Temakveld vår 2023

Date: 9. May, 2023

Time: 17:00, we start with mingling and an annual meeting followed by the talk

Format: Hybrid

Place: Abels utsikt (12th floor, Niels Henrik Abels hus - Universitetet i Oslo) and  Zoom https://uio.zoom.us/j/64170337111?pwd=dHJlcTA0SlRHQXRac3pVa1BXTGRsUT09 

Speaker: Nils Lid Hjort

Title: Bayes, Confidence Distributions, and US Pre-War Economy

Abstract:

Confidence distributions are frequentist analogues of the Bayesian posterior distributions, but without priors. Often these two approaches lead to similar results, regarding post-data intervals for primary parameters, etc. Situations with boundary parameters are different and more complicated, however. If your focus parameter is known to be nonnegative, a common Bayesian solution is to work with a flat prior from zero to infinity. I show that this might lead to very different and ostensibly wrong results, compared to confidence distributions. 

When Christopher Sims won the Sveriges Riksbank Prize in Economic Sciences im Memory of Alfred Nobel for 2011 (the so-called Nobel Prize in Economics), he chose for his Stockholm prize lecture to showcase how Bayesian modelling with MCMC worked for the analysis of a certain dataset on pre-war US macroeconomics (involving consumption, investment, government spending, 1929 to 1940). I illustrate the apparatus of confidence distributions with t-bootstrapping to analyse the same data and the same vector autoregressive time series model, but reaching rather different results when assessing the main parameters.

The essence of this story is contained in my book Confidence, Likelihood, Probability, with Tore Schweder, but I have more material and a different presentation of the main reasons for the differences with the Bayesian setup.