Answering the Skeptics: Yes, Standard Volatility Models Do Provide Accurate Forecasts
Read the paperopens doi.org in a new tab
What they found
Earlier studies had concluded that GARCH models forecast volatility poorly because their forecasts explained only a few percent of the variation in squared daily returns. The authors showed the problem was the yardstick: a single day's squared return is an extremely noisy measure of that day's true volatility. Using five-minute returns on the Deutsche mark and yen to build 'realized volatility', a far more precise measure, they showed that standard GARCH forecasts explain about half the variation in true daily volatility, which is about as good as theory says is possible.
What you can use
- Realized volatility from intraday data is a much better measure of what actually happened than the daily range or close-to-close return.
- Volatility forecasts are more accurate than they look if you judge them against a clean measure.
- Judge any volatility signal by a noisy proxy and you will conclude it is useless; that conclusion is often wrong.
Caveats
FX data from the early 1990s. Realized volatility requires high-frequency data and has its own microstructure noise issues at very fine sampling.
Tags: volatility, realized-volatility, forecasting, high-frequency
Summaries are our own reading of the paper, not the authors' words. Educational only, not advice. Discuss it in Book Club.