Employing decentralized liquidity pool data from UNISWAP-V2, we forecast the Value-at-Risk (VaR) and Expected Shortfall (ES) using the GARCH model with different error distributions, and, the deep learning probabilistic forecasting model algorithm DeepAR. The performances of the different forecasts are compared using an appropriate loss function. The GARCH model with normal distribution has been revealed to perform predominantly better, followed by the skewed t-student distribution when forecasting VaR. In contrast, the DeepAR model has demonstrated a poor forecasting capability for VaR in all cases - excluding liquidity pools with at least one stablecoin - however, prevails for the majority of the ES risk measures and data. Our findings recognize that, in part of the data, providing liquidity to a pair with at least one stablecoin is statistically significantly less risky than holding the same amount of crypto assets. Moreover, this research contributes to the development of new risk management tools and strategies for liquidity providers.
Comissão Organizadora
Anderson Odias da Silva
Claudia Yoshinaga
Ricardo D. Brito
Felipe Saraiva Iachan
Vinicius Augusto Brunassi Silva
Complete tool for submission, evaluation and publication of proceedings for scientific events.
