Learning Bayesian Statistics

#134 Bayesian Econometrics, State Space Models & Dynamic Regression, with David Kohns

Jun 10, 2025
Guest
David Kohns
David Kohns, a postdoctoral researcher at Aalto University, enriches the discussion with insights on Bayesian econometrics. He dives into the significance of setting appropriate priors to mitigate overfitting and enhance model performance. Dynamic regression is explored, emphasizing how it captures evolving relationships over time. State-space models are explained as a structured approach in time series analysis, which aids in forecasting and understanding economic dynamics. Kohns also discusses AI's role in prior elicitation, bringing innovative perspectives to statistical modeling.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Predictively Consistent Priors

  • Setting priors based on an interpretable statistic like Bayesian R-squared prevents overfitting in time series models.
  • Predictively consistent priors align prior expectations with prior predictive distributions, improving model reliability.
INSIGHT

Dynamic Regression with Time-varying Coefficients

  • Dynamic regression models allow regression coefficients to evolve over time following latent AR(1) processes.
  • This captures changing relationships, such as central bank policy responses that vary with economic conditions.
ADVICE

Avoid Inverse Gamma Priors

  • Avoid improper priors like inverse gamma on state innovation variances as they induce overfitting.
  • Use normal priors on state standard deviations for better-control over latent state variability.
Get the Snipd Podcast app to discover more snips from this episode
Get the app