any interesting papers in applied micro using bayesian techniques? a very good one i found was meager Understanding the Average Impact of Microcredit Expansions: A Bayesian Hierarchical Analysis of Seven Randomized Experiments.bump
Bayesian is popular in quant marketing, led by people like Peter Rossi at UCLA (formerly of Booth's Metrics & Stats) and Greg Allenby at OSU. Check out their Bayesm package and book: https://www.amazon.com/Bayesian-Statistics-Marketing-Peter-Rossi/dp/0470863676
Hierarchical bayes, once you substitute all the nested linear models for the beta parameters, end up as a standard linear model for the observed variables where you get very convoluted error structures, that introduce a lot of structure about the unobservable shock of the model, essentially a very complex random effect specification. This affects mostly to the efficiency, and if you are wrong you take the risk to get a very bad estimation. By contrast a fixed-effect OLS approach is simple and consistent (better from the Occam Razor perspective)
But then again, fixed-effect OLS implies linearity with varying intercept. That's a pretty strong assumption if you think about it.
https://doi.org/10.1017/S0266466600008719
Really? What can bayes add to TS that unit root testing, Granger Causality, and cointegration cannot answer?
Depends. Useful for macro (time series + estimation of DSGE models) but other fields are not much into bayesian stuff.
Hierarchical bayes, once you substitute all the nested linear models for the beta parameters, end up as a standard linear model for the observed variables where you get very convoluted error structures, that introduce a lot of structure about the unobservable shock of the model, essentially a very complex random effect specification. This affects mostly to the efficiency, and if you are wrong you take the risk to get a very bad estimation. By contrast a fixed-effect OLS approach is simple and consistent (better from the Occam Razor perspective)
Under regularity conditions, if the parameters have a finite dimension then the Bayesian posteriors asymptotically shrink to the maximum likelihood estimator, and therefore to the parameter of interest. Moreover, any centrality measure of the posterior (conditional mean, or mode) has the same asymptotic distribution as maximum likelihood. In practice, there is absolutely no advantage in one over another, they work the same. Any other statement is just cheap talk of illiterate people.
However, if the parameters have infinite dimension (a nonparametric curve modeled by sieves), Bayesian estimators can have important failures (inconsistent), there are well known examples, and only in some specific cases the Bayesian nonparametric estimators work well.
Agree with second paragraph. Disagree with first paragraph. Bayesian and frequentist methods can have dramatically different finite sample behavior - and in real life N is finite. Quoting a bunch of asymptotic results is sweeping this fact under the rug.
Under regularity conditions, if the parameters have a finite dimension then the Bayesian posteriors asymptotically shrink to the maximum likelihood estimator, and therefore to the parameter of interest. Moreover, any centrality measure of the posterior (conditional mean, or mode) has the same asymptotic distribution as maximum likelihood. In practice, there is absolutely no advantage in one over another, they work the same. Any other statement is just cheap talk of illiterate people.
However, if the parameters have infinite dimension (a nonparametric curve modeled by sieves), Bayesian estimators can have important failures (inconsistent), there are well known examples, and only in some specific cases the Bayesian nonparametric estimators work well.Agree with second paragraph. Disagree with first paragraph. Bayesian and frequentist methods can have dramatically different finite sample behavior - and in real life N is finite. Quoting a bunch of asymptotic results is sweeping this fact under the rug.
If parameters have infinite dimensions, frequentist methods completely collapse. You can't compare it to Bayesian methods, because there is nothing to compare it to.
Under regularity conditions, if the parameters have a finite dimension then the Bayesian posteriors asymptotically shrink to the maximum likelihood estimator, and therefore to the parameter of interest. Moreover, any centrality measure of the posterior (conditional mean, or mode) has the same asymptotic distribution as maximum likelihood. In practice, there is absolutely no advantage in one over another, they work the same. Any other statement is just cheap talk of illiterate people.
However, if the parameters have infinite dimension (a nonparametric curve modeled by sieves), Bayesian estimators can have important failures (inconsistent), there are well known examples, and only in some specific cases the Bayesian nonparametric estimators work well.Agree with second paragraph. Disagree with first paragraph. Bayesian and frequentist methods can have dramatically different finite sample behavior - and in real life N is finite. Quoting a bunch of asymptotic results is sweeping this fact under the rug.
If parameters have infinite dimensions, frequentist methods completely collapse. You can't compare it to Bayesian methods, because there is nothing to compare it to.
Are you drunk? Have you ever picked up book on non-parametric inference?
If parameters have infinite dimensions, frequentist methods completely collapse. You can't compare it to Bayesian methods, because there is nothing to compare it to.Are you drunk? Have you ever picked up book on non-parametric inference?
If parameters have infinite dimensions, frequentist methods completely collapse. You can't compare it to Bayesian methods, because there is nothing to compare it to.Are you drunk? Have you ever picked up book on non-parametric inference?
Irrelevant to quoted point.