Exploring the predictability of range‐based volatility estimators using recurrent neural networks

• Author: Gábor Petneházi,József Gáll
• DOI: http://doi.org/10.1002/isaf.1455
• Date: 01 July 2019
• Published date: 01 July 2019

RESEARCH ARTICLE

Exploring the predictability of range‐based volatility estimators

using recurrent neural networks

Gábor Petneházi

1

|József Gáll

2

1

Doctoral School of Mathematical and

Computational Sciences, University of

Debrecen, Debrecen, Hungary

2

Department of Applied Mathematics and

Probability Theory, University of Debrecen,

Debrecen, Hungary

Correspondence

Gábor Petneházi, Doctoral School of

Mathematical and Computational Sciences,

University of Debrecen. Debrecen, Hungary

Email: gabor.petnehazi@science.unideb.hu

Funding information

EFOP‐3.6.3‐VEKOP‐16‐2017‐00002; Pallas

Athéné Domus Scientiae Alapítvány

Summary

We investigate the predictability of several range‐based stock volatility estimates and

compare them with the standard close‐to‐close estimate, which is most commonly

acknowledged as the volatility. The patterns of volatility changes are analysed using

long short‐term memory recurrent neural networks, which are a state‐of‐the‐art

method of sequence learning. We implement the analysis on all current constituents

of the Dow Jones Industrial Average index and report averaged evaluation results.

We find that the direction of changes in the values of range‐based estimates are

more predictable than that of the estimate from daily closing values only.

KEYWORDS

long short‐term memory, range‐based volatility, recurrent neural network, volatilityforecasting

1|MOTIVATION

The volatility of assets has an important role in several areas of

finance. As a measure of riskiness, it is a key factor in, for example,

portfolio management and option pricing. A good understanding of

the nature and evolution of return volatilities is obviously valuable

for financial practitioners.

Volatility quantifies the dispersion of returns. Unfortunately, this

dispersion is not directly observable. Hence, we need to estimate it,

with not having a reliable benchmark.

Several studies have tried to explore and understand the nature of

this unknown volatility. One reasonable approach is sampling from the

price process frequently, so that we do not lose too much data. Ander-

sen, Bollerslev, Diebold, and Ebens, (2001) analysed the properties of

stock market volatility using 5 min returns. They reported that daily

variances significantly fluctuate through time, and their distributions

are extremely right‐skewed and leptokurtic, whereas logarithmic stan-

dard deviations approximate the normal distribution well. Engle and

Patton, (2007) outlined several stylized facts of volatility that have

emerged in previous studies:

•Persistence: large moves are usually followed by large moves, and

small moves are usually followed by small moves in the price

process.

•Mean reversion: usually, there is a normal level of volatility to

which it returns after uplifts and falls.

•Asymmetric impact of innovations: positive and negative shocks

have different impacts on volatility.

•Influence of exogenous variables: information outside the price

series (e.g. announcements) could have an impact on volatility.

These features suggest that, if we could measure volatility, it

should be somewhat forecastable. But we cannot measure it; the best

we can do is come up with reasonable proxies.

One such proxy is the standard deviation of returns—returns,

which are usually calculated from daily closing prices. It is obvious

that by sampling the asset's price more frequently we could make

better estimates of its unobservable true volatility. If, for example,

we measured the daily price ranges (i.e. daily high minus daily low),

we would already know a lot more about the unseen path of the

prices.

------------------------------------------------------- -- --- -- -- --- -- --- -- -- --- -- --- -- -- --- -- --- -- -- --- -- --- -- --- -- -- --- -- --- -- -

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided

the original work is properly cited.

© 2019 The Authors Intelligent Systems in Accounting, Finance and Management Published by John Wiley & Sons Ltd.

Received: 13 September 2018 Revised: 2 April 2019 Accepted: 13 June 2019

DOI: 10.1002/isaf.1455

Intell Sys Acc Fin Mgmt. 2019;26:109–116. wileyonlinelibrary.com/journal/isaf

109