The asymmetric confidence intervals are the real insight here, way more valuable than the flat point forecasts from AR(p) models. What's interesting is how the direction of skew differs between Italy-Germany (upside risk) and France-Germany (downside risk), which probably reflects market perception of fiscal sustainability trajectories. The Nelson Siegel approach is elegant for yield curve modeling, but the λ parameter choice (fixed from Diebold Li rather than estimated) seems like it could introduce specification error, especially post-ECB QE regime changes. Sieve bootstrapping is the right call for non-Gaussian residuals, though I'd be curious whether a DCC-GARCH overlay would capture volatility clustering better during stress periods. The lack of sub-2yr maturities definitly limits short-end accuracy, which matters alot for policy-sensitive spreads.
The asymmetric confidence intervals are the real insight here, way more valuable than the flat point forecasts from AR(p) models. What's interesting is how the direction of skew differs between Italy-Germany (upside risk) and France-Germany (downside risk), which probably reflects market perception of fiscal sustainability trajectories. The Nelson Siegel approach is elegant for yield curve modeling, but the λ parameter choice (fixed from Diebold Li rather than estimated) seems like it could introduce specification error, especially post-ECB QE regime changes. Sieve bootstrapping is the right call for non-Gaussian residuals, though I'd be curious whether a DCC-GARCH overlay would capture volatility clustering better during stress periods. The lack of sub-2yr maturities definitly limits short-end accuracy, which matters alot for policy-sensitive spreads.
a DCC-GARCH is an insightful extension indeed!