When it comes to dealership fraud, what makes one car dealer riskier than another?
More than $7 billion in originated loans that contained fraud or misrepresentation in 2020 alone. Essentially, almost every dealer will likely encounter at least one fraud case every year. Statistically speaking, for every 200 applications submitted by dealers to lenders, at least one will contain misrepresentation that could lead to default of the loan. These random applications represent a low percentage of the over $500 billion in total originations every year. They don’t necessarily make a dealer risky.
But how can lenders decide exactly when a fraudulent loan issue at a dealer is bad luck? And when it might be something more? What actually makes a dealer risky?
Machines Can Help Identify the Patterns of Dealership Fraud and Risk
To answer this question, it’s only natural to turn to machines and alternative data. These types of solutions can help reveal insights into patterns of dealer risk. With machine learning, we can identify signals in the data. These signals contain information about a dealer and the specific activity that indicate fraud.
Machine learning has been used for years to tackle the problem of fraud. Machine learning can apply thousands of calculations per second. So it is often the perfect technique to uncover statistical deviations that are key signals to fraud behavior.
For example, credit card issuers use machine learning to profile each customer’s typical credit card spending habits. Then they match those habits to every new transaction made on their account. If a fraudster steals a credit card, the spending pattern will change rapidly. Therefore the models will alert the bank that the card could be compromised.
Point Predictive has used those same machine learning techniques to understand dealer fraud risk. We analyzed our data set of more than 94 million historic applications submitted by nearly 100,000 car dealers nationwide. The data included detailed information on the performance and status of each application submitted. The data included not only performing loans, but also fraud, first payment default, early payment default and charge-off.
The goal of the study was to understand the key signals in the data that point to dealer risk. We wanted to know the common factors that could help predict fraud and early payment default. Essentially, we wanted to know which dealers submitted risky application, which did not, and how frequently they did.
8 Risk Factors Common in Dealership Fraud
Machine learning revealed the most common patterns in high-risk applications:
Risk Factor #1: High Average Application Fraud Scores
The most obvious sign of potential dealer fraud or risk can be found in the application-level fraud risk. This is a leading indicator of future default losses.
We relied on a score range of 1 (lowest risk) to 999 (highest risk). Then, we tracked the average application fraud score for every dealership in the data. We found that dealers with average application fraud scores greater than 900 had fraud and loss rates 66% higher than average. Conversely, dealers with average application fraud scores less than 100 had significantly lower rates of fraud and default than the average.
It’s worth repeating: The best predictor and signal of future default at the dealer level is their current application risk.
Risk Factor #2: High Variability and Fluctuations in Average Application Fraud Scores
Inconsistency in a dealer’s submitted application-level risk is another signal of potential fraud. So we tracked each dealer’s monthly fluctuations in average application fraud scores. Surprisingly, we found that dealers with a wider spectrum of risk (larger fluctuations in scores) tend to have fraud rates 48% higher than the average risk profile.
Risk Factor #3: Average Percentage of Down Payment Paid in Cash
Another signal of dealer risk is how much cash their borrowers are putting down on the car. There were some dealers whose applications indicated zero cash down on all their deals. Again unsurprisingly, those dealers have fraud and loss rates close to 40% higher than the average fraud rate.
Risk Factor #4: Ratio of First Payment Default Performance
First payment default is when the borrower fails to make any payments and then subsequently defaults. This is a key indicator of fraud risk at the dealer level. It indicates probable misrepresentation either by the finance manager, salesperson or the borrower that resulted in a bad deal.
First payment default is a lagging indicator for dealer fraud since it can take up to 180 days to discover a dealer issue.
Across the 96,000 dealers analyzed, there was a wide spectrum of first payment default levels. Lower risk dealers tended to have default levels between 0.25% and 1%. Conversely, higher risk dealers tended to have default levels between 4% and 25% (with some outliers exceeding 75%). The higher the ratio of first payment default at a dealership, the more likely the dealer is engaging in risky activity.
Risk Factor #5: Poor Consumer Experience
Consumers write reviews or submit complaints on popular ratings sites when they have been ripped off, scammed, or had a poor experience with a dealer.
We analyzed ratings and reviews of a large set of dealers with high fraud rates and another set of dealers with low fraud rates. Dealers with high fraud rates had much lower consumer ratings and far fewer reviews than low-risk dealers.
Reviews of high-risk dealers often contain keywords such as “fraud,” “scam” or “rip off”. This indicates potentially unethical dealer activity from a consumer standpoint.
Risk Factor #6: Employer Recycling and Forged Paystubs Rates
In 2020, the use of fictitious and fake employers on borrower applications for auto loans increased by 300%. Why? Higher rates of unemployment and more stimulus money flowing through the economy fueled this dramatic surge.
Point Predictive’s fraud analysts quickly noted that certain dealerships tended to have very high rates of falsified employment and forged paystubs on their applications. “Employer recycling” is the fraudulent practice of using fake income and employment information across multiple applications. This is a key indicator of systematic dealership fraud or misrepresentation.
Risk Factor #7: Cross-Lender Fraud Experience
A dealer’s historical fraud rate at one lender is a good predictor of future performance at another dealer. Dealerships terminated by lenders for too many bad loans typically continue the pattern with another lender. We tracked many dealers with persistently high fraud default rates over several years. One particular dealer submitted several fraudulent applications to four separate lenders over the course of three years.
Franchise dealers are often more established and have more fraud controls than independent dealers. On average, independent dealers have twice the fraud risk of franchise dealers.
Dealership Fraud Signals Come from Multiple Data Sources and Approaches
The overwhelming majority of dealers are not associated with high rates of fraud. Lenders should look for subtle and obvious fraud signals that are occurring in different sources of data, including:
- Lender’s own historical application and performance data.
- Fraud consortium data that collects cross lender dealer activity and performance.
- Dealer application-level fraud score or fraud alerts that track transactional patterns.
- Consumer sentiment data from reviews and complaints.
- Negative news or actions regarding dealers in their network.
- Dealer websites and their current and historic inventory levels.
Fraud signals can be complex and difficult to decipher – and that’s where machine learning can help. By using multiple data sources and harnessing the power of machine learning, lenders can dramatically improve their dealer management and monitoring.
Learn more about Point Predictive’s proprietary approach to fraud prevention using Artificial + Natural Intelligence™.