Ayliffe, K. and Rubin, T. (2020). A quantitative comparison of yield curve models in the MINT economies. EPFL Infoscience. URL: https://infoscience.epfl.ch/record/279314
Software: https://github.com/tomasrubin/yield-curve-forecasting
Description:
This work originates from the semester project written by Kelly Ayliffe written at EPFL under my supervision.
Abstract:
A yield curve is a line plotting bond yields (i.e.\ interest rates) as a function of their maturity date (their “expiration date”). When on a national scale, the yield curve represents the underlying interest rate structure of a country’s economy. It is fundamental in many areas of finance as it helps understand the expected evolution of an asset’s value, and therefore serves as a basis for pricing various assets and their derivatives, valuations of capital, determination of risk and more. Furthermore, yield curves are widely considered as a general indicator of a country’s overall economic health.
During this study, we conducted a series of quantitative comparisons of various yield curve models, specifically on the MINT (Mexico, Indonesia, Nigeria, Turkey) economies. We implemented the one-step and two-step Dynamic Nelson-Siegel methods as well as traditional time-series models such as vector autoregressions with single and multiple lags and autoregression, using the random walk as a benchmark for comparison.
The results confirm what was perhaps to be expected: forecasting yield curves is no easy tasks, and in some cases the random walk cannot be outperformed by the selected methods. There are, however, other cases where the presented methods thrive. In these situations, the Dynamic Nelson-Siegel methods in particular, both one-step and two-step, proved to be considerably superior to the random walk.