@Anonymous wrote:Hello TT. Thanks for the update on your ideas about scorecard assignment (SA). I know you had previously thought that AAoA was a factor in SA, based on Liz Weston's book I think. Appreciate you letting us know that you have changed your mind.
In both cases, however, it sounds like you thought there were exactly three factors for SA if the person had a clean profile. Is it still your guess that the three factors are binary -- since 2 x 2 x 2 = 8 (and there are 8 clean scorecards)?
If so, do you have any reasonable guesswork about where the binary switches are for each factor? E.g. if by "credit depth" we mean number of accounts, what's a reasonable guess as to where FICO defines the switch between a thin and non-thin profile? (E.g 1-2 = thin, 3 or more = non-thin.) Similarly (assuming that each of the three factors is indeed binary) where might the switch be for the other two factors?
Of course, it's possible that the fact that 2-cubed = 8 may be a numerological coincidence in the this case. I am curious whether you know of any evidence beyond that piece of math aesthetics to confirm that SA is done with three binary switches.
CGID - If "Non derogatory"" card colors are an indication, the below indicates only one "binary split". I'm pretty sure a 30 day late would be a mild delinquency and 90 day would be a major delinquency that would move a profile into a derogatory classification. Not sure about 60 day lates.
Liz wasn't really specific in defining the age component. My interpretation of what she said was wrong. As I recall you brought up the point that she might have been referring to oldest account as opposed to AAoA. Anyway, here is a screen print direct from a Fico presentation. Not sure how mild delinquency factors in. Also, not sure if "Enhanced Core System" refers to standard Fico 8 or some enhanced model. (I'll add the link if I can find it). Webinar document title is - Fico 8 Score: Get The Benefits Now!
Here's a summary of VS3 scorecards from a VS white paper.
@Thomas_Thumb wrote:
@Anonymous wrote:.
Also illuminating to see that a person with only mild delinquencies is considered part of the clean scorecards. My guess is that a profile with delinquencies is considered mild if they are only 30-day lates (no 60-day) and there are no recent lates (e.g. none in the last X number of months, perhaps 24). That's a guess, but it strikes me as reasonable. I agree that the chart is vague there. It seems as though maybe you have to be guilty of something as serious as a public record to be a non-mild derog, but I find it hard to believe that 60 or 90 day lates are considered mild.
CGID -
Some of the scorecard boundries may involve some fuzzy logic conditional statements particularly with regard to quantity and recentness of lates. Perhaps, one or two 30 day lates won't necessarily keep one on a "dirty" scorecard. As you say, an aged single 30 day late might be discounted enough to trigger re-asignment to one of the eight clean scorecards.
Note: Another mild delinquency could be collections under $100. I believe paid, under $100 collections, are not used in Fico 8 scoring but may influence scorecard assignment.
There are some articles/publications that discuss scorecard factor values by means of hypothetical profile examples. Some info mentioned includes:
1) File history: 3 years = young, 10 years = established [I use 10 years as a basis for a high threshold]
2) File thickness: 5 accounts = thin, 10 accounts = not thin, 15 to 20 accounts = mainstream. [I'm not sure on this but I went with 8 accounts or more as not thin]
3) youngest account = ? [I went with 1 year on this but it could be 2 years] Also even if 1 year, time could be conditioned to require more time (2 years) depending on QTY of "new" accounts.
Side nore: Scorecard assignment may be influenced by credit behavior during the most recent 12 to 24 month performance period relative to a prior period. Fico states it does look at performance period behavior. However, potential influence on scorecard assignment is pure speculation on my part.
The below link takes you to a table I have been working on in an effort to categorize scorecards. It's a constant work in progress.
http://ficoforums.myfico.com/t5/Understanding-FICO-Scoring/SCORECARDS/m-p/4361047#M102264
Tell ya in another year and a half re 30 day late comparisons 
FICO 8 ignores <$100 collections period, paid vs unpaid doesn't make any difference.
Not sure I agree with your assumptions on breakdowns but as this is all speculation none of my own rationale is really any more insightful (other than to suggest lenders viewed <4 tradelines and now <3 tradelines when we are talking modern mortgage market as thin).
