PaperlayerView exact revision on arXiv

Source-linked AI summary

Bitcoin market route to maturity? Evidence from return fluctuations, temporal correlations and multiscaling effects

Stanisław Drożdż, Robert Gębarowski, Ludovico Minati, Paweł Oświęcimka, Marcin Wątorek

arXiv:1804.05916v2q-fin.STcs.CEecon.EM

TL;DR

The paper examines whether Bitcoin’s high-frequency fluctuations approach statistical features associated with mature markets. It compares return tails, temporal correlations, Hurst estimates, and multiscaling across trading periods; later return tails approach the inverse cubic pattern, while multiscaling does not emerge. The reported patterns provide evidence that Bitcoin may have entered a phase of maturity.

Abstract

from arXiv · show

Based on 1-minute price changes recorded since year 2012, the fluctuation properties of the rapidly-emerging Bitcoin (BTC) market are assessed over chosen sub-periods, in terms of return distributions, volatility autocorrelation, Hurst exponents and multiscaling effects. The findings are compared to the stylized facts of mature world markets. While early trading was affected by system-specific irregularities, it is found that over the months preceding Apr 2018 all these statistical indicators approach the features hallmarking maturity. This can be taken as an indication that the Bitcoin market, and possibly other cryptocurrencies, carry concrete potential of imminently becoming a regular market, alternative to the foreign exchange (Forex). Since high-frequency price data are available since the beginning of trading, the Bitcoin offers a unique window into the statistical characteristics of a market maturation trajectory.

I. INTRODUCTION

The paper frames Bitcoin as a candidate currency whose market fluctuations can be assessed against statistical features observed in mature markets. It identifies return distributions, temporal correlations, persistence, and multiscaling as key measures.

I. INTRODUCTION

  • The paper assesses Bitcoin’s market dynamics through return distributions, volatility correlations, Hurst exponents, and multiscaling effects.These measures are used to compare Bitcoin with empirically identified features of established, large-capitalization, high-liquidity markets.Source: §I · p. 2
  • Mature markets are characterized by fat-tailed returns, near-immediate decay of return autocorrelation, persistent positive volatility autocorrelation, Hurst exponents near 0.5, and broad singularity spectra.The paper uses “mature market” in this statistical, econophysics sense.Source: §I · p. 2
  • The paper presents Bitcoin as a peer-to-peer monetary unit that does not require issuance by a central bank or government.The introduction situates this novel currency amid substantial social and scientific interest, including concerns about market regulation and stability.Source: §I · p. 2

II. BITCOIN RETURNS AND VOLATILITY

The study analyzes Bitstamp Bitcoin trading records sampled at one-minute intervals from January 2012 through March 2018. Early trading includes extended zero-return intervals, while recent USD-denominated volume enters a substantially higher regime.

II. BITCOIN RETURNS AND VOLATILITY

  • The Bitstamp dataset spans January 1, 2012–March 31, 2018, with price changes sampled every minute.The study also records BTC volume, USD value, and consecutive one-minute bins with zero returns.Source: §II · p. 2
  • Early trading contains frequent zero-return intervals lasting on the order of 10^3 minutes, plausibly reflecting an absence of transactions.Figure 1 includes the number of consecutive one-minute zero-return bins alongside price and trading-volume trends.Source: §II · p. 2, Figure 1 · p. 3
  • Recent Bitcoin trading enters a USD-volume regime averaging about one order of magnitude higher.Figure 1 charts price, BTC and USD exchange volumes in hourly bins, and consecutive one-minute zero-return bins.Source: §II · pp. 2–3, Figure 1 · p. 3

A. Distribution of return fluctuations

Bitcoin return tails approach the inverse cubic power law observed in mature markets in the later trading biennia, while the earliest years have substantially heavier tails.

A. Distribution of return fluctuations

  • γ ≈ 3.2 in 2014–2015 and γ ≈ 3.3 in 2016–2017, close to the mature-market inverse cubic power law.For EUR/USD, GBP/USD, and GBP/JPY in early 2018, the reported exponents are respectively γ = 3.1 ± 0.2, γ = 3.1 ± 0.2, and γ = 3.2 ± 0.1.Source: A. Distribution of return fluctuations · p. 3
  • γ ≈ 2.2 in 2012–2013, indicating substantially heavier return tails than in the later biennia.Including these early years leaves the exponent essentially unchanged from the full 2012–2017 period, so they dominate its tail thickness.Source: A. Distribution of return fluctuations · p. 3

B. Volatility autocorrelation

Bitcoin return autocorrelation decays almost immediately, whereas volatility autocorrelation persists over long periods. The latter is strongest and most persistent in the 2016–2017 trading period.

