Short-Term Risk and Adapting Covariance Models to Current Market Conditions

Covariance models of stock returns appear throughout the investment process, e.g., forecasting portfolio risk, hedging, constructing Markowitz return-risk optimal portfolios, and algorithmic trading. Typically built from historic time-series, they estimate the past but - because markets and regimes continually change - not the current and future.There are non-time-series, instantaneous ways to infer volatility, e.g., intra-period trading range, implied volatility, and big data. This paper presents a framework for incorporating these inferences into any linear factor covariance model so that it forecasts forward instead of echoing the past