The present invention relates to a system for processing credit applications in the non-prime market.
In the financial industry, credit applications for non-prime applicants are typically processed manually. A typical manual workflow for processing credit applications is depicted in
Some lenders have deployed rule-based guidelines 6, typically implemented in software, which determine whether an application conforms to the lender's guidelines. For example, a rule may determine whether the loan originates in a state where the lender has a license to operate. Over time, lenders have implemented more comprehensive rules and systems are now able to automatically approve credit applications for the prime and non-prime market if all conditions associated with a loan product are satisfied. However, credit applications which do not fit into the exact requirements of a certain loan product are either rejected or referred for manual processing by a loan officer.
In fact, many credit applications, especially in the non-prime market, do not meet all conditions expressed in a rules-based system and must be manually processed. Some lenders resort to implementing more complex rules, so that a larger percentage of submitted applications can be automatically approved. However, in the non-prime market, most credit applications have at least one exception that presents a rules-based system with a difficult challenge. More complex rule sets partially alleviate this problem but still capture only a small percentage of presented applications. Consequently, rules-based systems have become more complicated and the vast majority of credit applications in the non-prime market must still be processed manually. A typical rules-based system for the non-prime market may implement hundreds or even thousands of rules which are needed to cope with the complexity of the credit decisions. The maintenance of such systems is extremely complex, cumbersome, and error-prone. Even when such a system is implemented and properly maintained, a significant number of credit applications still fall outside of the defined rules set and must be processed manually.
In addition to the mortgage industry and non-prime credit applications with complex approval requirements, this problem also exists in auto finance, credit card, Home Equity Line Of Credit (HELOC) and other credit lines.
One embodiment of the invention is a system for processing credit applications comprising a computer-implemented adaptive decision engine that produces a score determined by the risk associated with a credit application. In one implementation, credit applications are automatically approved or rejected, or referred for manual processing, based on a comparison of the risk score with predetermined thresholds.
Another embodiment of the invention is a method for processing credit applications. A risk score associated with a credit application is calculated, and the credit application is automatically approved or rejected based on a comparison of the risk score with at least one predetermined threshold.
Another embodiment of the invention is a computer program product stored in a tangible computer-readable medium. The computer program product includes an adaptive decision engine that produces a score determined by the risk associated with a credit application. A rules system automatically approves the credit application if all conditions are met, and refers the application to the adaptive decision engine for further risk assessment if conditions are not met.
These and other embodiments of the invention are described in more detail in the following description, drawings and claims.
a is a diagram illustrating use of risk thresholds by the method of the present invention.
b is a diagram illustrating use of risk thresholds by the method of the present invention.
Credit applicant 12, with the assistance of broker 14, has a credit application prepared and submitted to automatic underwriting system 16. Alternatively, applicant 12 may submit an application directly to automatic underwriting system 16, in which case broker 14 may be eliminated. The submission of the application may take place either through the Internet or via a dedicated connection. Broker 14 may also enter the application data directly into system 16, such as through an Internet portal.
Once the credit application is entered into automated underwriting system 16, rules system 18 checks the application for exceptions. System 18 either (1) rejects the loan (for example, if the lender is not licensed in the state of interest); (2) approves the loan if all conditions are met and enters it into loan origination system 24; or (3) refers the application to adaptive decision engine 20 if not all conditions are met as stated in rules system 18. Adaptive decision engine 20 checks if the exceptions, generated when conditions were not met, can be automatically waived. This is a balance between how many exceptions are waived and the perceived risk of a loan. The number of exceptions does not necessarily reflect the actual risk level associated with a credit application since it may contain enough compensating factors to offset the risk posed by failed conditions. For example, while an exception may state that the credit history of the applicant is not good enough for a certain loan product, the same applicant may be able to show that he/she is working in a stable job that pays well. The applicant's job status may then be used to compensate for the bad credit history.
