Can anyone clarify credit card default definition given here?
\n\n\nThe target binary variable is calculated by observing 18 months performance window after the latest credit card statement, and if the customer does not pay due amount in 120 days after their latest statement date it is considered a default event.
\n
What does \"the latest\" statement mean? How often are such statements made? In my mind you can get such a statement at any point of time.
\nLet's say I've been approved for a credit card on 1st May 2022, but I'm not starting using it until 1st December 2023 when I decide to buy a laptop with the credit card. And 120 days later I can't pay due amount. Is it considered as a default? If so what are 18 months mentioned in the definition for?
\nSorry if my question is a bit stupid – I've never had any credit cards :)
", "messages": [ { "id": "1801561", "postDate": "05/25/2022 21:49:22", "content": "Can anyone clarify credit card default definition given here?
\n\n\nThe target binary variable is calculated by observing 18 months performance window after the latest credit card statement, and if the customer does not pay due amount in 120 days after their latest statement date it is considered a default event.
\n
What does \"the latest\" statement mean? How often are such statements made? In my mind you can get such a statement at any point of time.
\nLet's say I've been approved for a credit card on 1st May 2022, but I'm not starting using it until 1st December 2023 when I decide to buy a laptop with the credit card. And 120 days later I can't pay due amount. Is it considered as a default? If so what are 18 months mentioned in the definition for?
\nSorry if my question is a bit stupid – I've never had any credit cards :)
", "rawMarkdown": "Can anyone clarify credit card default definition given here?\n\n> The target binary variable is calculated by observing 18 months performance window after the latest credit card statement, and if the customer does not pay due amount in 120 days after their latest statement date it is considered a default event.\n\nWhat does \"the latest\" statement mean? How often are such statements made? In my mind you can get such a statement at any point of time.\n\nLet's say I've been approved for a credit card on 1st May 2022, but I'm not starting using it until 1st December 2023 when I decide to buy a laptop with the credit card. And 120 days later I can't pay due amount. Is it considered as a default? If so what are 18 months mentioned in the definition for?\n\nSorry if my question is a bit stupid – I've never had any credit cards :)", "votes": null }, { "id": "1802275", "postDate": "05/26/2022 15:57:17", "content": "I'm curious about this topic either🥲
", "rawMarkdown": "I'm curious about this topic either🥲", "votes": null }, { "id": "1802537", "postDate": "05/26/2022 20:41:09", "content": "The credit risk modelling domain usually works with the following way to calculate probability of default of customers:
\nThey set a performance period (18 months in this) for observations to use in modelling and estimates probability of default within certain period after the product is granted to the customer. So that you can standardize the period you follow credits performance since one customer may have credit card for 10 years and other for few months, which have different performance period to analyze and also different probability of default. This can tweak your models performance and may even create some bias.
For your case yes it should be considered as default. However, if you never used your credit card, probably you should be removed from the observations rather than kept as undefaulted since financial institution can not know whether you are a customer who can handle the risk amount or not.
Searching for this question in the Discussion for quite a while and I found @joyfulcorgi 's post very helpful.
\nThe short answer to understand this is that the default is calculated by whether the customer repays their bills in 120 days period after the latest credit card statement, and the 18 months window is just the time window we use to pull the historical data (which is the training data) to predict future credit default (i.e. The 18-month has nothing to do with the definition of the default)
\nBelow are some answers quoted from @joyfulcorgi 's post
\n\n", "rawMarkdown": "Searching for this question in the Discussion for quite a while and I found @joyfulcorgi 's [post](https://www.kaggle.com/competitions/amex-default-prediction/discussion/335158) very helpful. \n\nThe short answer to understand this is that the default is calculated by whether the customer repays their bills in 120 days period after the latest credit card statement, and the 18 months window is just the time window we use to pull the historical data (which is the training data) to predict future credit default (**i.e. The 18-month has nothing to do with the definition of the default**)\n\nBelow are some answers quoted from @joyfulcorgi 's [post](https://www.kaggle.com/competitions/amex-default-prediction/discussion/335158) \n> By Ali Abdin: I understand it the following:\n> 1. Each customer is observed for a 18 months window\n> 2. After each month, each customer gets a credit card statement and everyone has 120 days time to pay their due amount\n> 3. If a customer didnt pay within the following 120 days it is considered a default for that customer\n\n> By Denis Perracchio\n> You're looking for patterns/trends over time and across different dimensions (e.g. payment history, risk factors, spend, late payments, etc.). If you just had a snap shot of one period you wouldn't necessarily have enough information to predict if they'd default (I suppose if it was aggregated 18 month performance, that might help. Some Kaggle datasets have default info but it's an account level, not over time).\n\n> For the sake of illustration, let's say they were current for 12 months, then balances start going up, late payments, delinquency indicators start departing from \"norms\" (possibly observed from a credit report pull every month. ;-), Risk variables start trending up/spiking, you might think, yeah they're likely to default.", "votes": null } ], "comments": [ { "id": 1802275, "author_name": "sangwooooo", "author_url": "", "post_date": "05/26/2022 15:57:17", "content": "By Ali Abdin: I understand it the following:
\n\n
\n- Each customer is observed for a 18 months window
\n- After each month, each customer gets a credit card statement and everyone has 120 days time to pay their due amount
\n- If a customer didnt pay within the following 120 days it is considered a default for that customer
\nBy Denis Perracchio
\n
\n You're looking for patterns/trends over time and across different dimensions (e.g. payment history, risk factors, spend, late payments, etc.). If you just had a snap shot of one period you wouldn't necessarily have enough information to predict if they'd default (I suppose if it was aggregated 18 month performance, that might help. Some Kaggle datasets have default info but it's an account level, not over time).For the sake of illustration, let's say they were current for 12 months, then balances start going up, late payments, delinquency indicators start departing from \"norms\" (possibly observed from a credit report pull every month. ;-), Risk variables start trending up/spiking, you might think, yeah they're likely to default.
