Toward Regime-Aware Risk Forecasts
研究了行业风险模型中估计回溯窗口的校准问题，发现不同波动环境需要不同模型，并提出利用跨截面预测离散度实时指导模型切换，实现体制感知的波动率预测。
Estimation lookback window is a key calibration parameter in industry-standard risk models. It determines how readily the model incorporates new data to form volatility forecasts. Most volatility environments can be characterized as either slow-moving or fast-moving, and no single calibration generates consistently reliable forecasts. A model’s strength in one environment is often the reason it is ill-suited for other environments. The potential impact of using a suboptimal model can be measured in real time with a cross-sectional dispersion of forecasts from different calibrations. This information can be used to inform timely, and disciplined, transitions between slow-moving and fast-moving models—the makings of regime-aware volatility forecasting.