Yield Curve Analysis of Kazakhstan’s Government Bonds Using PCA and Time Series

Abstract

We apply Principal Component Analysis to the monthly panel of Kazakhstan's sovereign zero-coupon yields from June 2018 to January 2026, covering 26 maturities, and find that three principal components explain 99.9% of the cross-sectional variance, with loadings matching the canonical level, slope, and curvature factors documented for developed markets. The factor dynamics align closely with Kazakhstan's 2022 macroeconomic and geopolitical shocks: the level factor accelerates with the National Bank's tightening cycle, the slope plunges sharply after February 2022, and the curvature factor spikes during episodes of elevated uncertainty. Augmented Dickey-Fuller tests and AR(1) estimates reveal a clear persistence hierarchy: the level factor is near unit-root with a half-life of roughly 20 months, while slope and curvature are stationary with half-lives of 3.5 and 1.3 months respectively. In an out-of-sample forecast comparison, our PCA+AR(1) framework outperforms both Principal Components Regression and the Random Walk benchmark at the long end of the curve, with statistical significance at the 26-year maturity under the Diebold-Mariano test.