Taming Momentum Crashes

Stochastic Volatility Models with Skewness Selection

“…A traditional stochastic volatility model is defined in Equations ( 1)- (5). Equation (1) represents asset returns with a mean of zero and dynamic volatility of exp(h t /2).…”

“…Equation (2) indicates persistent log-volatility with level µ and persistence ϕ, having initial values given by the stationary distribution shown in Equation (3). In its simplest form, both measure and state equations have Gaussian errors as represented in Equations ( 4) and (5).…”

“…More recently, refs. [5,20] used GAS models to model Bitcoin returns and manage momentum-based portfolios. However, unlike our proposal, GAS models are suitable for including time-varying higher moments within a GARCH-like framework, while our proposal is designed for SV models.…”

“…Ref. [5] is one key example of the empirical benefits of adding such feature for modeling crosssectional stock momentum. While the momentum factor is known for delivering good mean-variance compensation, it is also subject to a long period of negative performance.…”

“…By capturing such prolonged periods of likely negative returns via dynamic skewness, ref. [5] improves the performance of the stock momentum factor compared to traditional approaches that neglect skewness.…”