Bjørn Eraker is a financial econometrician at the Department of Econometrics and Statistics, Graduate School of Business, University of Chicago (at time of publication; Norwegian background, thesis from Norwegian School of Economics and Business Administration, 1997). His research focuses on Bayesian MCMC methods for continuous-time models with latent factors, particularly diffusion and jump-diffusion processes applied to asset pricing and interest rates. His 2001 JBES paper adapts data augmentation MCMC to general Itô diffusions and demonstrates that CEV interest-rate models are misspecified relative to stochastic volatility models.
Key Contributions / Features
Eraker (2001) data-augmented MCMC for Itô diffusions: Introduces m−1 auxiliary latent data points between each pair of discrete observations (Δt = 1/m) to eliminate Euler-discretization bias; Brownian bridge AR-MH proposal draws each latent Yi∼N(21(Yi−1+Yi+1),21σi−12Δt), exact for constant drift/diffusion (Proposition 1) and converging as Δt → 0 in general (Proposition 2); 6-step Gibbs sampler for CEV model (β≈0.75 vs. GMM ~1.5); block Gibbs for CEV+SV two-factor model (κ_z → first-order AC ≈ 0.976; σ_z ≈ 0.27); reparameterization trick (Appendix D) subtracts running chain mean Zˉ(g) to eliminate downward bias in log-volatility persistence κ_z; SV model dramatically outperforms CEV on Q-Q fit and conditional variance dynamics for 2,288 weekly US 3-month T-bill yields 1954–1997.