Forecasting bitcoin volatility: Evidence from the options market

• DOI: http://doi.org/10.1002/fut.22144
• Date: 01 October 2020
• Published date: 01 October 2020
• Author: Dirk G. Baur,Lai T. Hoang

J Futures Markets. 2020;40:1584–1602.wileyonlinelibrary.com/journal/fut1584

|

© 2020 Wiley Periodicals LLC

Received: 16 March 2020

|

Accepted: 18 May 2020

DOI: 10.1002/fut.22144

RESEARCH ARTICLE

Forecasting bitcoin volatility: Evidence from the options

market

Lai T. Hoang

1,2

|Dirk G. Baur

1

1

UWA Business School, The University of

Western Australia, Crawley,

Western Australia, Australia

2

National Economics University, Hanoi,

Vietnam

Correspondence

Lai T. Hoang, UWA Business School, The

University of Western Australia, Crawley,

WA 6009, Australia.

Email: lai.hoang@uwa.edu.au

Abstract

This paper studies a large number of bitcoin (BTC) options traded on the

options exchange Deribit. We use the trades to calculate implied volatility (IV)

and analyze if volatility forecasts can be improved using such information. IV

is less accurate than AutoRegressive–Moving‐Average or Heterogeneous Auto‐

Regressive model forecasts in predicting short‐term BTC volatility (1 day

ahead), but superior in predicting long‐term volatility (7, 10, 15 days ahead).

Furthermore, a combination of IV and model‐based forecasts provides the

highest accuracy for all forecasting horizons revealing that the BTC options

market contains unique information.

KEYWORDS

bitcoin, bitcoin options market, forecasting, implied volatility, realized volatility

JEL CLASSIFICATION

G12; G13; G14

1|INTRODUCTION

Bitcoin (BTC) has grabbed the attention of academics since its inception in 2008 and particularly since its bubble‐like

price rise in 2017. This has led to a large number of papers examining various aspects of BTC, such as market efficiency

(e.g., Kristoufek, 2018; Tiwari, Jana, & Roubaud, 2018; Urquhart, 2016), its properties compared to other assets (e.g.,

Baur, Hong, & Lee, 2018; Corbet, Lucey, Urquhart, & Yarovaya, 2019; Dyhrberg, 2016; Yermack, 2013), and the linkages

with other cryptocurrencies (e.g., Baur & Hoang, 2020; Katsiampa, Corbet, & Lucey, 2019; Yi, Xu, & Wang, 2018).

Recent literature also studied BTC futures after the Chicago Board Options Exchange (CBOE) and the Chicago Mer-

cantile Exchange (CME) introduced futures contracts on BTC in December 2017.

1

Interestingly, there is no study on BTC options despite its informational role and its potential relevance to enhance

our understanding about BTC trading and risk preferences of BTC investors. Options have advantages over futures as

they have relatively short maturities and different strike prices (e.g., see Kelly, Pastor, & Veronesi, 2016). Another key

advantage of options over futures and spot markets is that options prices carry information of investors' expectations

about future volatility (i.e., implied volatility [IV]), which allows a direct examination on the relevance of such unique

information.

1

Baur and Dimpfl (2019) use intraday data from regulated futures exchanges and find that the BTC spot price leads the futures price, indicating that

the information content on those futures markets is negligible. In contrast, Alexander, Choi, Park, and Sohn (2020) show for the unregulated futures

exchange BitMEX that it plays a leading role in the price discovery process.

The lack of studies on BTC options and IV of BTC

2

is in stark contrast to the growing literature on the volatility of

BTC and volatility prediction (e.g., Katsiampa et al., 2019; Klein, Thu, & Walther, 2018; Shen, Urquhart, & Wang, 2019).

It may also be surprising given the role of options‐IV for forecasting volatility and for investors to anticipate risks and

implement appropriate hedging strategies (Frijns, Tallau, & Tourani‐Rad, 2010).

Hence, this paper contributes to the literature on BTC, options trading and options‐IV. We assess the accuracy of IV

derived from BTC options prices in forecasting future volatility and compare the relative forecasting power of IV with

that of forecasts generated by time‐series models (i.e., Generalized AutoRegressive Conditional Heteroskedasticity

(GARCH), AutoRegressive–Moving‐Average (ARMA) and Heterogeneous Auto‐Regressive (HAR)). We also study

whether IV contains any incremental information beyond model‐based forecasts.

