<p>This paper proposes a Bayesian multivariate smoothing spline (BMSS) model to jointly estimate multiple yield curves, addressing the limitations of existing methods that either treat Treasury and corporate bonds separately or rely on intermediate credit spread modeling. By introducing a partially improper prior structure as a multivariate conditional autoregressive distribution (MCAR), our approach decomposes the precision matrix into two components: one for cross-sectional smoothing of contemporaneous data and another for temporal correlation, which captures the correlation of second-order derivatives of multivariate cubic splines under a stationary Gaussian process. This innovation enables information pooling across bond types-such as leveraging Treasury bond term structures to enhance corporate bond yield curve estimation-while directly handling sparse data scenarios (e.g., corporate bonds with minimal observations). Applied to simulated data and the Chinese fixed-income data, the model demonstrates robust performance with a root mean square prediction error, outperforming the Nelson–Siegel benchmark. Key advantages include ease of implementation, intuitive correlation interpretation, the elimination of credit spread modeling, and applicability to both theoretical research and practical fixed-income market analysis.</p>