B. Volatility autocorrelation

  • Return autocorrelation drops almost immediately to zero and can fall below zero, forming a “correlation hole.”The below-zero region is associated with a tendency for returns to change sign between adjacent time steps; Figure 3 contrasts return and absolute-return autocorrelation.Source: B. Volatility autocorrelation · p. 3, Figure 3 · p. 4
  • Volatility autocorrelation extends over roughly 10^5 minutes, or two months, and is strongest and most persistent in 2016–2017.The fall in autocorrelation beyond that horizon reflects the average time span of underlying volatility clusters.Source: B. Volatility autocorrelation · p. 4, Figure 3 · p. 4

III. NONLINEAR TEMPORAL CORRELATIONS

The section introduces detrended scaling analysis of varying-order fluctuation moments to study nonlinear temporal correlations in single or multiple time series. Its general method is Multifractal Cross-Correlation Analysis (MFCCA).

III. NONLINEAR TEMPORAL CORRELATIONS

  • MFCCA studies temporal correlations by inspecting scaling properties of varying-order fluctuation moments after detrending.The passage presents it as a general method for single-series and multiseries analysis.Source: §III · p. 4
  • The method is presented as suitable for nonlinear and multifractal correlations in single time series as well as among multiple time series.Source: §III · p. 4
  • MFCCA is carried out through the steps that follow in the section.Source: §III · pp. 4–5

A. Methods based on detrending

The methods construct cumulative signal profiles, detrend them in segments, and examine how fluctuation functions scale with time scale and moment order. They also derive multifractal spectra and a detrended cross-correlation measure.

A. Methods based on detrending

  • The series are converted to signal profiles, divided into segments from both ends, and locally detrended with fitted polynomials before cross-covariance is calculated.For each series, the profile uses deviations from its time-series average; polynomial order m = 2 is typical.Source: A. Methods based on detrending · p. 5
  • Fractal cross-dependencies appear as scaling relations in the qth-order covariance function, with q-dependent scaling exponents indicating multifractal rather than monofractal correlations.For monofractal correlations, the scaling exponent is q-independent.Source: A. Methods based on detrending · p. 5
  • MFDFA is the single-series special case of MFCCA, obtained by setting x_i = y_i; its generalized Hurst exponent h(q) supports calculation of the singularity spectrum f(α).The spectrum is derived from h(q) using the relations given in the section.Source: A. Methods based on detrending · p. 5
  • The multifractal spectrum’s width Δα quantifies complexity, while its asymmetry indicates whether multifractality is more developed for large or small fluctuations.The asymmetry index is defined from the left and right widths around the spectrum maximum.Source: A. Methods based on detrending · pp. 5–6
  • The qDCCA coefficient measures detrended cross-correlation across time scales and identifies, through q, which fluctuation amplitudes are most strongly correlated.Source: A. Methods based on detrending · p. 6

B. Hurst exponent

The BTC/USD Hurst exponent, estimated in monthly windows from July 2013, rises from strong early anti-persistence toward 0.5. The authors interpret its recent proximity to 0.5 as evidence that the market is approaching maturity.

B. Hurst exponent

  • BTC/USD’s Hurst exponent gradually approached 0.5 from below and recently became very close to it, which the authors interpret as a strong indication of approaching market maturity.Source: B. Hurst exponent · p. 6
  • Early BTC/USD H values were far below 0.5, indicating strong anti-persistence; estimates used one-month windows starting in July 2013 because earlier windows were biased by excessive consecutive zero returns.Each window comprised more than 40,000 data points to limit estimation error.Source: B. Hurst exponent · p. 6
  • For Jan 2–Mar 30, 2018, H was 0.48 ± 0.001 for EUR/USD, 0.49 ± 0.001 for GBP/USD, and 0.50 ± 0.001 for GBP/JPY.Source: B. Hurst exponent · p. 6
  • The BTC Hurst exponent dropped sharply around events apparently linked to uncertainty, while the authors associate the 2017 $1,000 threshold and BTC futures launch with maturation.Source: B. Hurst exponent · p. 6

C. Multiscaling

BTC’s return series develops increasingly clear scaling behavior across trading periods, culminating in mature-market-like multifractality in October 2017–March 2018. The earliest period’s zero returns distort small-fluctuation analysis, while surrogate tests remove the latest period’s multiscaling effects.

