Network causality structures among Bitcoin and other financial assets: A directed acyclic graph approach

https://doi.org/10.1016/j.qref.2018.05.016Get rights and content

Highlights

  • Study relations between Bitcoin and other assets via DAG.

  • Contemporaneous analysis indicates that Bitcoin is quite isolated.

  • Lagged relationships exist between Bitcoin and some assets, especially when the former is in bear state.

  • Integration between Bitcoin and other assets varies over time.

Abstract

We use a data-driven methodology, namely the directed acyclic graph, to uncover the contemporaneous and lagged relations between Bitcoin and other asset classes. The adopted methodology allows us to identify causal networks based on the measurements of observed correlations and partial correlations, without relying on a priori assumptions. Results from the contemporaneous analysis indicate that the Bitcoin market is quite isolated, and no specific asset plays a dominant role in influencing the Bitcoin market. However, we find evidence of lagged relationships between Bitcoin and some assets, especially during the bear market state of Bitcoin. This finding suggests that the integration between the Bitcoin and other financial assets is a continuous process that varies over time. We conduct forecast error variance decompositions and find that the influence of each of the other assets on Bitcoin over a 20-day horizon does not account for more than 11% of all innovations.

Introduction

With the emergence of Bitcoin as an investment asset (see Baur, Hong, & Adrian, 2017; Bouri, Molnár, Azzi, Roubaud, & Hagfors, 2017a), an increasing number of international participants have become involved in this market, which is the largest of today’s cryptocurrencies. Recent spikes in transaction and trading volumes evidence this trend. Furthermore, Bitcoin’s market capitalisation has increased exponentially, from 4.5 billion USD at the end of 2014, to more than 41.6 billion USD as of June 2017. Despite the rising scholarly interest in the economics and finance of Bitcoin, the extent to which the Bitcoin market has integrated into the markets of other asset classes remains largely unexplored. Specifically, there is a perceived threat that the Bitcoin market represents a potential source of financial instability, which suggests the need to monitor its integration into the global financial system (European Central Bank, 2012). Additionally, we must enhance our limited understanding of Bitcoin’s market integration with other financial assets for several other reasons. First, it affects the design and implementation of policies for maintaining financial stability. Second, it influences the decisions of policy makers in countries that are likely to consider Bitcoin as an official digital currency or as part of their foreign reserves. Third, it affects investor inferences regarding asset allocation and risk management.

The few existing studies considering the relations between Bitcoin and other economic and financial assets have mostly relied on unconditional correlations (e.g. Baur et al., 2017; Brière, Oosterlinck, & Szafarz, 2015) or are limited to the hedging ability of Bitcoin (e.g. Bouri et al., 2017a; Bouri, Jalkh, Molnár, & Roubaud, 2017c). In the present study, we offer a broad view of contemporaneous causal flows (see Awokuse & Bessler, 2003) between Bitcoin and several asset classes (i.e. equities, bonds, currencies and commodities) via the use of a purely data-driven approach, called the directed acyclic graph (DAG). We endogenously detect structural breaks and derive the forecast error variance decompositions (FEVDs). We also calculate network centrality (i.e., the importance of one market in a network relative to other markets), based on the work of Ahern and Harford (2014).

Our research contribution arises from two main aspects. First, the application of the DAG approach allows us to map the causal order without relying on ad-hoc network structures while avoiding unsubstantiated assumptions. Second, the enrichment of the discussion concerning the relationship between the largest cryptocurrency – Bitcoin – and other financial assets provides new empirical evidence under different market situations. Interestingly, our methodological approach allows for uncovering significant differences in the relationship across three sub-periods, with less levels of segmentation shown during the bear-market state. Such findings extend our limited understanding of Bitcoin integration by revealing a time-varying nature of market integration that seems to contradict the general view in the current empirical literature that Bitcoin is isolated from the global financial system.

In our empirical analyses, we consider several financial assets, including the more conventional investments, such as international equities, bonds, and currencies, as well as commodities. In choosing these financial assets, we refer to prior studies that consider the relationship between Bitcoin and key asset classes from the global financial system (Baur et al., 2017; Bouri et al., 2017a,c), which can provide useful implications for the stake of investors and policy makers. In examining the equity market, we consider a global equity index and pay particular attention to Chinese equities, given that Chinese investors and users represent an important group of stakeholders in the Bitcoin market (Bouoiyour & Selmi, 2015; Bouri et al., 2017a). We focus on a general commodity index and on gold prices, as several studies refer to Bitcoin as a ‘digital commodity’ or ‘digital gold’ (Baur et al., 2017; Dyhrberg, 2016). We also include energy commodities in the empirical analysis, because energy, particularly in the form of electricity, represents the main input in Bitcoin mining (Li & Wang, 2017; Bouri et al., 2017c; Hayes, 2017). Investment-grade bonds are also part of the analysis because their role as a proxy for sovereign risk might contradict with the role of Bitcoin as new asset class independent from sovereign authorities (Brière et al., 2015; Baur et al., 2017; Bouri et al., 2017a). We also focus on the US Dollar Index given the use of Bitcoin as a (digital) currency (Baur et al., 2017; Bouri et al., 2017a; Polasik, Piotrowska, Wisniewski, Kotkowski, & Lightfoot, 2015).

