Title: Country-specific uncertainty indices and machine learning: The case of Russia
Authors: Wojciech Charemza - Vistula University (Poland)
Svetlana Makarova - University College London (United Kingdom) [presenting]
Krzysztof Rybinski - Vistula University Warsaw (Poland)
Abstract: Problems related to the construction of country-specific geopolitical and economic policy uncertainty indices have been identified and analysed. The methodology is based on the textual analysis of data extracted from local language newspapers through unsupervised machine learning. The problems include (1) identification of the external and internal (idiosyncratic) factors affecting uncertainty; (2) crowding out and covering up the uncertainty-related topics by other news items; (3) identification of country-specific effects of global geopolitical and economic policy uncertainties. Problems (1) and (2) have been tackled by using the Latent Dirichlet Allocation algorithms with various settings to recognise economic and policy-related topics and applying Word2Vec model to categorise uncertainty-related terms. For problem (3), quantile correlation techniques have been applied to find a relationship between changes in the global (worldwide) uncertainty and country-specific local language indices. The data used are collected from four main Russian newspapers and cover the period from 1992 to 2018.