Cryptocurrency Risks and Asset Allocation: What do Quantile Regressions, Spectral Analysis, and M-GARCH-based Dynamic Principal Components Say?
42 Pages Posted: 19 Sep 2019
Date Written: September 12, 2019
Simple optimizations unsurprisingly produce meaningful allocations to Bitcoin (XBT), even for modest levels of optimal portfolio volatility, given its extraordinary appreciation as well as minimal correlation with major asset classes from July 20, 2010 through August 30, 2019. Strictly speaking, this baseline result supports passive strategic allocation, gross of transaction costs, but what about the motivation and scope for active management? To start, GARCH-based methods that account for substantial time-varying XBT volatility, as well as dynamic covariance, do imply sizeable allocations across levels of risk. But optimal weights change meaningfully over time. Also, quantile regressions suggest that conditional XBT returns with respect to the S&P 500 are modestly convex in absolute terms as well as more positively skewed compared to other assets. Yet conditional symmetry is also hardly inert. Plus, spectral analysis shows XBT volatility primarily owes to higher-frequency cycles, of one week or shorter length, similar to assets such as shares. Nonetheless, the band-spectrum betas of XBT relative to the S&P 500 are substantially greater over longer cycles of at least one month, compared to shorter cycles, which suggests that XBT has been a less effective strategic, but no less useful a tactical hedge. Finally, among five major digital coins, dynamic principal components analysis produces substantial variation in the factor structure of cryptocurrencies.
Keywords: bitcoin, cryptocurrency, asset allocation, portfolio optimization, spectral analysis, quantile regressions
Suggested Citation: Suggested Citation