forecasting
The Heterogeneous Autoregressive model of Realized Volatility (HAR-RV) is a common benchmark for forecasting realized volatility (RV) Verified Answer #1Verified Answer #2. It is specifically designed to predict realized measures rather than latent conditional variance or close-to-close return variance Verified Answer #1.
Model Structure and Persistence
The HAR-RV model uses an additive structure to capture the persistence of volatility across different time horizons Verified Answer #1Verified Answer #2. It approximates long-memory behavior by incorporating three primary components: daily, weekly, and monthly realized volatility Verified Answer #2. The model typically forecasts the next period's volatility using the most recent daily RV, a recent weekly average, and a recent monthly average Verified Answer #2. This simple structure allows the model to remain robust while capturing short-, medium-, and long-term persistence without the complexity of full long-memory models Verified Answer #2.
Advantages Over Return-Based Models
Realized-measure models like HAR-RV generally outperform standard GARCH models when the target is realized volatility Verified Answer #1Verified Answer #2. While standard GARCH models rely on daily close-to-close returns, HAR-RV utilizes intraday information, which is more informative about ex post volatility Verified Answer #1Verified Answer #2. Realized volatility is calculated by summing squared high-frequency returns within a specific day Verified Answer #2. Because HAR-RV uses these lagged realized measures directly, it is structurally better aligned with the forecasting target than models inferred from daily returns Verified Answer #2.
Applications in Cryptocurrency
For cryptocurrency markets, HAR-RV is considered a best practical baseline for forecasting realized volatility Verified Answer #1Verified Answer #2. Enhanced versions of the model are often recommended to account for market-specific features like jumps and asymmetry Verified Answer #1Verified Answer #2.
- Recommended Upgrades: Practical upgrades for crypto include log-HAR-RV-J, HAR-CJ, or HAR-RS, which incorporate jump and semivariance terms Verified Answer #2.
- Alternative Models: Realized GARCH serves as a strong alternative when a full return-volatility system is required Verified Answer #1.
- Fallback Options: If intraday data is unavailable and only daily returns are accessible, GJR-GARCH or EGARCH are the preferred fallback models Verified Answer #1Verified Answer #2.
- Machine Learning: Machine learning or ensemble models are considered "challengers" and should only be used if they demonstrate superior out-of-sample performance compared to HAR Verified Answer #2.