har-rv-volatility-forecasting

The Heterogeneous Autoregressive model of Realized Volatility (HAR-RV) is a benchmark framework used for predicting realized volatility Verified Answer #1. Realized volatility is calculated from high-frequency intraday returns rather than daily returns Verified Answer #1. Because HAR-RV models utilize lagged realized-volatility measures directly, they are considered better matched to the forecasting target than classic GARCH models Verified Answer #1. While GARCH models are designed for latent conditional variance inferred from daily returns, HAR-RV captures the persistence of volatility across different time horizons Verified Answer #1.

Model Structure and Components

The HAR-RV model approximates the long-memory behavior of volatility using a simple structure of three components Verified Answer #1:

This configuration allows the model to capture short-, medium-, and longer-horizon persistence without the complexity required by full long-memory models Verified Answer #1.

Applications and Variations

For cryptocurrency markets, jump-robust and asymmetry-enhanced versions of the model are preferred over the standard version Verified Answer #1. Recommended variations for these assets include log-HAR-RV-J, HAR-CJ, and HAR-RS Verified Answer #1.

In scenarios where only daily returns are available instead of intraday data, GJR-GARCH or EGARCH models serve as the best fallback options Verified Answer #1. Machine learning models or ensembles built on HAR features are considered "challenger" models and are typically only utilized if they demonstrate superior out-of-sample performance compared to the standard HAR model Verified Answer #1.