C. Multiscaling

  • Across 2012–2017, scaling in F(q, s) improves and its observable scale range expands, but fully developed multifractality does not emerge in those biennia.The latest biennium retains residual distortions associated with trading around the US presidential election.Source: C. Multiscaling · pp. 6–7, Figure 5 · p. 8
  • In October 2017–March 2018, BTC shows scaling across the examined scales, q-dependent h(q), and a left-asymmetric singularity spectrum with Aα ≈ 0.33, resembling mature markets.The multiscaling effects disappear in shuffled and Fourier phase-randomized surrogates.Source: C. Multiscaling · p. 7, Figure 6 · p. 8
  • In 2012–2013, zero-return intervals make fluctuation functions singular beyond s > 10^3; scaling appears only afterward and h(q) shows little q-dependence.The singularity regime shrinks by about one order of magnitude in 2014–2015 and largely disappears in 2016–2017.Source: C. Multiscaling · p. 7, Figure 5 · p. 8

D. Volatility vs. volume traded

The study finds cross-correlations between BTC volatility and trading volume that strengthen over time and are strongest for large fluctuations in the latest period. Smaller fluctuations show no scaling, and comparable effects are not observed between returns and volume.

D. Volatility vs. volume traded

  • In 2016–2017, volatility–volume cross-correlation reaches maximum synchrony for large fluctuations (q between 3 and 4), while smaller fluctuations show no scaling.For q < 0.4, the cross-correlation fluctuation functions fluctuate between positive and negative values.Source: D. Volatility vs. volume traded · p. 9, Figure 7 · p. 9
  • Volatility–volume correlations strengthen across biennia, shifting from relatively weak in 2012–2013 to large-fluctuation-dominated in 2016–2017.The complementary ρq(s) analysis covers scales from 10^2 to 10^3 minutes.Source: D. Volatility vs. volume traded · p. 9, Figure 7 · p. 9
  • Replacing volatility with returns yields no cross-correlation effects or scaling in Fxy(q, s) for any q-value.Source: D. Volatility vs. volume traded · p. 9

E. Log-periodic Bitcoin price patterns

The paper fits log-periodic oscillations to BTC’s price reversal on December 16, 2017, using a parameterized form for rising and falling phases. The fits reproduce the observed oscillatory patterns and are presented as additional evidence that BTC may have entered a mature-market phase.

E. Log-periodic Bitcoin price patterns

  • BTC reversed from a strongly increasing trend to a sharply decreasing one on December 16, 2017, with smaller-scale oscillations accompanying both trends.The paper describes such patterns as known in mature financial markets and related to discrete scale invariance.Source: E. Log-periodic Bitcoin price patterns · p. 9
  • The price parameterization combines a power-law term with log-periodic oscillations, using ω = 2π/log(λ) for the bubble phase and changing the sign of tc−t for the anti-bubble phase.Source: E. Log-periodic Bitcoin price patterns · p. 10
  • The log-periodic fits reproduce oscillations around the December 16, 2017 reversal, with a data-fit Spearman rank-order correlation of ρ = 0.96.The fits use λ = 2 and exclude activity before April 2017 because other markers indicated that the market was not yet mature.Source: E. Log-periodic Bitcoin price patterns · p. 10, Figure 9 · p. 10

IV. COMMENTARY

The authors conclude that BTC has rapidly developed statistical hallmarks observed in mature markets, including stocks, commodities, and Forex. They suggest other cryptocurrencies may follow a similar trajectory and that BTC’s high-frequency history offers a way to study market maturation.

IV. COMMENTARY

  • BTC has recently and rapidly developed statistical hallmarks observed in mature markets such as stocks, commodities, and Forex.The authors say other cryptocurrencies may follow this trajectory and that BTC’s high-frequency data since inception provide a window into market maturation.Source: §IV · p. 11
Loading 1804.05916v2…