Can wavelets produce a clearer picture of weak-form market efficiency in Bitcoin?

Abstract

This study proposes a new approach for testing for random walk behavior in daily Bitcoin returns (19/07/2010–03/03/2022) by contextualizing the Dickey-Fuller test in time-frequency space using continuous complex wavelet transforms. By splitting our full sample into smaller sub-sample periods segregated by Bitcoin halving dates, we find that Bitcoin returns are most predictable or least market efficient (i) at higher frequency or short-run cycles of between 2 and 16 days, (ii) between November-February months, (iii) during ‘bubble’ periods, (iv) across the consecutive halving dates, (v) during the ‘Black Swan event’ caused by financial market turmoil arising from the COVID-19 pandemic, and (vi) subsequent to the announcements of new COVID-19 variants. Altogether, our findings have important policy implications for different stakeholders in Bitcoin markets.