Banking on data and machine learning-driven credit analysis


May 5, 2020

We are just past the half time in the earnings season. The March quarter results of banks across US, Europe and other developed markets post a fairly grim picture. Provisioning is up resulting in an over 50% decline in profits. The results factor in approximately only one month of lock down.

COVID fallout on the banking sector will fully unfold over the next 6-9 months. Analysts believe that the worst is yet to come and June quarter results will see profits plunge further. Unlike other financial crises in the past, the current one is driven by a sudden, unforeseen and significant impact on the real economy. The 2008 crisis for example emanated from the financial sector and unfolded over 18 months across the wider economy which resulted in the NPL ratio of banks increasing by 3-5x across developed countries.

The current crisis is like a time lapse unfolding across the real economy in a matter of a few weeks, with over a third of the global population in some form of lockdown. The unemployment levels in the US which are projected to reach 16%, have not looked this bad since 1929 – when the country’s population was a third of what it is today. It would not be farfetched to assume that the banking NPLs – which is a reflection of the real economy would be similar if not worse. It is yet another reminder, if we needed one, of the impact of the COVID-19 pandemic on the global economy and poses three unique complications for banks:

  1. Severe and rapid liquidity crunch
    The combined supply and demand shocks have pushed borrowers to draw on their existing credit lines. This dramatic increase in loan drawdowns occurred amid a sharp increase in risk aversion and financial markets disruption.

  2. Unknown knowns
    While banks, in general, are reasonably well-capitalized and have liquidity, they will be severely tested with respect to handling defaults.
    Furthermore, COVID's impact has been heterogeneous across sectors and regions, and is rapidly evolving. Hence, banks which are inundated with borrower requests for restructuring, waivers, increase in credit limits etc. are required to undertake detailed, yet rapid reviews, of every request to determine the best course of action.
    The sheer number of requests means that their credit functions are overwhelmed. Normal underwriting criteria and credit risk models are not applicable currently, and over-reliance on "best guess" or “opinion of internal sector experts” unsubstantiated by data-driven granular analysis, could result in distressed lending.

  3. Operational capacity constraints
    Governments around the world have announced numerous economic packages and programs for businesses most affected by the pandemic including the Paycheck Protection Program (PPP)and the Main Street Lending Program in the US, the Coronavirus Business Interruption Loan Scheme and Bounce Back Loan Scheme in the UK, the Sonderfonds Finanzmarktstabilisierung (“SoFFin”) - the special financial market stabilization fund in Germany, etc. These much-needed liquidity lifelines for businesses are being channelled through the banking system which was already operationally strained before the crisis.

The irony is that the traditional credit models and underwriting criteria have broken down and don’t apply in the current circumstances. So, what should banks do?

We drew upon our experience at OakNorth Bank in the UK, and the experience of having run over $30bn of loans through the OakNorth Credit Intelligence Suite for our client banks globally since the onset of the crisis in Jan '20, to build a COVID Vulnerability Rating (CVR) Framework. The CVR framework enables banks to rapidly screen their existing portfolio and supplement their internal sectoral expertise with a data and machine learning-driven approach. The CVR framework, which uses forward-looking 2020 forecasts, enables banks to quickly determine the impact of COVID-19 on their portfolio, conduct credit analysis on those businesses which have been materially impacted, and then monitor those credits. The framework operates at three key levels:

  1. Screening: The screening tool helps banks to conduct, within days, a granular segmentation analysis of their entire Commercial and Industrial and Commercial Real Estate loan books based on COVID vulnerability (mapping the most vulnerable borrower segments or prospective growth niches). This is a prioritization tool.

  2. Credit Analysis: Post CVR screening, the suite performs borrower-specific credit analysis with revised forward-looking views on those borrowers which have been materially impacted.

  3. Monitoring: The suite proactively monitors the analyzed portfolio providing actionable credit alerts, news alerts, and event alerts. This effectively ‘flips’ the role of a portfolio management team from preparing annual reviews to actioning proactive alerts.

OakNorth’s CVR framework integrates domain (industry) specific COVID-19 stress case scenarios with built-in assumptions for impact on key financial metrics such as revenue, operating costs, working capital and CAPEX for FY20.

These industry specific scenarios are built leveraging OakNorth's core IP:

OakNorth Credit Intelligence has built 126 domain models which are further mapped to 1,600+ sub-sector groups consisting of an exhaustive list of business KPIs and financial metrics specific to that industry. The suite has further developed 34 COVID scenario’s mapped onto domain models to generate cash flow projections e.g. how fast (and to what extent) do typical borrowers in a specific domain adjust operating costs / working capital requirements with revenue contraction. These scenarios are continuously calibrated to the evolving nature of the pandemic by machine learning algorithms. Machine learning models such as competition identifier, clustering, driver analyzer, sentiment analyzer, etc. trained by OakNorth credit scientists draw upon millions of traditional (e.g. company data, industry and macro data) and alternative (geo-location, web-scraping, web traffic, consumer sentiment, point of sale, surveys) datapoints to overlay borrower specific idiosyncrasies into the credit analysis.

Never has it been more critical to understand not only cash flow but also borrower actions to reduce cash burn during this severely disruptive period. Different businesses are reacting differently with each management team having significantly diverse views on the right management actions to apply – one borrower’s view of ‘cutting to the bone’ can be another borrower’s view of not going far enough. Furthermore, each management team might have a different view on how long the crisis will last and what the recovery period will look like. It, therefore, becomes even more crucial for banks to have a consistent view on the scenarios they would like to apply consistently across their loan book. Given the break in historic correlations, it is essential for banks to have a revised forward-looking view on borrowers.

While travel, hospitality, F&B, CRE etc. are considered high-impact sectors, the impact of COVID-19 on others is not intuitive. If we look at machine and spare parts manufacturing as a category for example there could be one company that does machine and spare parts manufacturing for the food industry, such as cane crushing machine parts, or parts that go into a rice mill; and one company that does machine and spare parts manufacturing for the automobile industry, such as making engine crankshafts or sheet metal for the body of the car. On the face of it, these two companies belong to the same sector (machine and spare parts manufacturing), but because one caters to the food industry which is a non-discretionary spend industry, it will revive much faster than the company that caters to the automobile industry – a discretionary spend – as fewer people will be purchasing or upgrading cars given lockdown measures. Lenders, therefore, need to factor in that both of these sectors will have very different COVID stress and reboot scenarios, in addition to borrower-specific idiosyncrasies, while evaluating the credits.

Even within a particular sector, building forward-looking scenarios to factor cash burn, funding gap, debt capacity, and business profitability can help a credit analyst determine the viability of each company, and the extent of debt support they would require to reboot post-COVID.

The OakNorth Credit Intelligence Suite enables lenders to analyze such aspects within their portfolio and for individual credits at speed and scale and provides proactive monitoring resulting in better credit outcomes.

As the financial sector braces itself to support the stabilization and revival of the economy, while also preserving its asset quality, it would do well to remind itself that unprecedented challenges require extraordinary solutions.

About the author