Project description
New probabilistic models could help manage unstructured data
The amount of data generated each day is overwhelming. Most of it is unstructured, meaning that it cannot easily be stored in databases. It is therefore more difficult to analyse and is not searchable. Common examples of unstructured data include text files, transaction data, images and web browsing histories. The EU-funded UnStruct project plans to build novel probabilistic models that link unstructured data directly to relevant economic parameters. Studies will focus on three main themes: how information about economic conditions is dispersed amongst agents, and how they aggregate it through interactions; the evolution of an economy hit by multiple shocks; and transaction payments between firms.
Objective
Most usable data is unstructured. Examples include text, transaction data, images, and web browsing histories. Although rich and plentiful, most economists do not use unstructured data. The few that do generally quantify it with off-the-shelf algorithms that are unrelated to the economic environment in which it is generated, which makes connecting it to economic models difficult. I instead propose to build novel probabilistic models of unstructured data that link it directly to relevant economic parameters. This powerful approach will use the information in unstructured data to test and estimate economic models in a way that is not currently possible with existing methods.
I will focus on three distinct themes. The first studies how information about economic conditions is dispersed among agents, and how they aggregate it through interactions. This process it at the heart of the policymaking process, and the use of text data provides a unique opportunity to structurally model this information in innovative ways.
The second theme jointly models unstructured data and the evolution of an economy hit by multiple, unobserved shocks. This will provide a novel forecasting tool, which is of key interest to policymakers. But it will also use unstructured data to estimate equilibrium models of the macroeconomy, and hence recover economic fundamentals.
The final theme will use transaction payments between firms, and extend probabilistic models of network formation to create new definitions of markets that go well beyond anything in the current literature. This will contribute to measuring market power and the transmission of economic shocks, both questions of fundamental importance.
Beyond these specific themes, my research will also pave the way for the use of probabilistic machine learning that combines novel data with clear economic models. The frameworks I introduce will provide a template for others to follow in the future.
Fields of science
Programme(s)
Funding Scheme
ERC-COG - Consolidator GrantHost institution
WC1E 6BT London
United Kingdom