独家优惠奖金 100% 高达 1 BTC + 180 免费旋转

What makes a credit card holder DEFAULT?

An image by BankBazaar Blog

The Taiwanese Financial Institution, or TFI, is the provider of credit cards. It has noticed an increase in client defaults. This situation is having a significant negative impact on the company’s revenue, but they are aware that they can take action if they can foresee which credit card customers would miss their subsequent payment. However, the management of TFI has emphasised that understanding why consumers are defaulting is as crucial to them as predicting which customers would do so. Hence, the purpose behind this project includes answer to following questions:

Figure 1: Probability of defaults by gender

We can see from the plot that the default rate in males is marginally ahead of its counterpart by less than 4%. However, the dataset comprises 20% more females with roughly 12k males and a little over 18k females. Let’s dig a bit deeper to see what age group in both genders defaults the most.

Here you can see the plot is right-skewed, meaning the distribution follows a decreasing trend as the age of customers increases. The majority of payers & non-payers in both genders lie in the age group of approximately 20–35. This suggests that the bank issued credit cards to adults more often than the elderly since 75% of the customers in the dataset have aged less than 41 years.

In the chart above, only a difference of 2.54% observed between married and single customers in default. A further analysis is required to understand whether the marriage attribute relates to default payment.

Figure 3: Probability of default in Education Level.

It is very tempting to infer from the plot that ‘Others’ category has the least default rate of all groups. However, the attribute has values that were listed as unknowns (probably filled in for missing values) and some values undocumented in the data dictionary. Hence those categories were grouped as ‘Others’. Nonetheless, there are insignificant differences found between High School and University graduates. Unsurprisingly, grad school clients are fewer defaulters since graduates from professional schools are more likely to get a high salary jobs that might make them pay their dues.

2. What age group do people with high lines of credit belong to?

To answer this question we need to define what high line of credit is. I assume credit limit greater than 75th percentile is what it means. So, let’s try to see the credit limit distribution of payers vs non-payers.

The density plot illustrates people with high credit limit have significantly lower default percentage. Intuitively, this makes sense since a bank must have verified applicants history before issuing such high credit.

Figure 5:

The bar chart clearly shows the high credit holders belong to the age group of 31–40 (around 46%), while the age groups 41–50 and 21–30 stands at second and third place respectively. Moreover, the default proportion follows the same trend as the non default proportion in decreasing order with age group 31–40 at top. We can try to find what proportion of this age group (31–40) have completed grad school because the default ratio is lower for grad school graduates and only after careful consideration by the bank, it issues an applicant a high credit limit. Based on the analysis, I found more than half the population (of age group 31–40) have concluded grad school. Around 36% have finished University education while the remaining categories stand at roughly 8%.

3. Which variables are the strongest predictors of default payment?

Looking at the heatmap of correlation matrix, we can see there aren’t any features highly correlated with the target feature. However the next month’s default prediction is dependent on the repayment status of the last six months, particularly the repayment status of recent month. The multicollinearity between repayment status features is also apparent. Further analysis is required to identify features that are strongest predictors of default payment.

In this article, we looked at interesting findings by asking questions from the dataset.

Add a comment

Related posts:

What freelancing has taught me about money

Going freelance will teach you a lot, not just certain skillsets that you accumulate when you jump into the cold water, but also valuable learnings about yourself — and finances. In this article, I’m…

Why Paid Commercial Web Accessibility Plugins Are Better Than Free Plugins

Building a website these days is one thing, building an accessible and usable website is another thing entirely as it requires some expertise and use of accessibility tools. These days, you can…