Covid-19 Recession – Integrating New Risk Driver Data into Risk Models


By Robert Phelan, CFA, GARP

( , 973-2727-3603,

April 4, 2020

A very successful risk leader said, “the data will set you free.”  By that, he meant, you do not need to know fundamental credit analysis because the patterns in historical data using machine learning will reveal most of what is required to predict credit risk.  Maybe not with Covid-19.  Thanks to the intertwined global economy, this pandemic has created a demand recession and health crisis like the world has never seen.  The bottom line is that the risks are NOT YET in the data.

Before I get to solutions that I am proposing for clients, I want to highlight some shortcomings of how the US is addressing the issue:

  • The most significant gap is that Congress is attempting to support workers without regard to the sustainability of the businesses that employ them. Congress requires companies to maintain employees on their payrolls until the loan is repaid.  Business owners are reluctant to take the SBA loans because of the uncertainty of when the economy will improve.  JPMorgan and other banks have said they will delay their participation in the Small Business Paycheck Protection Program because the requirements are not clear.  In other words, the banks are worried that small business owners may not adhere to the terms required for the ultimate forgiveness of these loans by the government.  The banks will then have losses, as no business owner will want to repay a loan requiring payroll for an indefinite period.  The process of getting loans is ineffective. Banks do not want to lend to small businesses at a 1% interest rate, given the uncertainty.
  • The size of the SBA loan package is only $350 Billion, even though there are 23 million Small and Medium Enterprises (SMEs) in the US. Effectively, Congress has only allocated $15,217 for each SME.  Ignoring the rent, utilities, and overhead needs of SMEs, the available money would only cover two weeks of pay. (assuming 20 employees as an average SME). Obviously, most SMEs will not survive on these loans alone if the economy does not restart soon.
  • Many of the larger businesses are furloughing a large portion of their workers. With millions going on unemployment, the States are not prepared to process a large number of claims and get money to laid-off workers promptly.

How should lenders to SMEs adjust their risk models? Most large banks are using machine learning risk models to predict the default risk of their small business customers.  Clients have found it easier to add risk drivers into a logistical regression model as a quick way to detect SME weakness, rather than rebuilding their machine learning models. As described in the attached article and R programming by Matt Mills, the default prediction of an existing gradient boosting machine model could be combined with new risk drivers as input into a new logistic regression model designed to predict the risk of SMEs impacted by new risk drivers or in this case, Covid-19.

The following data elements below are used as independent variables into a new logistics regression model to identify which SMEs are most vulnerable to Covid-19 using very recent data.  The dependent variable is the probability that an existing SME’s balance will go 30 days past due and delinquent.  Another approach would be to scale up the prediction of credit and liquidity risk by using the following indicators of Covid-19.

Some of the independent variables are:

  • An indicator of SMEs in any impacted sectors: Airlines, Oil & Gas, Travel, Hotels, Automobile Dealers, Advertising, Luxury Goods, Restaurants, Consulting, Retail, and any suppliers to the impacted sector.
  • An ordinal indicator of the number of employees at each SME (1= SME with less than five employees, 2= SME’s with 5 to 30 employees, 3= SME with greater than 30 employees)
  • Indicator (1=Yes, 0=No) if the SME is in an impacted Metro Area: NYC, DC, Seattle, LA, New Orleans, etc
  • Total SME Debt to Trailing 12 Month (TTM) average spend
  • An indicator of the strength of the most recent fundamental credit rating
  • Ability to capture monthly bank account balance or pull payments (YES=1 is a positive risk offset)
  • Current billing period spend versus TTM average spend (prorated)

While I have managed credit and fraud through three major credit events, many firms have risk managers who have not managed consumers or SMEs through a recession.  The economy is in a recession. The key for any risk manager is acting quickly to reduce exposure through proper alignment of balance limits, billing cycles, and extended repayment options that the customer can manage.  Of course, acquisitions should be aligned with conservative underwriting to rebalance the portfolio toward a larger mix of firms with steady revenue and strong balance sheets. Do not cut off customers with viable businesses as that action accelerates defaults and losses.

Carefully assessing, monitoring, and acting to reduce risk vulnerabilities protects the firm’s long-term economics but may increase other risks.  Inaction and internal focus expose a firm to obsolescence. Not addressing risks threatens survival and diminishes competitive advantage. Getting an independent risk assessment creates a dialogue on the prudent levels of risk action for any given environment.



Simon, Ruth; Rudeger, Peter; Omeokwe, Amara. “Confusion Veils Loan Plan for Businesses.”  The Wall Street Journal, Dow Jones & Company, 2 Apr. 2020,

Mills, Matt. “gradient boosting vs. logistic – free photo” Google Search, 3-Apr.2020

Stat Mills.  “Using Boosted Trees as Input in a Logistic Regression in R” Stat Mills, 9 Aug. 2016,

Author: RiskDirector, LLC

Risk Management Advisory and Consulting

Leave a Reply