Data science for the public good

International assessment of events leading up to the last financial crisis drew attention to an urgent need to improve financial statistics, to better support financial stability and prudential analysis.

One of the key areas for improvement in financial statistics, is to improve the granularity of the sector breakdowns. In the current publications of UK financial statistics, more than 100,000 organisations in the UK financial sector are grouped into three economic sub-sectors:

• monetary financial institutions – banks, building societies, Central Bank
• insurance companies and pension funds
• other financial institutions – all other types of financial institution, including money lenders, hedge funds, collective investment schemes, securities dealers, financial auxiliaries etc.

The ‘other financial institutions’ sub-sector clearly includes businesses with a wide range of different activities. If these institutions can be classified into these sub-sectors, it will underpin  more effective analysis of financial stability risk, and inform financial policy.

Our work

For the purpose of improving classification of the UK financial sector,a range of machine learning methods were applied to evaluate the possibility of predicting the company classification from available survey and administrative data. This consisted of three separate datasets: the Financial Services Survey, conducted by the Office for National Statistics (ONS), ONS’ Interdepartmental Business Register and the Financial Conduct Authority’s register of regulatory approvals for financial activity.

What’s the data science?

 The FinBins project targeted improvement of the UK financial sector classification through: finer-grained reporting, automatic anomaly detection in the (current) company classifications to increase the accuracy of reporting, and automatic re-classification of companies based on changes in their economic behavior as captured by surveys.

 What’s the impact?

Overall, the experiments have shown that although in general the achieved accuracy from currently available data is not sufficient for fully automatic classification of financial service companies, anomaly detection in classification of certain types of companies is possible. It is expected that linking-in of additional data sources will increase the range of companies that can be automatically monitored and flagged for reclassification if over time the data about them points to a change of their economic activities.

Our partners

ONS’ Enhanced Financial Accounts development, who, in close collaboration with the Bank of England, are working on improving the quality, coverage and granularity of the UK’s financial statistics