Also I haven't ever gotten a mysterious bump when tradelines turned one year, zero zip nada even when it was the last one with nothing more recent so, yeah. If there is a boundary it's probably at 2 years based on my data anyway... not convinced there's any seasoning period in modern models but again it's a purely speculative assertion at this point. To note though, I'm in the dirty buckets and there are a lot more clean buckets in terms of scorecard divisions if you are of the opinion it's just bucketing rather than sorting within a scorecard (which is probably the same between every file).
@Revelate wrote:
@Thomas_Thumb wrote:
@Anonymous wrote:.
Also illuminating to see that a person with only mild delinquencies is considered part of the clean scorecards. My guess is that a profile with delinquencies is considered mild if they are only 30-day lates (no 60-day) and there are no recent lates (e.g. none in the last X number of months, perhaps 24). That's a guess, but it strikes me as reasonable. I agree that the chart is vague there. It seems as though maybe you have to be guilty of something as serious as a public record to be a non-mild derog, but I find it hard to believe that 60 or 90 day lates are considered mild.
CGID -
Some of the scorecard boundries may involve some fuzzy logic conditional statements particularly with regard to quantity and recentness of lates. Perhaps, one or two 30 day lates won't necessarily keep one on a "dirty" scorecard. As you say, an aged single 30 day late might be discounted enough to trigger re-asignment to one of the eight clean scorecards.
Note: Another mild delinquency could be collections under $100. I believe paid, under $100 collections, are not used in Fico 8 scoring but may influence scorecard assignment.
There are some articles/publications that discuss scorecard factor values by means of hypothetical profile examples. Some info mentioned includes:
1) File history: 3 years = young, 10 years = established [I use 10 years as a basis for a high threshold]
2) File thickness: 5 accounts = thin, 10 accounts = not thin, 15 to 20 accounts = mainstream. [I'm not sure on this but I went with 8 accounts or more as not thin]
3) youngest account = ? [I went with 1 year on this but it could be 2 years] Also even if 1 year, time could be conditioned to require more time (2 years) depending on QTY of "new" accounts.
Side nore: Scorecard assignment may be influenced by credit behavior during the most recent 12 to 24 month performance period relative to a prior period. Fico states it does look at performance period behavior. However, potential influence on scorecard assignment is pure speculation on my part.
The below link takes you to a table I have been working on in an effort to categorize scorecards. It's a constant work in progress.
http://ficoforums.myfico.com/t5/Understanding-FICO-Scoring/SCORECARDS/m-p/4361047#M102264
Tell ya in another year and a half re 30 day late comparisons
FICO 8 ignores <$100 collections period, paid vs unpaid doesn't make any difference.
Not sure I agree with your assumptions on breakdowns but as this is all speculation none of my own rationale is really any more insightful (other than to suggest lenders viewed <4 tradelines and now <3 tradelines when we are talking modern mortgage market as thin).
Also I haven't ever gotten a mysterious bump when tradelines turned one year, zero zip nada even when it was the last one with nothing more recent so, yeah. If there is a boundary it's probably at 2 years based on my data anyway... not convinced there's any seasoning period in modern models but again it's a purely speculative assertion at this point. To note though, I'm in the dirty buckets and there are a lot more clean buckets in terms of scorecard divisions if you are of the opinion it's just bucketing rather than sorting within a scorecard (which is probably the same between every file).
Revelate, as you mention, you have a dirty file and land in one of the 4 derogatory scorecards. The factors for the 4 dirty scorecards are different - those you can test. We know one of the factors is public record (yes/no) and another is worst rating. I have speculated recency and/or credit behavior in the last 24 month performance period may also come into play.
Unfortunately, you are unable to test scorecard factors associated with clean files. Multiple sources list the factors associated with the 8 clean scorecards. By example, 5 files is thin and 3 years for oldest is young. Also 10 years for oldest is not young. For clean scorecards I strongly suspect reassignments may often not cause a step change in score - that's one reason why the boundries are elusive.
[For example, if weight on # cards reporting or weight on individual card UT% changes, you may not see a score shift if you typically report balances on 2 oo 9 cards and UT on any card never exceeds 30%. However, if you were reporting on 5 of 9 or 2 of 9 with a couple over 50% then perhaps a score shift when re-assigned]
Edit add: Thickness of file may be a strict count thing for scorecard assignment. On the other hand, credit mix is a scoring factor.
