How do i Forecasting Balance, charge off, and unit decay over time?

[Kojimasan]

Hi all,

I have this scenario, where i have 22 months of data of balances, chargeoffs, and units open over month of book/time. If i were to add 800 more accounts to my portfolio, what would the forecasted balance, chargeoffs, and units decay over time look like?

I appreciate all the help.

MOB	OpenSum	Total_CurrentLimits	TotalBalances	Total_NetRevenue	TotalCO
0	8555	$14,204,580	$724,597	$18,741	$0
1	8083	$13,231,680	$2,044,468	$47,335	$0
2	7641	$12,750,080	$2,404,794	$44,057	$0
3	7170	$12,097,780	$2,494,425	$40,122	$0
4	6739	$11,603,780	$2,425,099	$37,327	$0
5	6345	$11,072,580	$2,362,642	$36,457	$0
6	6038	$10,658,880	$2,304,736	$32,787	$2,056
7	5703	$10,251,480	$2,238,227	$31,514	$18,960
8	5238	$9,312,180	$2,141,285	$28,905	$13,549
9	4852	$8,785,900	$2,064,912	$26,378	$13,538
10	4497	$8,422,600	$1,958,612	$24,109	$9,433
11	4157	$8,108,800	$1,904,706	$22,480	$13,278
12	3862	$7,709,800	$1,814,142	$22,245	$5,844
13	3527	$7,327,900	$1,700,450	$33,129	$12,425
14	3216	$6,913,400	$1,595,645	$30,890	-$3,671
15	2958	$6,420,500	$1,497,002	$28,749	$4,226
16	2732	$5,991,600	$1,401,854	$26,669	$9,857
17	2475	$5,587,950	$1,294,763	$24,121	$5,523
18	2249	$5,219,850	$1,189,160	$21,408	$8,021
19	1966	$4,740,700	$1,032,205	$19,220	$1,854
20	1601	$3,758,900	$841,423	$15,005	$739
21	1304	$3,056,400	$663,451	$12,444	$1,557
22	1004	$2,378,250	$498,282	$9,151	$2,810
23	724	$1,727,500	$363,072	$6,625	$2,638
24	521	$1,328,000	$254,276	$4,463	$1,017
25	254	$615,100	$123,321	$1,987	.
26	55	$148,300	$31,027	$544	.

[PaigeMiller]

Well, this is a very broad question, and the answers really depend on what assumptions you make about those other 800 accounts, and what you mean by "forecasting".

You need to probably sit down with some subject matter experts at your institution to get some direction on how to do this.

[Kojimasan]

The accounts are of the same class, so let's say the original data set are only accounts in Vermont. The 800 accounts would be additional accounts only from Vermont as well we could've added but we didn't because they were rejected an account based on credit bureau data. 

And by forecasting, it could probably just be back-testing ?

[PaigeMiller]

Well, if they were rejected based on credit bureau data, you still have to make some assumptions about how these loans will perform. I can't do that for you. 

No, backtesting is meaningless here until you have a model, which you don't have. Forecasting here is meaningless as well, you don't have a model.

[Kojimasan]

What do i need in terms of performance data to develop a model?

[PaigeMiller]

Well, this is out of the realm of SAS at this point. You need to talk to someone who has expertise regarding this problem, and then make (hopefully) realistic assumptions about how these new 800 loans will perform.

[Kojimasan]

Turns out no one around me as implemented a model or any sort of project like this before at my job. It's a new portfolio. I have all the charge off data.

[PaigeMiller]

---

@Kojimasan wrote:

Turns out no one around me as implemented a model or any sort of project like this before at my job. It's a new portfolio. I have all the charge off data.

---

But this doesn't change my answer. Someone (not me) has to make assumptions about how the 800 loans will perform.

[Ksharp]

Maybe you could try Mixed Model or GEE Model.

Check 

For Mixed Model:

PROC MIXED  

PROC GLIMMIX

For GEE Model:

PROC GEE

PROC GENMOD