We structure the remainder of this paper as follows: Section 2 discusses Bitcoin and a selection of its related literature, Section 3 describes the materials and methods, Section 4 provides the empirical results, and Section 5 summarises the conclusions.

Section snippets

Bitcoin and asset classes

Bitcoin is an electronic scheme that facilitates the transfer of value between parties. Based on peer-to-peer networking and cryptographic protocols, it allows users to make anonymous transactions, just as with cash, but through the Internet and without the need for financial intermediaries. In this sense, Bitcoin is fully decentralised without the intervention of third parties, such as central banks or government financial agencies (Weber, 2016). Interestingly, the design of its protocols

Materials and methods

We studied the interdependence between Bitcoin prices and the other financial variables through the application of the vector autoregression (VAR) and error correction model (ECM) techniques. We use the DAG approach to identify the contemporaneous causality among the examined variables, and then we estimate the FEVDs based on the causal order we determined from the DAG results. Due to the wide application of VAR/ECM techniques in the empirical literature, we paid special attention to the DAG

Sub-period I

We construct a VAR model and apply the Johansen cointegration test (Johansen & Juselius, 1990). Results in Table 3 Panel A show that the null hypothesis (i.e. no cointegrating vectors) cannot be rejected, suggesting a lack of cointegration among these variables during sub-period I. We further estimate a VAR model in first differences (i.e. without cointegration) to obtain the contemporaneous correlation matrix of innovations. According to the Akaike information criterion (AIC) and the

Conclusions

The present study contributes to the debate surrounding the causal relationships between Bitcoin and several financial assets (i.e. equities, bonds, currencies and commodities) using a DAG-based approach and FEVDs.

Focusing on the contemporaneous causality between Bitcoin and all the asset classes under study, empirical results suggest the isolation of the Bitcoin market. However, based on the time-lagged causality structure, the causal relationships seem to be time-variant. Specifically, there

Acknowledgements

The first author acknowledges the support from the National Natural Science Foundation of China under Grant No. 71774152, 91546109; and the Youth Innovation Promotion Association of the Chinese Academy of Sciences (Grant No. Y7×0231505).

References (44)

  • X. Li et al.

    The technology and economic determinants of cryptocurrency exchange rates: The case of bitcoin

    Decision Support Systems

    (2017)
  • G. Pieters et al.

    Financial regulations and price inconsistencies across bitcoin markets

    Information Economics and Policy

    (2017)
  • K.R. Ahern et al.

    The importance of industry links in merger waves

    The Journal of Finance

    (2014)
  • T.O. Awokuse et al.

    Vector autoregression, policy analysis, and directed graphs: An application to the US economy

    Journal of Applied Economics

    (2003)
  • J. Bai et al.

    Estimating and testing linear models with multiple structural changes

    Econometrica

    (1998)
  • J. Bai et al.

    Computation and analysis of multiple structural change models

    Journal of Applied Econometrics

    (2003)
  • D.G. Baur et al.

    Bitcoin: Medium of Exchange or speculative assets?

    Journal of International Financial Markets, Institutions and Money.

    (2017)
  • R. Böhme et al.

    Bitcoin: Economics, technology, and governance

    Journal of Economic Perspective

    (2015)
  • J. Bouoiyour et al.

    What does bitcoin look like?

    Annals of Economics and Finance

    (2015)
  • E. Bouri et al.

    Bitcoin and global financial stress: A copula-based approach to dependence and causality in the quantiles

    Quarterly Review of Economics and Finance

    (2018)
  • E. Bouri et al.

    Bitcoin for energy commodities before and after the december 2013 crash: Diversifier, hedge or more?

    Applied Economics

    (2017)
  • M. Brière et al.

    Virtual currency, tangible return: Portfolio diversification with bitcoins

    Journal of Asset Management

    (2015)
  • Cited by (216)

    View all citing articles on Scopus
    View full text