Regime Detection Based Risk Modeling of Asset Classes

Abstract

In this work, we have found a risk model that improves the performance of Risk Targeting. Risk Targeting in portfolio construction is implemented to improve capital utilization in growing markets and systematically step away from risk scenarios. However, the performance of risk targeting varies with different implementations of risk estimation. Risk Targeting using recent backward volatility estimates is the most popular risk targeting mechanism but it could not have anticipated a deep crisis like 2008 and it would hurt in range bound situations like the February 2018 drop and bounce back or the December 2018 drop and January 2019 bounce back. The drawbacks of recent volatility are that in such a risk model, short-term volatility is being used to detect a long-term risk event with and short term mean reversion effects are ignored .In this work, we will try to separate risk estimation into two risk models, a long-term risk model that predicts extreme risk scenarios based on macroeconomic data and a short-term risk model that adjusts risk based on short-term mean reversion effects. We then combine the output of the two risk models into a risk measure that enables a risk targeted allocation strategy to outperform static allocation in both crisis periods like 2008 and mean-reverting periods like 2018.Note that risk for a real-money investor is not an expectation of volatility but a measure of the probability of losing money if one is allocated to that asset class. Hence, we try to forecast a risk value which, if interpreted as a probability of loss, outperforms a baseline estimate of risk.