Dependent Metaverse Risk Forecasts with Heteroskedastic Models and Ensemble Learning

Abstract

Metaverses have been evolving following the popularity of blockchain technology. They build their own cryptocurrencies for transactions inside their platforms. These new cryptocurrencies are, however, still highly speculative, volatile, and risky, motivating us to manage their risk. In this paper, we aimed to forecast the risk of Decentraland’s MANA and Theta Network’s THETA. More specifically, we constructed an aggregate of these metaverse cryptocurrencies as well as their combination with Bitcoin. To measure their risk, we proposed a modified aggregate risk measure (AggM) defined as a convex combination of aggregate value-at-risk (AggVaR) and aggregate expected shortfall (AggES). To capture their dependence, we employed copulas that link their marginal models: heteroskedastic and ensemble learning-based models. Our empirical study showed that the latter outperformed the former when forecasting volatility and aggregate risk measures. In particular, the AggM forecast was more accurate and more valid than the AggVaR and AggES forecasts. These risk measures confirmed that an aggregate of the two metaverse cryptocurrencies exhibited the highest risk with evidence of lower tail dependence. These results are, thus, helpful for cryptocurrency investors, portfolio risk managers, and policy-makers to formulate appropriate cryptocurrency investment strategies, portfolio allocation, and decision-making, particularly during extremely negative shocks.