Estimate lost income per mortgage condition

[Theherald]

Good morning

I work in Compliance for a mortgage company, and have no real background in statistics. My company closes loans and sells them to other investors. Between the time that we close a loan and an investor buys a loan there are issues with the loans (called conditions) that have to be resolved before investors will buy the loans.

Each day that passes without the investor buying a loan there is a cost (usually substantial) that occurs, calculated as Expected Revenue (minus) Actual Income. This cost represents the loss of income recieved for the loan. We can’t sell the loan until all of the conditions are cleared, so conditions can be tracked based on the number of days they take to clear (see note at the end of email about the relationship between number of days for conditions and loss of income).

For example, let’s say there are 20 loans closed (a Sample) in a month with Conditions A, B, C. Condition A (A) appears on 10 loans for a total of 100 days, an avg of 10 days per loan. In the end I want to find a way to estimate which of the conditions are the most costly, so we can focus training and a portion of bonuses on the performance of the most costly conditions.

I have been trying to calculate several things: 1) find a way to assign an estimated value to each condition even though there may be several conditions per each loan, each lasting a variable number of days before a final condition is satisfied (I’ve attempted applying a weighted metric derived from percentage of loss from total loss of the whole sample but this isn’t quite doing it) 2) if #1 is too complex then just the best way to Estimated Loss per loan by # of days per each condition. Thank you for any assistance you can provide.

A note: while in most cases the # of days that it takes to satisfy all of the conditions, or the final condition, will determine the Loss of Income for each loan, there are circumstances where no investor will buy a loan and we have to sell it at a greatly reduced price that skews our numbers.

I appreciate any direction you can provide.

Best regards

Jim Lyons

 

[staassis]

1. Build a non-linear model where

_ Cost of Loan is the dependent variable,
_ Condition Set, other contractual provisions, other characteristics of the reference entity(ies) and macro-environment are the predictors.

The modeling methods depend on home many loans you have in the data set.

2. Run sensitivity analysis of Cost of Loan on Condition Set using the estimated model. Try two risk management frameworks:

_ 2.1 Changing presence of condition A does not change presence/absence of conditions B and C.
_ 2.2 Changing presence of condition A changes the probability of B and/or C being present, according to the joint distribution implied by the data set.

 

[Theherald]

Thank you. I have no experience with statistics. Could you provide me a very “dumbed-down” version of

_ 2.1 Changing presence of condition A does not change presence/absence of conditions B and C

 

[staassis]

The statement above is equivalent to saying: the decision to include A in a contract is independent of the decisions to include B or C in the contract. There are no relationships among A, B and C in the sense:

- if A is included, B (or C) is redundant,
- A and B (or C) work well / inefficiently together.

In summary, environment 2.1 is idealized. Most likely, environment 2.2 is closer to reality.

 

[giovannynolan]

Good morning

I work in Compliance for a mortgage company, and have no real background in statistics. My company closes loans and sells them to other investors. Between the time that we close a loan and an investor buys a loan there are issues with the loans (called conditions) that have to be resolved before investors will buy the loans.

Each day that passes without the investor buying a loan there is a cost (usually substantial) that occurs, calculated as Expected Revenue (minus) Actual Income. This cost represents the loss of income recieved for the loan. We can’t sell the loan until all of the conditions are cleared, so conditions can be tracked based on the number of days they take to clear (see note at the end of email about the relationship between number of days for conditions and loss of income).

For example, let’s say there are 20 loans closed (a Sample) in a month with Conditions A, B, C. Condition A (A) appears on 10 loans for a total of 100 days, an avg of 10 days per loan. In the end I want to find a way to estimate which of the conditions are the most costly, so we can focus training and a portion of bonuses on the performance of the most costly conditions.

I have been trying to calculate several things: 1) find a way to assign an estimated value to each condition even though there may be several conditions per each loan, each lasting a variable number of days before a final condition is satisfied (I’ve attempted applying a weighted metric derived from percentage of loss from total loss of the whole sample but this isn’t quite doing it) 2) if #1 is too complex then just the best way to visit https://fitmymoney.com/best-personal-loans/ per loan by # of days per each condition. Thank you for any assistance you can provide.

A note: while in most cases the # of days that it takes to satisfy all of the conditions, or the final condition, will determine the Loss of Income for each loan, there are circumstances where no investor will buy a loan and we have to sell it at a greatly reduced price that skews our numbers.

I appreciate any direction you can provide.

Best regards

Jim Lyons

Sounds to me like something confusing and complicated.