MetaChat - More fascinating excerpts.

MetaChat is an informal place for MeFites to touch base and post, discuss and chatter about topics that may not belong on MetaFilter. Questions? Check the FAQ. Please note: This is important.

18 July 2007

More fascinating excerpts. →[More:]
From a document called Basel II Clustering Algorithms for Portfolio Pool Aggregation:

"Risk Segment Clustering Specifics

The portfolio risk segments, delinquency, MOB, and, where applicable, score and LTV, constitute tiers (clustering dimensions) for portfolio pool aggregation (clustering). Clustering by delinquency band provides segmentation not only for all bands less than the default, but also for each band separately. Score band clustering provides segmentation for delinquency bands 0, 1 and 2. Risk segment clustering provides segmentation at the portfolio level. Multiple report-period back tests generate the data used for clustering.

The multi-dimensional clustering procedure follows these main steps:

1. Preprocessing

Selection of tiers (clustering dimensions)
Definition of tier sequence (hierarchical structure)

2. Clustering process

Determination of cluster numbers
Assignment of stability checks and cluster cutoffs

Portfolio pool aggregation by risk segment clustering does not permit breaking the sequence of aggregated pools, so standard SAS clustering procedures are inappropriate. Standard SAS clustering also requires special procedures for determining both dimensions and hierarchical structures and uses specific segmentation criteria unsuitable for risk segment clustering."

Hey, I wrote that! And another ten pages of increasingly complicated nonsense!

By the way, mullacc: on Friday I mentioned I worked on a Basel II score consolidator for a $30 billion portfolio(by "billion" I mean 10⁹; I gather there are many places where "billion" means something else entirely). That seemed small by a factor of ten, and it was: I meant $300 billion.

It's all about "lift," my friends.

post by: Hugh Janus at: 10:59 | 12 comments

*blank stare*

posted by jokeefe 18 July | 13:05

You think that was bad, check out this load of bull:

"Concurrent consideration of the AR and stability metrics for clustering permits optimization of group count and cutoff calculation. AR maximization can sometimes lead to isolation of high-noise data pools in separate clusters; the stability check prevents this process and allows acquisition of stable cutoffs only. Formal concurrent maximization of these two metrics requires heuristically defined metrics weights combined with arbitrarily selected function structures. To eliminate this uncertainty, the clustering algorithm uses the AR maximization procedure only and verifies the achievement of a certain level of stability (max P-Value < 0.05).

"The optimization algorithm is nonlinear: it contains multiple local maximums and seeks a global maximum. Local maximums are stabilized AR maximums for each considered group number. The final cutoff set corresponds to the AR maximum of all local AR maximums (see Appendix 1.2).

"Thresholds play a significant role in the clustering segmentation process. Algorithms use three main thresholds: AR “min” for tier determination, AR “lift” for tier group number, and hierarchical tier structure determination and P-Value “max” for stability check. Procedures to identify thresholds and assesses their influence on the optimization process are currently under analysis."

posted by Hugh Janus 18 July | 13:27

Okay. I'll see you that and raise you this:

The holophrastic is situated at the initial point of the recognition of the self as separate being— Lacan’s famous mirror stage— and the newly invented/discovered process of creating meaning through utterance. Kristeva references the holophrase in her discussion of the semiotic and the early acquisition of language: “The child’s first so-called holophrastic enunciations include gesture, the object and vocal emission. Because they are not yet sentences, (NP-VP), generative grammar is not readily equipped to account for them. Nevertheless, they are already thetic in the sense that they separate an object from the subject, and attribute to it a semiotic fragment, which thereby becomes the signifier.” (98) Holophrastic utterances then do carry meaning and signification, of a type that is both fragmentary and whole, in the sense that the fragment here functions as synecdoche. A holophrastic utterance can be seen as the minimal gesture of meaning, performing a fundamental recognition of the self and other, as when a child generalizes the sound of an object to naming the object (Kristeva provides an example here: “all animals become ‘woof-woof’’; otherwise, all trucks can become ‘beep-beep’, and so on).

posted by jokeefe 18 July | 14:08

Eeew, gross!

I'm referring to the use of the word "references" as a verb, which disgusts me as much as does "utilize." Then again, I used the word "heuristically" in mine, so I have little room to talk.

posted by Hugh Janus 18 July | 14:23

*snerk*

There's more where that came from, too.

posted by jokeefe 18 July | 15:07

Ditto.

posted by Hugh Janus 18 July | 15:22

*smacks down*

Bring it, biatch.

posted by jokeefe 18 July | 15:42

"Modeling result charts can be found in Appendix 2. These preliminary modeling results illustrate general approaches to clustering algorithms and can be helpful in understanding and evaluating them. They are not final modeling results.

"For portfolio XXXX004, the clustering algorithm creates five score groups for delinquency 0, four score groups for delinquency 1, and three score groups for delinquency 2. Two MOB groups are created for delinquency 3. There is no multi-tier clustering for this portfolio. MOB groups are not created for any score groups. Stability PD of score groups over time (report periods) is shown on corresponding graphs for each delinquency separately. A similar result was obtained for portfolio XXXX008. The differences in results reflect volatility in PD score groups over time, specifically for delinquency 0. As the corresponding graph demonstrates, highly sensitive cutoff determinations remain stable through time.

"For portfolio XXXX001, the clustering algorithm performs a multi-tier clustering procedure. Two corresponding graphs show the dynamic of AR change when MOB groups are added to each score group. MOB groups are not created for score group 3 based on the AR “lift” evaluation. The graphs for portfolio XXXX001 (delinquency 0) demonstrate the level of PD MOB group stability over time for each score group."

posted by Hugh Janus 18 July | 15:48

Oooooof! Argh!

*staggers slightly, but recovers, breathing heavily*

posted by jokeefe 18 July | 16:20

*parries*

"This is what literary modernism itself viewed, at least in part, as its project— the expression of contemporary interiority, the subjectivity of temporal experience and the conditions of urbanity, and the conscious search for a “new” language with which to represent these conditions. Pound’s famous dictum (“make it new”) and his critical writings from the first decade of the twentieth century, along with the various manifestoes of the Vorticists, the Dadaists, and so on are artifacts of modernism’s self-consciousness: as Sara Danius writes, modernism, as “it wanted to understand itself” was as “autonomous aesthetic performances at a safe remove from the encroachments of technology, massification, and commodification” (26) The question of modernism’s engagement with mass culture—critically obscured by modernism’s own insistence on its position on the high culture side of the ‘split’ between the avant garde and the popular, between high art and low—is a compelling one, and one which casts a light on Paris: A Poem, a title which in itself transforms the living city to a text, where the whole of the urban is translated into a single aesthetic phrase."

posted by jokeefe 18 July | 16:26

The new Harry Potter book now has the status IN PROCESS. We have already found an item level Hold on a copy (which we have transferred to a bib Hold). This must have been done by a staff member because a patron cannot place an item level hold on this item through the online catalog. Please be sure that everyone in your branch knows that they are not to put an item level Hold on this book (or any item that is IN PROCESS unless it is a multi-volume set). Either [one of my coworkers] or I will be checking the record everyday for awhile to transfer item Holds to bib Holds, but you can help cut down on the problem by making sure that your staff knows how to properly place Holds for patrons.

posted by box 18 July | 17:50

Hugh, you should translate this into I CAN HAS CHEEZBURGER-ese and then submit that as the report.

posted by jason's_planet 18 July | 18:05

← RING!!!!! || Anyone CAN Cook. →

HOME || REGISTER || LOGIN