DNS–JSV: Stochastic Volatility with Jumps in a Dynamic Nelson-Siegel Model

This thesis develops a Dynamic Nelson–Siegel term structure model with Jump Stochastic Volatility (DNS–JSV), in which the stochastic volatility process incorporates a time-varying and state-dependent jump component. The model is evaluated in an out-of-sample density forecasting exercise, using daily U.S. Treasury yield data. The results indicate that DNS–JSV delivers superior predictive performance relative to Dynamic Nelson–Siegel and Dynamic Nelson–Siegel–GARCH benchmarks over evaluation windows extending up to one month following major episodes of financial stress. These findings suggest that incorporating state-dependent jumps in the volatility process can improve density forecasts in the aftermath of periods of financial stress, at the expense of a modest reduction in predictive accuracy during more tranquil market conditions.