Our Analysis

We worked with Trans Union LLC (TransUnion®) access and analyze 1,000,000 anonymous credit reports and VantageScore® 3.0 credit scores to get an understanding of what credit usage is like in the US and how that credit usage impacts customer credit scores. We have looked at this data in a number of new and different ways to share with consumers what credit is like and help them better understand how they can improve their credit. The articles and tools on this site combine the learnings from this analysis with other publicly available information about credit and our conversations with credit bureaus and credit scoring companies.

Our Data

The data we have is based on depersonalized credit files as of August 2019. For security and data privacy reasons, we did not receive any consumer personal information or full consumer credit reports. Our data files consisted of about 200 summary variables that Transunion created based on consumer credit reports. That means we see data points from consumer credit reports like the following: 1 mortgage with a balance of $127,345; 3 credit cards with a total balance of $2,456; 1 account that was 30 days past due in the last 90 days; etc. More importantly, what we do not have are any consumer details, including no names, no address information, no age information, no social security numbers, and no names of individual creditors.

This data is a random sample of US consumers who have a credit file with Transunion and a VantageScore® 3.0 credit score. We do not have data on people who do not participate in the credit system. We choose to use 1,000,000 random credit reports to ensure that the file was a representative sample of the US population who have a VantageScore® 3.0 credit score.

Score 40 Methodology

In order to determine the estimated impact of credit behavior on the VantageScore® 3.0 credit score, we used a differential analysis rather than creating a model to estimate the credit score. What this means is that we analyzed both the detriments and benefits to a credit score that would arise from different credit behavior. This is because we are trying to understand the relative impact of consumer action on the credit score, not to determine an actual credit score. All of our efforts were therefore focused on plausible customer actions. For example, we would try to understand the impact of reducing credit utilization by 10%, because consumers can pay down their debts and it would be useful to know the impact on VantageScore® 3.0 of that action. We did not determine the impact of removing a bankruptcy from a credit report as this is not something a consumer can usually do.

We began our efforts with open knowledge. That is, VantageScore Solutions states that payment history, credit utilization, credit mix, age of credit, and new accounts (this is assumed to include inquiries) are the key influencers of the credit score. Taking this into account, we created variables to summarize the credit score influencers into simple but important key variables.

  • Months since most recent 30 day or worse past due status
  • Total dollar amount of third-party collections
  • Number of credit accounts
  • Number of credit cards
  • Utilization rate of credit cards
  • Time since an account has been charged off
  • Time since last inquiry

In each of these groups, classifications were created (rather than using exact amounts). For example, collections was broken down into “No collections / $0,” “Less than $100,” “$100 to $500,” and “Over $500.”

We then use these in different combinations to paint a picture of how credit scores change based on the overall factors. In this there is a balance to strike. Add too many factors and you have too little data. Not enough, and you cannot know what is impacting the changes. Therefore, this is done on multiple levels with summary statistics. Below is a small example table with discussion of what we learned from it. Note that this was done with thousands of rows of data to complete our program.

In this example, we have 3 variables accounted for.

  • Number of open accounts, in this case it’s always 3 or more
  • Open credit card, yes or no
  • Total amount of 3rd party collections, all options shown

While other factors are sure to play a role in these differences, for example, a higher percentage of people with collections are also currently past due than people without collections, we can see some patterns in these statistics. This is why we look at the distribution of credit scores rather than an average. In every case you can see a steady improvement in credit score as the amount of collections is lowered to the next tier, regardless of whether someone has an open credit card.

Another point to see is that those without an open credit card have lower scores than their counterparts with a credit card. However, we cannot say this is a direct result of the existence of a credit card account. It’s possible that the difference is an indirect result of how credit cards are used differently than other forms of credit.

This includes:

  • Credit cards typically have the longest credit age aside from mortgages
  • Credit cards are more likely to have low utilization compared to loans
  • Credit cards provide an additional item in the credit mix variable
  • And possibly, people are more likely to have a credit card because they have good credit (and this type of product is available to them) rather than the credit card causing a good credit score

By reviewing these at multiple levels, each across three to four variables at once, we are able to determine estimated, relative impacts of the actions that consumers may take to improve their credit. These estimates are based on the action and the current content of the credit report.

Background on VantageScore® 3.0

It is important to note that mathematically, VantageScore® is modeling the likelihood of a customer default within the next two years. That is, based on what is known today, what is the relative chance, compared to other consumers, that an individual consumer will fail to repay a debt over the course of the next two years? Statistics cannot predict an individual’s future, but rather it gives an expectation of how likely something is within a group of similar individuals, i.e., will 5% or 50% of similar individuals default? We use this understanding in our analysis as well.

Access a PDF of this information here.

Published March 30, 2021