Utility of Smoothing Techniques in Yield Curve Modeling for Non-Steady State Data of Sri Lanka Capital Market

Traditional yield curve models such as Nelson–Sigeal parsimonious model represent the interest rate term structure at high accuracy with smoothed yield curve data. The model proposed by Nelson–Sigeal has been adopted by many countries across the world due to the economic interpretability of the parameters. Since developed capital markets do not experience turbulent economic conditions very often, the smoothness of the yield curve data does not bring challenges to modeling the yield curves. Countries like Sri Lanka used to face economic downturns frequently and the impact of such conditions influence the country’ substantially. Adopting the most flexible and interpretable yield curve model such Dynamic Nelson–Sigeal model is a challenge under such conditions. The yield curve data from January 2010 to May 2022 was examined and clustered into steady-state and non-steady-state data based on inflation and exchange rate movement. It was found the accuracy of the Nelson–Sigeal model deteriorates substantially in the non-steady state compared to the steady state based on the R-Squared and MAD. Further, the smoothness of the data in these two statuses is different, where smoothness is higher in a steady state over a non-steady state based on the autocorrelation function. Several smoothing techniques such as Spline Smoothing, Super Smoothing, Loess Smoothing, Lowess Smoothing, and approx Smoothing were employed for both statuses. Lowess smoothing seems to do a better job of smoothing for steady-state data, while the Super smoothing tool does a better job in non-steady states. Spline smoothing performed better than other smoothing tools on model accuracy tests (R-Squared, MAD, MAPD, Correlation of original vs predicted, Nash–Sutcliffe model efficiency, and Index of Agreement).