@Thomas_Thumb wrote:
@Revelate wrote:
@Thomas_Thumb wrote:
@Anonymous wrote:.
Also illuminating to see that a person with only mild delinquencies is considered part of the clean scorecards. My guess is that a profile with delinquencies is considered mild if they are only 30-day lates (no 60-day) and there are no recent lates (e.g. none in the last X number of months, perhaps 24). That's a guess, but it strikes me as reasonable. I agree that the chart is vague there. It seems as though maybe you have to be guilty of something as serious as a public record to be a non-mild derog, but I find it hard to believe that 60 or 90 day lates are considered mild.
CGID -
Some of the scorecard boundries may involve some fuzzy logic conditional statements particularly with regard to quantity and recentness of lates. Perhaps, one or two 30 day lates won't necessarily keep one on a "dirty" scorecard. As you say, an aged single 30 day late might be discounted enough to trigger re-asignment to one of the eight clean scorecards.
Note: Another mild delinquency could be collections under $100. I believe paid, under $100 collections, are not used in Fico 8 scoring but may influence scorecard assignment.
There are some articles/publications that discuss scorecard factor values by means of hypothetical profile examples. Some info mentioned includes:
1) File history: 3 years = young, 10 years = established [I use 10 years as a basis for a high threshold]
2) File thickness: 5 accounts = thin, 10 accounts = not thin, 15 to 20 accounts = mainstream. [I'm not sure on this but I went with 8 accounts or more as not thin]
3) youngest account = ? [I went with 1 year on this but it could be 2 years] Also even if 1 year, time could be conditioned to require more time (2 years) depending on QTY of "new" accounts.
Side nore: Scorecard assignment may be influenced by credit behavior during the most recent 12 to 24 month performance period relative to a prior period. Fico states it does look at performance period behavior. However, potential influence on scorecard assignment is pure speculation on my part.
The below link takes you to a table I have been working on in an effort to categorize scorecards. It's a constant work in progress.
http://ficoforums.myfico.com/t5/Understanding-FICO-Scoring/SCORECARDS/m-p/4361047#M102264
Tell ya in another year and a half re 30 day late comparisons
FICO 8 ignores <$100 collections period, paid vs unpaid doesn't make any difference.
Not sure I agree with your assumptions on breakdowns but as this is all speculation none of my own rationale is really any more insightful (other than to suggest lenders viewed <4 tradelines and now <3 tradelines when we are talking modern mortgage market as thin).
Also I haven't ever gotten a mysterious bump when tradelines turned one year, zero zip nada even when it was the last one with nothing more recent so, yeah. If there is a boundary it's probably at 2 years based on my data anyway... not convinced there's any seasoning period in modern models but again it's a purely speculative assertion at this point. To note though, I'm in the dirty buckets and there are a lot more clean buckets in terms of scorecard divisions if you are of the opinion it's just bucketing rather than sorting within a scorecard (which is probably the same between every file).
Revelate, as you mention, you have a dirty file and land in one of the 4 derogatory scorecards. The factors for the 4 dirty scorecards are different - those you can test. We know one of the factors is public record (yes/no) and another is worst rating. I have speculated recency and/or credit behavior in the last 24 month performance period may also come into play.
Unfortunately, you are unable to test scorecard factors associated with clean files. Multiple sources list the factors associated with the 8 clean scorecards. By example, 5 files is thin and 3 years for oldest is young. Also 10 years for oldest is not young. For clean scorecards I strongly suspect reassignments may often not cause a step change in score - that's one reason why the boundries are elusive.
[For example, if weight on # cards reporting or weight on individual card UT% changes, you may not see a score shift if you typically report balances on 2 oo 9 cards and UT on any card never exceeds 30%. However, if you were reporting on 5 of 9 or 2 of 9 with a couple over 50% then perhaps a score shift when re-assigned]
Edit add: Thickness of file may be a strict count thing for scorecard assignment. On the other hand, credit mix is a scoring factor.
I would not have pointed out my limitations if I weren't awre of them .
Which sources? (and please don't say Liz Weston 
)
Factors shown in above Fico slide. Link provided above the slide.
I think you and Liz should meet. Perhaps my-own-fico can arrange it. 