\n
I'm curious about this topic either🥲
", "votes": null, "replies": [] }, { "id": 1802537, "author_name": "efehandanisman", "author_url": "", "post_date": "05/26/2022 20:41:09", "content": "The credit risk modelling domain usually works with the following way to calculate probability of default of customers:
\nThey set a performance period (18 months in this) for observations to use in modelling and estimates probability of default within certain period after the product is granted to the customer. So that you can standardize the period you follow credits performance since one customer may have credit card for 10 years and other for few months, which have different performance period to analyze and also different probability of default. This can tweak your models performance and may even create some bias.
For your case yes it should be considered as default. However, if you never used your credit card, probably you should be removed from the observations rather than kept as undefaulted since financial institution can not know whether you are a customer who can handle the risk amount or not.
Searching for this question in the Discussion for quite a while and I found @joyfulcorgi 's post very helpful.
\nThe short answer to understand this is that the default is calculated by whether the customer repays their bills in 120 days period after the latest credit card statement, and the 18 months window is just the time window we use to pull the historical data (which is the training data) to predict future credit default (i.e. The 18-month has nothing to do with the definition of the default)
\nBelow are some answers quoted from @joyfulcorgi 's post
\n\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1801561": "Can anyone clarify credit card default definition given here?\n\n> The target binary variable is calculated by observing 18 months performance window after the latest credit card statement, and if the customer does not pay due amount in 120 days after their latest statement date it is considered a default event.\n\nWhat does \"the latest\" statement mean? How often are such statements made? In my mind you can get such a statement at any point of time.\n\nLet's say I've been approved for a credit card on 1st May 2022, but I'm not starting using it until 1st December 2023 when I decide to buy a laptop with the credit card. And 120 days later I can't pay due amount. Is it considered as a default? If so what are 18 months mentioned in the definition for?\n\nSorry if my question is a bit stupid – I've never had any credit cards :)", "1802275": "I'm curious about this topic either🥲", "1802537": "The credit risk modelling domain usually works with the following way to calculate probability of default of customers:\n\n- They set a performance period (18 months in this) for observations to use in modelling and estimates probability of default within certain period after the product is granted to the customer. So that you can standardize the period you follow credits performance since one customer may have credit card for 10 years and other for few months, which have different performance period to analyze and also different probability of default. This can tweak your models performance and may even create some bias.\n\n- For your case yes it should be considered as default. However, if you never used your credit card, probably you should be removed from the observations rather than kept as undefaulted since financial institution can not know whether you are a customer who can handle the risk amount or not.", "1893430": "Searching for this question in the Discussion for quite a while and I found @joyfulcorgi 's [post](https://www.kaggle.com/competitions/amex-default-prediction/discussion/335158) very helpful. \n\nThe short answer to understand this is that the default is calculated by whether the customer repays their bills in 120 days period after the latest credit card statement, and the 18 months window is just the time window we use to pull the historical data (which is the training data) to predict future credit default (**i.e. The 18-month has nothing to do with the definition of the default**)\n\nBelow are some answers quoted from @joyfulcorgi 's [post](https://www.kaggle.com/competitions/amex-default-prediction/discussion/335158) \n> By Ali Abdin: I understand it the following:\n> 1. Each customer is observed for a 18 months window\n> 2. After each month, each customer gets a credit card statement and everyone has 120 days time to pay their due amount\n> 3. If a customer didnt pay within the following 120 days it is considered a default for that customer\n\n> By Denis Perracchio\n> You're looking for patterns/trends over time and across different dimensions (e.g. payment history, risk factors, spend, late payments, etc.). If you just had a snap shot of one period you wouldn't necessarily have enough information to predict if they'd default (I suppose if it was aggregated 18 month performance, that might help. Some Kaggle datasets have default info but it's an account level, not over time).\n\n> For the sake of illustration, let's say they were current for 12 months, then balances start going up, late payments, delinquency indicators start departing from \"norms\" (possibly observed from a credit report pull every month. ;-), Risk variables start trending up/spiking, you might think, yeah they're likely to default." }, "source": "meta" }By Ali Abdin: I understand it the following:
\n\n
\n- Each customer is observed for a 18 months window
\n- After each month, each customer gets a credit card statement and everyone has 120 days time to pay their due amount
\n- If a customer didnt pay within the following 120 days it is considered a default for that customer
\nBy Denis Perracchio
\n
\n You're looking for patterns/trends over time and across different dimensions (e.g. payment history, risk factors, spend, late payments, etc.). If you just had a snap shot of one period you wouldn't necessarily have enough information to predict if they'd default (I suppose if it was aggregated 18 month performance, that might help. Some Kaggle datasets have default info but it's an account level, not over time).For the sake of illustration, let's say they were current for 12 months, then balances start going up, late payments, delinquency indicators start departing from \"norms\" (possibly observed from a credit report pull every month. ;-), Risk variables start trending up/spiking, you might think, yeah they're likely to default.
\n