R.C. Giltner Services, Inc. (RCGS)

Digital Lending Underwriting,

Model Governance Control and Validation

Financial modeling for underwriting and fully automated digital lending are increasingly important to the banking industry. Many institutions, from Suntrust’s Lightstream to Eastern Bank’s Express loan for small business,are now using models and digital lending all under their underwriting and control.

Our digital lending technology and automated underwriting model presented here focus on specific loan segments: unsecured loans under $100k for small business underwritten by the owner/guarantor’s credit and the business cash flow, and unsecure consumer loans under $30k. Our technology allows our clients the ability to fully set and automate the underwriting standards using the same underwriting credit report standards they use today.

These segments have historically been marginally profitable at best for FIs, generated a large number of expensive declines and presented material compliance risks.As a result, FIs have seen a decline in market share of these key relationship segments.


The underwriting for these loan segments often produced low losses, but the cost of underwriting and delivery made the loan product unprofitable at the outset. The use of automated underwriting models combined with the efficiencies of digital delivery allow FIs to pursue these smaller market segments profitably, including digital automation of adverse action and declines and comprehensive automated compliance. For example, the new profitable value chain is shown below for a small business loan of $38k as “Digital” compared to the existing revenue and cost value chain for “Manual” for FIs. A similar model for consumer loans follows. With the delivery efficiencies provided by fully automated digital lending for these segments, FIs can profitably accept losses documented now with 30 years of history based on credit report information as used with credit cards.

Automated Underwriting and Digital Delivery of a Small Business Loan vs Manual

Automated Underwriting and Digital Delivery of a Consumer Loan vs Manual

These tools represent a significant advance for the industry, and the models and digital delivery represent a new source of risk management for FIs . In this document, we provide the following information for documenting and managing this risk associated with the underwriting models used in RCGS digital lending applications. This information builds on the detail provided in FDIC guidance and regulatory documents Supervisory Insights on Scoring and Modelingand Scoring and Modeling.

This document is organized into four parts:

  1. Model Scope of Risk
  2. Model Governance
  3. Model Control Practices
  4. Model Validation
  1. Model Scope of Risk

The FDIC governance and regulatory guidance is most concerned with risks created in the application of underwriting models. The emphasis in regulatory review of use of underwriting models is directly related to the level of risk loans generated by the underwriting models might create for the financial institution. Financial institutions with national credit card portfolios and large regional consumer lending strategies for specific market segments will come under a high level of scrutiny because these loan portfolios will be a material part of total loans for the institution. In our digital lending focus working with community and regional banks, the total loans generated by our digital lending platform are less than 1% of the total loans outstanding for the bank. Thus the risk assessment of our models in the institution’s ALLL risk management is not as material as applied in most large-bank applications.

Nevertheless, our clients are often generally new to using models for underwriting decisions. Even for the small volume of total loans we generate, we want to follow good model procedures of Model Governance, Model Control Practices and Validation.

  1. Model Governance

Model governance assures the following principles are applied in use of our models in digital lending:

  • Your board reviews and approves policies providing oversight throughout the organization commensurate with overall reliance on models. We provide the relevant information for your board and management to review and to approve the models we use in our digital lending platform.
  • Your business line managementreviews and approves policies and controls over each of our model's uses.This includes an analysis involving the FI’s imperial data against the digital credit criteria to validate expected outcomes, adding detail to this analysis. The FI needs to delineate what the outcome of the new digital platform will look like vs. the existing portfolio. This effort should include adding program guardrail parameters such as:“If delinquencies exceed ___ or losses exceed ____ the program will be discontinued.”
  • Our implementation has all parts of our model under the control and approval of your line of business executives.
  • Your bank staff with appropriate independence and expertise periodically validate that the model is working as intended. We provide at the outset model validation and performance expectations based on well-documented industry experience, and then provide reporting and tracking of loans and losses according to our model factors for your ongoing review.
  • Your internal audit may test model control practices and model validation procedures to ensure compliance with established policies and procedures. We provide the documentation and reporting for your internal audit your control practices and validation procedures.

Specifically, in our implementation we provide model policies for our three models as presented below as examples only.

