Driving the economic wellbeing of individuals, the performance of companies and the policy options open to government, the economy involves complex networks of financial relationships: from international trade flows and everyday spending decisions, to supply chain connections and employment arrangements.
The aim of the Turing’s economic data science research programme is to apply data science and AI techniques to how the economy works, and to use this insight to address challenges of national and international importance.
The research programme is funded through a five-year strategic partnership between the Institute and HSBC.
The research programme has the following three challenges.
Economic activity over the business cycle
How to analyse new economic activity data
New data, whether from the government or private sector, has made it possible to measure economic activity – outputs, employment, wages, spending, regional or international trade etc – faster, more precisely and at a more disaggregated level than has previously been possible.
This will mean that individual, business and government decisions can be based on a much more complete and up-to-date picture of what is going on in the economy.
Additionally, it will allow a better understanding of economic networks and the drivers of productivity differences. It will also enable researchers to address issues relating to household consumption and saving decisions, company financing over the business cycle and the economic impact of Brexit.
Changing nature of work
How to measure and account for the changing nature of work
Recent decades have seen important changes in the labour market across four dimensions.
First, there have been shifts in how long people spend at work. This includes the growth in part-time work as well as changes in participation in various forms for men and for women.
Second, how work is organised has changed. The size and structure of employers has shifted, and outsourcing has grown in importance. Alongside this, a growing number of people are on zero-hours contracts or are self-employed (some in the so-called “gig economy”) or in new forms of informal employment.
The third change is in what work involves. This includes both the types of occupation available and the tasks performed while people are at work.
And finally, how people return to work has changed. Wage growth since the financial crisis has been subdued. Those lower down in the distribution have tended to fare better than those higher up, but similarly, there have been important developments among the very rich – with implications for inequality. Occupations and skill levels have also altered substantially.
These changes raise challenges for measurement and create a need to understand what is driving the changes, analysing the implications for welfare and establishing what role policy has to play.
Machine learning for economic data
How can machine learning improve economic models?
Econometrics uses statistical models to describe economic systems and to assess the impact of policy interventions. Modern econometrics emphasises causal inference, drawing conclusions about cause and effect, and counterfactual analysis, comparing what actually happened and what would have happened in the absence of a particular intervention.
Typical econometrics models are built for quantitative datasets of a limited size. However, data in various different forms is increasingly available. Observations now regularly appear with hundreds or even thousands of recorded variables, whilst text, satellite images, and web search profiles have non-standard data structures, containing vast amounts of information.
Machine learning methods that train a computer to learn the mapping between inputs and outputs without explicitly being told what it is have developed in the last decade in response to the growth of such data. These methods usually prioritise computationally efficient algorithms that can accurately make predictions about outcomes beyond the data that was used to train them.
The goal of this theme of the Economic Data Science programme is to develop an understanding of the role of these machine learning techniques in empirical economics, gaining of economic insights from direct and indirect observation.
The first part is methodological. For example, when and how do economists need to adapt machine learning algorithms to address their empirical challenges? What are the statistical properties of the estimates that algorithms produce?
The second part is to apply machine learning methods within specific fields to answer economic questions that traditional data cannot.
For more information, please contact Programme Manager Dr Mahlet (Milly) Zimeta [email protected]