Estimating and forecasting the yield curve : Sri Lankan government securities market

Abstract

In this study, I evaluate two versions of the Nelson and Siegel (1987) model, namely the Nelson-Siegel model using the methodology presented in Diebold and Li (2006) and Nelson-Siegel-Svensson model (1994), with the purpose of fitting the current yield curve and forecasting the yield curve for the Sri Lankan government securities market.

The study finds that using the Svensson model which has an additional curvature factor compared to the Nelson -Siegel (Diebold and Li model) leads to a better in-sample fit of the term structure, and thus a better fit of the yield curve is observed. The superior in-sample fit of the Svensson model is clearly visible in the graphical outputs obtained and is further supported by the higher 𝑅 2 and lower RMSE associated with the Svensson model. The results obtained are robust for recent events such as the COVID -19 pandemic that affected the country.

Forecasting performance of the two models, indicated opposite results compared to results obtained in the estimation of yield curves. Yield curves from Nelson-Siegel (Diebold and Li) model are predicted better compared to the Svensson model under both the short forecast horizon of one month and longer forecast horizon of six months. This is clearly exhibited in the lower RMSE associated with the Nelson -Siegel (Diebold and Li) model under the rolling window forecasting design that was applied using an AR(1) forecasting model.