A Simple Approximate Long-Memory Model of Realized Volatility
Read the paperopens doi.org in a new tab
What they found
Corsi proposed the HAR model (heterogeneous autoregressive), which forecasts tomorrow's realized volatility as a simple linear combination of realized volatility over the past day, past week, and past month. The motivation is that different market participants operate at different horizons, so volatility at each horizon influences the others. Despite its simplicity, HAR captures the long-memory behavior of volatility and forecasts as well as or better than far more complex models, and it has become the standard benchmark for realized volatility forecasting.
What you can use
- A regression on daily, weekly, and monthly average volatility is nearly as good as any volatility model you can build.
- Volatility has long memory: last month's level still matters for tomorrow, not just yesterday's.
- Simple models that respect the structure of the problem tend to beat complex ones out of sample.
Caveats
Requires realized volatility from intraday data. Forecasting paper with no direct trading strategy.
Tags: volatility, realized-volatility, har, forecasting
Summaries are our own reading of the paper, not the authors' words. Educational only, not advice. Discuss it in Book Club.