Predicting
pandemics
Read more
case-studies-predicting-pandemics

Can we use London stock market data to identify late payments during Covid-19? How the innovative use of surrogate data led to a 70% improvement in late and no-payment rates.

The Challenge

With a global pandemic uprooting economic norms, stifling industry and contracting the job market, the need to predict late payments and identify defaults was crucial in managing the post-Covid credit cycle in the UK.

Initial focus was to predict which loans would fail their next payment, but with the advent of furlough schemes, self-employed income support and payment holiday support, the focus shifted to managing loan recovery.

The Purpose

  • Our model needed to identify which customers would resume payments after the UK’s payment holiday.
  • Distinguishing between those unable to pay, and those who would pay with prompt and without prompt, was essential for business planning and for revising collection strategy.
  • Findings would be applied to creating re-usable tools and methodologies for loan recovery.

The Method

First, we used open-source data to identify late payers and potential clients who had been put on furlough. Data collection points included the London Stock Exchange, The Companies House and online business listings, among other sources. By identifying industry, job title and company, the model could map likelihood of furlough or retrenchment.

Using application and origination data, Pepper could identify late payers by integrating key features like collection method – whether payment was made by live direct debit or by cheque; how many months in arrears they were before the payment holiday, how much of their household income came from self employment, and the age of the loan.

The Findings

  • Industry level stock market performance can be used to identify customers likely to face payment difficulties.
  • Self-employment is not a meaningful risk factor. The proportion of household income that comes from self-employment is of more use in credit decisions.

The Outcome

Using our data science model, our UK business launched a successful campaign that saw late payment fall from 14% to 2.5%, and no-payment decrease from 5% to 1.5%