Model A Policy: Small Business Lending Less Than $50,000

Definition: The model uses the following variables in defining the availability of a loan,

loan amount and interest rate, or the denial of the loan:

  1. The business loan guarantor is valid and authorized to apply for and guarantee the loan.
  2. The credit score and related credit information of the business loan guarantor.
  3. Late payment experience by the guarantor in terms of 30 day and 60 day late payments within the standards set by the financial institution.
  4. Age of checking account in use with the financial institution as a proxy for the minimum time in business.
  5. Deposit activity through the checking account measuring coverage ratio algorithms for deposit levels and required monthly loan repayment levels.
  6. NSF activity affecting the accuracy of “true” deposits and account management.

Model B Policy: Consumer Lending Less Than $30,000

Definition: The model uses the following variables in defining the availability of a loan,

loan amount and interest rate, or the denial of the loan:

  1. The loan recipient and guarantor is valid and authorized to apply for and guarantee the loan.
  2. The credit score and related credit information of the loan recipient and guarantor identifying his or her overall level of debt relative to his or her ability to repay debit.
  3. Late payment experience by the guarantor in terms of 30 day and 60 day late payments within the standards set by the financial institution.
  4. Age of checking account in use with the financial institution.
  5. Deposit activity through the checking account and coverage ratio algorithms for deposit levels and required monthly loan repayment levels.
  6. NSF activity affecting the accuracy of “true” deposits and account management.

Organization controls and validation polices and materials for Model Governance are:

  1. We provide documentation and training for your board, management, operating members and other team members in terms of the acquisition and use of the model through our Vendor Due Diligence documentation, implementation manual materials and onsite training.
  1. We install and train on how our digital lending tools provide your management complete control of the settings and management of the variables above.
  1. We provide testing that the digital lending platform is behaving as predicted with access and calculation using the model variables.
  1. We provide detailed documentation of how these model variables have been used with a long history of accurately predicting risk and loss levels in general populations like the customer base of the bank prior to the bank implementing our digital lending platform and model.
  1. We provide ongoing tracking of results for comparison to expected losses according to our model factors.
  1. Model Control Practices

Our model control practices are presented below in the following sections:

  1. Model theory
  2. Model operating procedure
  3. Data reconciliation procedures
  4. Security and change control procedures
  1. Model Theory

Credit Score Based Loans

Our Model A small business loan (< $50,000) and Model B consumer loan (<$30,000) are underwritten primarily on the individual guarantor or consumer credit report. In this way, the loan is underwritten similarly to a credit card. We add to this information, validation of ability to repay the deposit information going through the checking account over time.

Credit score and credit information has a 30-year history of assessing risk levels of losses in making a loan. We provide our clients expected losses according to the credit score and credit information they allow in our model so they can predict risk. Numerous studies have validated a similar pattern as shown by the CFPB publicly available study, “Analysis of Differences in Consumer Credit Scores, 2012”. Many publications by industry provides from Fair Issacs to Trans Union have published similar results in proprietary studies. Based on credit score information alone, we project our population range of credit scores in our model to be between 670 and 800, with a weighted midpoint of 710. Using this lost history information, we projecta 5% 90-day delinquency ratio and a 3.5% loss ratio without inclusion of deposit information.

Many, of course, continue to search for unique factors to further clarify risk, but our models use these validated methods for financial institutions to make loans to small businesses or consumers.

Further, default risk and credit score information is not static but follows macroeconomic trends. We look at credit card default rates over time to adjust the mean estimated loss rate for the credit score range and distribution projected based your risk choices. This history of a long macroeconomic period for all credit card loans is shown below. For example, while losses in current years may be close to 2.5% and well below our estimate above. However, losses have been as high as 7% in the last decade. We start with our expected delinquency over a period of time with credit score only information at 5%.

Source: FDIC

We add to this information checking deposit information. Many have documented that a deposit relationship and deposit information further reduces credit losses compared to just using credit score information alone. Deposit information not only tells us more about the cash flow of the business and coverage ratios in ability to repay, but linking the loan to a checking deposit relationship also strengthens the loan. Unlike a credit card where a customer can continue transacting with a separate checking account even if they are in default on the credit card, in our model, defaulting on a line of credit linked to a checking account can bring transaction account closure and risk evaluation. With checking account information, our loss projection for clients is reduced.

