Econometric tools for interdisciplinary applications with big data



'Big Data' provide both opportunities and hazards. The former arise from the vast amounts of potential information; the latter from the substantial chance of finding spurious connections. An approach to benefit from the former and minimise the latter is presented, focusing on time-series data where there are many more candidate explanations than available observations. To construct viable models requires jointly addressing not only the set of likely explanatory variables but also potential outliers, location shifts and trends, lagged responses and non-linearities. Appropriately formulated automatic model specification and selection approaches can tackle this huge task, and check for substantive links. The applications described include modelling UK real wages over 1860-2014; ascertaining pollution responses to policy changes; establishing that atmospheric CO2 changes are anthropogenic; modelling UK CO2 emissions as first-in and one of the first out of the Industrial Revolution; and detecting the impacts of volcanic eruptions in global temperature reconstructions over the last millennium.


<< Atrás