Measuring priorsensitivityandpriorinformativeness in largeBayesianmodels

In largeBayesianmodels,suchasmodernDSGEmodels,itisdifficulttoassesshow much theprioraffectstheresults.Thispaperderivesmeasuresofpriorsensitivityand prior informativenessthataccountforthehighdimensionalinteractionbetweenprior and likelihoodinformation.Thebasisforbothmeasuresisthederivativematrixofthe posteriormeanwithrespecttothepriormean,whichiseasilyobtainedfromMarkov Chain MonteCarlooutput.Weillustratetheapproachbyexaminingposteriorresultsin the smallmodelof LubikandSchorfheide(2004) and thelargemodelof Smets and Wouters(2007).