On Unspanned Latent Risks in Dynamic Term Structure Models
We explore the importance of information hidden from the yield curve and assess how valuable the unspanned risks are to a real-time Bayesian investor seeking to forecast excess bond returns and maximise her utility. We propose a novel class of arbitrage-free unspanned Dynamic Term Structure Models (DTSM), that embed a stochastic market price of risk specification. We develop a suitable Sequential Monte Carlo (SMC) inferential and prediction scheme that guarantees joint identification of parameters and latent states and takes into account all relevant uncertainties. We find that latent factors contain significant predictive power above and beyond the yield curve, providing improvement to the out-of-sample predictive performance of models, especially at shorter maturities. Most importantly, they are capable of exploiting information hidden from the yield curve and translate the evident statistical predictability into significant utility gains, out-of-sample. The hidden component associated with slope risk is countercyclical and links with real activity.
READ FULL TEXT