Work detail

Range-based volatility estimation and forecasting

Author: Mgr. Daniel Benčík
Year: 2012 - summer
Leaders: doc. PhDr. Jozef Baruník Ph.D.
Work type: Finance, Financial Markets and Banking
Language: English
Pages: 108
Awards and prizes: M.A. with distinction from the Dean of the Faculty of Social Sciences for an excellent state-final examination performance.
Abstract: In this thesis, we analyze new possibilities in predicting daily ranges, i.e. the differences
between daily high and low prices. The main focus of our work lies in investigating how
models commonly used for daily ranges modeling can be enhanced to provide better forecasts.
In this respect, we explore the added benefit of using more efficient volatility measures as
predictors of daily ranges. Volatility measures considered in this work include realized
measures of variance (realized range, realized variance) and range-based volatility measures
(Parkinson, Garman & Klass, Rogers & Satchell, etc). As a subtask, we empirically assess
efficiency gains in volatility estimation when using range-based estimators as opposed to
simple daily ranges. As another venue of research in this work, we analyze the added benefit
of slicing the trading day into different sessions based on trading activity (e.g. Asian, European
and American session). In this setting we analyze whether whole-day volatility measures
reliably aggregate information coming from all trading sessions. We are led by intuition that
different sessions exhibit significantly different characteristics due to different order book
thicknesses and trading activity in general. Thus these sessions are expected to provide
valuable information concealed in the aggregate volatility measure.
Next, we investigate the possibility to gradually update daily volatility forecasts by
incorporating all up-to-date information. That means once a trading sessions ends its volatility
and trading activity measures are used for updating the current day's volatility forecast. These
updated forecasts exhibit very strong gains in terms of goodness-of-fit and thus short-term
traders active in later sessions of the day can gain a significant advantage over traders active
early in the day.
The array of models within which we investigate the aforementioned effects include the
heterogeneous autoregressive model, conditional autoregressive ranges model and a vector
error-correction model of daily highs and lows. Models performing well in terms of in-sample
fit are challenged on out-of-sample, one-day-ahead forecasting. Contrary to results presented in
literature, models based on co-integration of daily highs and lows fail to produce good quality
forecasts. When one strives for the best one-day-ahead daily ranges forecasts a HAR model
using realized ranges as predictors with a GARCH volatility-of-volatility component is the
preferred option.
Downloadable: Diploma Theses of Benčík


McKinsey & Company