DNS/AFNS Yield Curve Modeling Enforcing the ZLB

A very nice paper by Feunou et al., “Tractable Term Structure Models,” just appeared in Management Science (link here).  Basically it develops a highly-tractable DNS/AFNS-style approach to yield curve modeling that enforces the zero lower bound (ZLB).

[The following is adapted from private correspondence from one of the authors.  Any errors or misunderstandings are mine.] 

They introduce a new framework that facilitates term structure modeling that has both (1) positive interest rates, and (2) flexible time series dynamics, but which is also tractable, meaning quick and robust estimation.

 

They build on the dynamic Nelson-Siegel (DNS) and arbitrage-free Nelson-Siegel (AFNS) models, as in DL2006, DRA2006, CDR2011 and the references therein.  They seek a DNS/AFNS-type model that (1) enforces the zero lower bound, (2) allows for time-varying volatility, (3) is computationally tractable, and (3) enforces freedom from arbitrage.

 

They find what they're looking for, almost.  From the introduction: 

...this broad family of models delivers bond prices that are free of dominant trading strategies in the sense of Rothschild and Stiglitz (1970) and Levy (1992). Intuitively speaking, the no-dominance (ND) property rules out almost all arbitrage opportunities, and we show that any remaining bond trading arbitrage opportunities cannot be self-financing and do not survive in the presence of nonzero short-selling costs. Overall, this analysis suggests that the practical relevance of DNS models generalizes to a broader...framework [that includes imposition of the ZLB].

Sounds good to me.  Simple imposition of the ZLB is a very welcome addition to the  DNS/AFNS toolkit.