Earlier studies that compare the forecasting power of option‐IV and GARCH‐type models generally conclude that

the former outperforms the latter (e.g., Day & Lewis, 1992 in stock markets and Jorion, 1995 in foreign exchange

markets). Instead of daily data used in GARCH‐type models, recent studies increasingly use models based on intraday

data and realized volatility (e.g., ARMA or ARFIMA) and find that such models generate comparable or even better

forecasts than IV (e.g., Kambouroudis, McMillan, & Tsakou, 2016; Martens & Zein, 2004; Pong, Shackleton, Taylor, &

Xu, 2004). However, IV still provides some incremental information and thus improves the forecasting performance

when it is combined with model‐based forecasts (Busch, Christensen, & Nielsen, 2011; Haugom, Langeland, Molnár, &

Westgaard, 2014). Since all prior studies focus on either stocks, foreign exchange or commodities markets, we con-

tribute to the literature with new evidence from the cryptocurrency market.

We use options data from Deribit—the largest albeit unregulated BTC options exchange as of February 2020.

During 2019, the daily trading volume of options contracts on Deribit regularly reached 5,000 BTC which was

equivalent to about 50 million USD. Although BTC options were also launched on the regulated exchange CME on

January 13, 2020, its trading volume is still much lower than that of Deribit,

3

suggesting that the CME options exchange

is in an early stage with very limited data available rendering an empirical analysis difficult. We also observe that

trading volume on Deribit dominates all other existing BTC options exchanges.

4

This leading role of Deribit justifies the

focus on this exchange for BTC options.

Our results indicate that option‐IV is generally less accurate than parametric time‐series models in predicting 1‐day

ahead volatility, but superior in long‐term prediction (7, 10, 15 days ahead). In addition, regardless of forecasting

horizons, IV consistently contains incremental information that is not available in model‐based forecasts. Thus, a

combination of IV and parametric volatility models is able to improve the forecasts. It is important to note though that

this finding is not necessarily a surprise as combinations of forecasts generally improve forecasts (e.g., see Bates &

Granger, 1969; Fameliti & Skintzi, 2020; Hendry & Clements, 2004; Kambouroudis et al., 2016; Liang, Wei, &

Zhang, 2020; Metaxoglou & Smith, 2017).

Despite the fact that the options exchange considered in this study (Deribit) is unregulated, BTC traders are

attracted by the investment opportunities provided by the exchange and appear to tolerate the lack of regulation. This

finding is consistent with Alexander et al. (2020) in the BTC futures market and also in line with the fact that BTC itself

is unregulated.

In addition, we contribute to another growing strand of the BTC literature, that is, volatility modeling and fore-

casting. Since BTC is extremely volatile, predicting its volatility is crucial for risk management. Although there is a large

body of literature assessing volatility forecasts in stocks, fiat currencies and commodities markets (e.g., Andersen,

Bollerslev, & Meddahi, 2005; Patton, 2011; Poon & Granger, 2003, among many others), research on BTC is relatively

scarce. Early studies attempt to model BTC volatility using several GARCH‐type models with daily returns and

compare the models' performance (e.g., Ardia, Bluteau, & Rüede, 2019; Katsiampa, 2017), but mainly focus on the in‐

sample goodness‐of‐fit of the models. Hence the conclusions on the “best”model do not necessary hold in out‐of‐

sample forecasts. Recently, exploiting the availability of intraday data, Shen et al. (2019) consider a comprehensive set

of 18 HAR‐type models and find that a HARQ‐F‐J model provides the best out‐of‐sample forecasts. We extend those

studies by examining various types of models, including GARCH, ARMA and HAR, and find that ARMA‐type models

generally provide better forecasts than GARCH‐type models for both 1‐day and 7‐day horizons. Interestingly, the

HARQ‐F‐J model, despite being documented in Shen et al. (2019) as the best among all HAR‐type models, performs

worse than all GARCH, ARMA and IV models.

2

One possible reason for this void in the literature is that BTC options are relatively young and thus may have been overlooked by researchers.

3

For example, trading volume on February 11, 2020 on Deribit was 159 million USD, while that on CME was 2 million USD.

4

A comparison of daily trading volume across BTC options exchanges is presented in Appendix A.

HOANG AND BAUR

|

1585