We are training an XGBoost model to predict tomorrow's VIX changes. Because our options portfolio is highly sensitive to volatility spikes, a 5-point spike hurts us ten times more than a 5-point drop. How do you handle this asymmetry directly in the model training phase?

我们在训练一个 XGBoost 模型来预测 VIX 明天的变动。由于我们的期权组合对波动率飙升极其敏感，VIX 涨5个点对我们的伤害是跌5个点的十倍。你如何直接在模型训练阶段处理这种非对称性？