Specifically, studies like “The State of Small Business Lending,” Harvard Business School, 2016, document:

“Interpreting cash flow data using the Demand Deposit Account (DDA) as part of assessing credit worthiness is a major innovation brought by online lenders.

Banks could have formidable advantages if they can better tap DDA data, given they have such data on every small business customer which has an account with them.”

Further, having DDA access in relation to the loan for the business or consumer addresses a number of critical underwriting compliance, fraud and risk management factors.

–Simplifying BSA/CIP validation

–Validating income

–Determining deposit-to-loan ratio (50%-300%)

–Simplifying disbursement speed and eliminating fraud risk by requiring stablished deposit account

–Improving risk with right of offset

–Providing early warning of loan problems

–

For these small business loan segments, credit monitoring, watch list and risk rating can be automated with benefits from tracking deposits as a proxy for revenue and deposit factors such as NSF and loan draw, which are better early identifiers of challenges than just the owner-guarantor credit. For example, reporting we track shows the watch list notification for a loan with no owner-guarantor credit report changes, but significant changes in the deposit cash flow and NSFs.

Credit Monitoring Example – Small Business Loan

In our financial modeling and pro forma, we apply sensitivity analysis in using charge off levels as high as 6%. Even in this extreme case, our average loan of $15,000 generating revenue of approximately $2,000 and losses at 6% of $900 is still profitable for the bank even for the bank with our technology providing delivery and maintenance costs of under $200 annually. Such small loans today with traditional, more expensive underwriting, management and maintenance costs are marginally profitable. Finally, even in these most extreme cases, recall that these loans amount to less than 1% of the financial institution’s total loans.

A sample underwriting matrix for business loans and consumer loans are shown in the tables below:

Business Underwriting RCG Manage Site Sample Data Example

Consumer Underwriting RCG Manage Site Sample Data Example

The individual components are set by the financial institution. The underwriting process is a linear process evaluating components within the matrix beginning with the credit score and ending with the sustained level of monthly deposits, which approximate the businesses gross cash flow.

If the applicant passes the parameters in the matrix listed above, the cash flow of the business (monthly deposits) is used to determine the line of credit granted. For example, this is largely determined through an acceptable level of debt service coverage as shown below for a business loan.

The underwriting measures the cash flow on a 33-day basis as well as an average monthly deposit level based on a 96-day basis. The lower of these two cash flow values is used to determine the sustained monthly deposit level of the business.

The line of credit is based on this level of sustained monthly deposits. The sustained monthly deposit figure is then multiplied by a factor. This factor is determined by the guarantor’s credit score. The better the credit score, the higher the factor used in determining the amount of the loan.

A debt service coverage is then calculated based on the monthly repayment of the loan. The sustained level of deposits is multiplied by a Profit margin percentage. This profit margin assumes a level of cash available to service the debt. A debt service coverage ratio is calculated using the cash available to pay the debt compared to the monthly payment.

Loan Grade

Based on the underwriting matrix above, the business loan is then assigned a credit grade consistent with the bank’s internal loan grading system.

Ongoing credit monitoring and loan grade migration

An institution should have robust risk management practices to identify, measure, monitor, and control credit risk in its lending activities. The individual loans generated through this underwriting process are dynamically monitored on a quarterly basis.

The system will pull a Clear Fraud Score for the guarantor with the initial credit pull and quarterly thereafter. The Clear Fraud Score includes a larger range of credit parameters than the standard credit score. The Clear Fraud Score is designed to project defaults earlier than a typical credit score.

Examples of our loan monitoring are shown below for business loans. Note the one loan message exception on NSFs.

Drill down analysis tracks key small load risk data like guarantor credit score, deposits as a revenue proxy, loan balance and NSFs.

Model Settings and Functioning.

The design and functioning of our model and settings are documented in our implementation manual.

  1. Model operating procedure

Our model works through our installed technology platform and data gathered daily from the bank’s core system and credit bureau information. We provide detailed information on the data request and model matrix in our manuals, test and assure the technology is working as planned and testing validation that the calculations of our technology are accurate.

  1. Data reconciliation procedures

Our testing and implementation procedures assure data is captured and analyzed accurately. Once our technology is installed and functioning, we do ongoing tests of accuracy of daily imports from your core, as well as our integration and data collection with our credit bureau. Our systems are built to maintain the integrity of data and external audits of our processes document our data integrity.