Title: Probabilistic Bayesian updating of IOTs
Authors: Vladimir Potashnikov - RANEPA (Russia) [presenting]
Oleg Lugovoy - RANEPA (Russia)
Andrey Polbin - Russian Presidential Academy of National Economy and Public Administration (Russia)
Abstract: Efforts on developing and application of probabilistic method(s) for updating IO tables are summarized. The core of the methodology is the Bayesian framework which combines an information from observed data, additional beliefs (priors), and related uncertainties into the posterior joint distribution of input-output table (IOT) coefficients. The framework can be applied to various IOT problems, including updating, disaggregation, evaluation of uncertainties in the data, and addressing incomplete/missing observations. The flexibility of the methodology is partially based on sampling techniques. We apply modern Markov Chains Monte Carlo (MCMC) methods to explore the posterior distribution of IOT coefficients. Estimating IO tables by Bayes method is a computationally complex problem. The aim is to propose a modification of the algorithm, allowed to partially solve the problem of the curse of dimension. We also compare results with mainstream methods of updating IOT to investigate its performance. Various indicators of performance and application to various data suggest different results. The overall performance of the method is similar or comparable with mainstream techniques. The main advantage of the proposed methodology is an estimation of the full profile of joint probability distribution of unknown IOT matrices. The method can be also combined with any other techniques through prior information.