Loss Choice or Model Choice? The Role of Forecast Level in Cryptocurrency Volatility Forecasting

Battle Count: 0

Highly relevant for quantitative traders and risk managers. It clarifies how training loss choice impacts volatility forecasts and downstream VaR estimates, helping practitioners distinguish between model superiority and mere level shifts, which is critical for position sizing and capital allocation.

Moderate complexity. It requires implementing multiple loss functions, several distinct model architectures (including neural networks and gradient boosting), and a specific validation-based alignment procedure. The walk-forward cross-validation setup adds to the engineering effort.

The paper provides detailed specifications for data sources (Binance), models, training procedures (walk-forward, seeds), and evaluation metrics. However, no explicit GitHub repository link is provided in the text, which slightly limits immediate code reproducibility.