Realized Variance Versus Standard Deviation for Volatility

Summary

The document compares annualized standard deviation of returns with realized volatility computed as the square root of summed squared high-frequency returns. It says the realized variance approach can make better use of intraday observations than a standard deviation based on a coarser sample, and may perform better for measuring volatility when high-frequency data are available.

The answer also notes limits: market microstructure noise can bias realized variance, so noise-robust estimators and methods that account for jumps may be needed. It mentions heterogeneous autoregressive models as one way to forecast volatility from realized variance. The discussion is brief and does not specify sampling frequency, estimator choices, data or a direct empirical comparison, so it does not establish that realized variance is always more accurate or best for every use case.

Key ideas

• Realized variance sums squared high-frequency returns and its square root gives realized volatility.
• High-frequency observations can provide more information than a plain standard deviation calculated from less frequent returns.
• Market microstructure noise and price jumps can affect realized variance estimates.
• HAR models can use realized variance as an input to volatility forecasts.

Tags

# Realized Volatility Methods


# Realized Volatility Methods












Can someone explain to me which of these two methods is more accurate or commonly used to calculate Realized Volatility? I'm seeing both used, but I get very different results from them.

1) Standard deviation of log returns x Sqrt of 252.

2) Sqrt of the Realized Variance based on the summed squares.

Thank you

## Answer by Kevin (score 1)

https://quant.stackexchange.com/a/48672

In econometrics, if you have access to high-frequency (HF) data, then the realised variance approach works better than simply computing the standard deviation. The reason is that you use much more data and thus can utilise the additional information HF data carries, thus RV typically performs better than, say, GARCH models and a plain standard deviation. There are of course many extensions of realised variance which address biases (noise robust estimators) arising from market microstructure and take jumps into account. Furthermore, using heterogeneous autoregressive (HAR) models, you can use RV to forecast volatility.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.