Adaptive decision engine 20 may be a computer program implemented in computer hardware or software. The core of decision engine 20 is a predictive algorithm that is presented with credit application characteristics from historical data. Once trained, decision engine 20 is able to discriminate between high and low risk applications. Predictive algorithms that may be used by adaptive decision engine 20 to estimate the risk of an application include, without limitation, artificial neural networks, statistical algorithms such as linear and logistic regression, fuzzy logic, genetic algorithms, decision trees, or any other algorithm that is able to extract knowledge from data, i.e., data-driven algorithms. Adaptive decision engine 20 may also leverage predictive models, historical performance data, theoretical assumptions or any combination thereof.
If, for example, adaptive decision engine 20 implements a linear regression algorithm, it will learn during training the relationship between each loan characteristic and the target (low or high risk) associated with the entire credit application. In the case of a mortgage application, features such as “Length of Time in Residence”, “Debt-to-Income Ratio”, “Loan Term”, “Applicant's State”, etc. are weighted in relation to the target variable (low or high risk). If, for example, a higher “Debt-to-Income Ratio” correlates with a higher chance that the application is high risk, the linear regression algorithm will likely raise the coefficient associated with this feature during training so that, once deployed, the scores generated by adaptive decision engine 20 are higher whenever a high “Debt-to-Income Ratio” is encountered.
Adaptive decision engine 20 can learn both on-line and off-line. That is, decision engine 20 may adapt and learn based on data currently being processed on-line, and may also be “tuned” off-line based on historical data and model parameters. So, engine 20 may be tuned off-line based on historical data and model parameters and then applied to current real-time data in an on-line environment.
Using these methods, adaptive decision engine 20 produces a score that varies depending on the perceived risk associated with a credit application. For example, if the risk score is a value between 0 (low risk) and 100 (high risk), a threshold TH1 (28) for automatically approving loans can be set to, for example, 20 (
Applications with a risk score greater than the threshold TH1 are referred to loan officer 22 for manual review and decision to either approve or reject the application. Through use of adaptive decision engine 20 and risk thresholds, however, the number of loans that must be manually processed by loan officer 22 is much smaller than in current systems.
A second threshold TH2 (29) may also be set to automatically reject credit applications exceeding TH2. This threshold is useful to eliminate high risk credit applications which loan officer 22 does not need to waste time to consider. In
TH1 and TH2 may be set to the same value, so that all loans are either approved or rejected automatically. This eliminates the need for manual processing and allows processing of all credit applications in real time. Such a configuration is ideal for Internet-based implementations where applications are submitted online, and can be used by brokers (business-to-business transactions, also known as “wholesale” or “correspondent lending” in the mortgage industry) as well as credit applicants (business-to-consumer model, also known as “retail”). Because of the immediate response to a submitted credit application, this real time system will have excellent usability and high user acceptance.
The present invention embraces implementations of adaptive decision engine 20 using TH1 only, TH2 only, or both TH1 and TH2. In addition, decision engine 20 may be placed before rules system 18 in the workflow. The decision engine result would then be submitted as additional input to rules system 18 and facilitate risk-based exception handling in a similar fashion.
One advantage of the present invention is that rules system 18 and adaptive decision engine 20 can give immediate feedback to broker 14, who will then know immediately whether a loan is approved or rejected. The only time delay occurs when manual processing is needed.
The decision to approve or reject a credit application can be based solely on risk score (as calculated by adaptive decision engine 20) or on a combination of risk score and total number of exceptions. For example, an application may be rejected if it contains an excessive number of exceptions according to rules system 18, even if it has a relatively low risk according to adaptive decision engine 20.
In another embodiment, a lender may not have a rules system 18 and may use adaptive decision engine 20 solely to determine whether credit applications should be approved or rejected. In this case, adaptive decision engine 20 is trained with historical credit data applicable to a specific loan product. A variant of this situation could be a simplified rules system in combination with decision engine 20.
Other embodiments and implementations of the invention will be or will become apparent to one of ordinary skill in the art. All such additional embodiments and implementations are within the scope of the invention as defined by the accompanying claims.