|
Abstract:
|
Motivated by the successes of the Billion Prices Project in predicting inflation, we investigate whether and how more information can be extracted from price data harvested from the internet. More specifically, we look at how expert knowledge regarding price dynamics, inter-product price dependencies and index construction allows us to construct models leading to accurate, uncertainty-qualified predictions for official index figures before they are published. The methodology developed for computing predictions makes use of a large number of ARIMA models related to each other according to a hierarchical aggregation structure. Treating the structure as a Bayesian hierarchical model, we produce coherent and computationally tractable algorithms for adjusting index estimates in light of new data.
The work presented is a continuation of a project carried out in collaboration with the UK Office for National Statistics, in which product-level indices are estimated from internet price data.
|
