HAR-RV Model of Realized Volatility
Also known as: HAR-RV, heterogeneous autoregressive realized volatility, Corsi HAR model, HAR-RV Modeli (Heterogeneous Autoregressive Realized Volatility)
The HAR-RV model, introduced by Fulvio Corsi in 2009, forecasts realized volatility by decomposing it into daily, weekly, and monthly components. It is a simple linear regression that mirrors how market participants with different investment horizons react to volatility, and it naturally captures the long-memory behaviour of volatility.
Key highlights
- Simpler and easier to estimate than GARCH while naturally capturing the long-memory persistence of volatility.
- Estimated by plain ordinary least squares, so coefficients and forecasts are fast and transparent.
- Decomposition into daily, weekly, and monthly horizons has a clear economic interpretation tied to heterogeneous market participants.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use HAR-RV when you have a long high-frequency financial series (at least about 250 daily observations) and want to forecast or predict realized volatility. It fits when realized variance can be computed from intraday returns and when you prefer a transparent, easy-to-estimate alternative to GARCH that still captures volatility's long memory. It is less appropriate when only low-frequency (daily close) data are available or when intraday microstructure noise is left uncorrected.
Strengths & limitations
- Simpler and easier to estimate than GARCH while naturally capturing the long-memory persistence of volatility.
- Estimated by plain ordinary least squares, so coefficients and forecasts are fast and transparent.
- Decomposition into daily, weekly, and monthly horizons has a clear economic interpretation tied to heterogeneous market participants.
- Requires high-frequency intraday data to compute realized variance; daily close prices alone are not enough.
- Intraday microstructure noise can bias realized variance and may need a kernel or pre-averaging correction.
- Needs a long sample (about 250 or more observations) for stable estimates.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is HAR-RV different from GARCH?
GARCH models the conditional variance of returns with a nonlinear recursion, while HAR-RV is a simple linear regression on realized volatility computed from intraday data. HAR-RV is easier to estimate and captures long memory through its daily, weekly, and monthly components without an explicit long-memory parameter.
What exactly is realized volatility here?
Realized volatility for a day is built from the realized variance, the sum of squared intraday returns over that day. HAR-RV then uses this daily series and its 5-day (weekly) and 22-day (monthly) averages as predictors.
Why daily, weekly, and monthly components?
They represent market participants acting on different horizons. Combining short-, medium-, and long-horizon averages reproduces the slowly decaying memory of volatility in a parsimonious way.
Do I need to correct for microstructure noise?
Often yes. At very high frequencies, bid-ask bounce and other microstructure effects bias realized variance, so kernel-based or pre-averaging corrections are commonly applied before fitting the model.
Sources
- 1.Corsi, F. (2009). A Simple Approximate Long-Memory Model of Realized Volatility. Journal of Financial Econometrics, 7(2), 174–196.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). HAR-RV Model. ScholarGate. https://scholargate.app/finance/har-rv-model