Comparing Five Historical Volatility Estimators as Persistence Forecasts
Summary
The study compares Close-to-Close, Parkinson, Garman-Klass, Rogers-Satchell, and Yang-Zhang variance estimates as one-session-ahead persistence forecasts. It evaluates those forecasts against a next-session EURUSD realized-variance proxy built from the close-to-open jump and returns reconstructed from minute closes. The main comparison uses a 20-session estimation window; 10- and 40-session windows are sensitivity checks. The article describes the formulas, chronological data handling, shared target mask, scoring, and supplied MQL5 tools for reproducing the analysis.
The evidence is historical broker output for 2020–2025: 1,560 candidate sessions were reduced to 1,472 after common minute-data coverage checks. Mean QLIKE is the primary criterion, and a paired circular moving-block bootstrap assesses the Parkinson–Yang-Zhang difference. A separate analysis applies past-only rolling scale calibration. The findings are specific to one instrument, broker history, interval, and target proxy; they do not establish a universally best estimator or a trading strategy. The target is an imperfect volatility proxy, and the sample selection and persistence forecast rule also limit interpretation.
Key ideas
- Historical volatility estimators differ in whether they use closing returns, intraday ranges, or the full OHLC record.
- The experiment uses each rolling variance estimate directly as a one-session-ahead persistence forecast.
- All estimators are scored on a shared set of EURUSD sessions after chronological alignment and minute-data coverage checks.
- Mean QLIKE is the primary comparison, with a block bootstrap used to assess the main pairwise difference.
- The results describe one broker, instrument, period, and realized-variance proxy rather than a universal ranking.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.