Estimating Interval Realized Volatility from High-Frequency Returns

Summary

The document asks how to estimate realized volatility for individual time intervals rather than for an entire day. It describes realized variance at a chosen sampling frequency as the sum of squared log returns over that span. The example considers a daily estimate based on 30-minute returns and asks whether a 30-minute estimate can instead be built from the squared one-minute returns within the interval.

This frames a useful distinction between aggregating returns over a daily window and measuring variation within shorter windows. However, the text contains only the question, not an answer or empirical comparison, so it does not confirm the proposed procedure or address choices such as sampling frequency, microstructure noise, or whether to report realized variance or its square root. It is a starting point for understanding interval-based volatility measurement, not a complete estimator specification.

Key ideas

• Realized variance over a span is described as the sum of squared log returns sampled within that span.
• A daily estimate can aggregate returns observed at a chosen intraday frequency.
• The document asks whether shorter-period realized variance can use higher-frequency returns within each interval.
• It gives no answer about sampling choices, market microstructure noise, or converting variance to volatility.

Tags

# Calculating realized volatility of high-frequency data


# Calculating realized volatility of high-frequency data












I am wondering how to calculate the realized volatility. Sources such as say that the realized volatility is the sum of squared log returns sampled at a given frequency. So using 30 minute frequency the realized volatility per day would be the sum of 16 log returns observed during the day. However, I am interested in having a realized volatility estimate for each of the periods, in this case for each of the 30 minutes intervals. I do not intend to do forecasting, I am simply interested in knowing how volatile the asset is in a given period of time.

After reading this am I right to assume that, for example, a 30 minute RV would be estimated by summing up volatilities from higher frequency data of the preceeding period, so let's say the sum of 30 1-minute squared log returns?

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.