Using machine learning to predict bankruptcy

In an unstable business landscape better predictive models are needed. ML is already showing promise

At the start of this year, new research revealed that the UK was expected to face the largest increase in bankruptcies of any EU economy during 2017, with the number of businesses going insolvent predicted to reach 20,254. And sadly, 2017 has not been short of stories of high-profile UK companies collapsing - particularly in the retail industry. Just this month, Multiyork, and Feather and Black announced they would be falling into administration, joining the likes of high-profile retail stores such as Jaeger which went under earlier this year. Toys R Us, too, hit the headlines and surprised the markets by filing for bankruptcy in September.

Assessing the risk

In this rather bleak landscape, companies, lenders, and investors have undertaken considerable amounts of research to look into how they can use statistical models to proactively monitor and predict the likelihood of bankruptcy.

By taking data - such as gross margin, debt to assets, and cash flow - and plugging it into programs to process the information, financial analysts have been able to create parametric and non-parametric models for assessing risk. These rather traditional models allow companies to determine whether they should continue working with certain partners, vendors, or clients. Banks and other lenders also use these insights to assess whether an organisation is a suitable candidate for a commercial loan, while investors can decide if they are going to buy or sell stock in a company.

However, while hugely valuable, these models require extensive analysis which is not only time consuming but also the source of inaccuracies. With the ability to process and analyse large amounts of data quickly, recognise patterns and form intelligent decisions, could artificial intelligence (AI) and machine learning provide a better alternative?

Predictive patterns

There can be no denying that AI and machine learning (ML) have fast become the buzzwords of 2017. While often used in the same vein, the two technologies are not the same thing. Machine learning is an approach to AI whereby a system learns or "trains" itself using an initial dataset, then improves from experience without explicit programming. It's all about sorting through vast amounts of data to identify meaningful patterns, predict news ones, and anticipate upcoming problems.

With this in mind, our team of researchers at Genpact recently collated 11 years' worth of financial data on 267 bankrupt companies and 585 healthy organisations across the IT, industrial and healthcare sectors in order to show whether machine learning could, in fact, predict bankruptcy in advance of its occurrence.

At first, we normalised the large dataset and implemented it into trained learning models to recognise the warning signs of bankruptcy. The models sifted through the data and found a direct relationship between financial metrics and bankruptcy risk. We then reverted back and selected companies that had filed for bankruptcy to see if the models were correct.

Our research revealed that the team could accurately predict bankruptcies two years ahead of an event occurring - a task which would have been nearly impossible by manually reviewing the data and using traditional statistical methods.

Digitally dependent decision making

Predictive analytics has the potential to completely transform virtually every industry. In healthcare, for example, a ML model can train on a database of patient records to understand patients' symptoms and identify illnesses early on. Insurers can use ML to analyse data on an applicant's driving record, their personal information and other variables, compare them with similar people and calculate the best insurance policy with the lowest amount of risk.

It's all about leveraging the data at our fingertips. We are told that, today, data output is around 2.5 quintillion bytes a day - and that's only going to increase as the world becomes more digitally-focused. But all this data is meaningless if we can't extract the value it holds.

Machine learning can uncover hidden patterns in this data and give businesses the insight they need. Advancements in this technology allow an organisation to move from saying from ‘I think' to ‘I know'. What's more, predictive analytics will eventually enable businesses to move away from assessing ‘what happened' to instead confidently say ‘I know what will happen', allowing them to stay one step ahead of the game.

Srini Bharadwaj is CTO AI products at Genpact