Of herding 1 million hissing cats onto a carousel somewhere a few blocks north of Bryant Park in New York. It must be said though that the music on this particular carousel had stopped (and Edith Piaf had most definitely left the building).

As part of the unwind of Lehman Brothers’ derivatives portfolio for the post-bankruptcy Estate (a portfolio of over 1 million derivatives trades); the team conducted a classification exercise of the products in the portfolio (with underlying covering all major asset classes; and instruments running the gamut of complexity from vanilla single factor single asset class flow products to highly exotic structured multi-factor hybrid products).

The context and objective was valuing them in the shortest possible time, in the most efficient manner possible (given limited to no infrastructure), in the most defensible manner possible (given their eventual day in Bankruptcy court) over the days in September/October 2008 when they were unwound.

This classification exercise was essential to developing and driving the valuation strategy for the portfolio, covering:

• Team selection;
• Valuation platform, model and method selection;
• Computational resource provisioning;
• Market data requirements definition;
• Developing the netting and hedging assumptions needed to take a view on reasonable bid/offer and transaction costs;
• And conducting self-consistency checks of the valuations.

A complex and gargantuan valuation exercise that could only be accomplished by intelligent abstraction of product commonality through classification.

It is always easy to find fault with a classification. There are a hundred ways of arranging any set of objects, and something may almost always be said against the best, and in favour of the worst of them. But the merits of a classification depend on the purposes to which it is instrumental.

John Stuart Mill
Auguste Comte and Positivism

Classification as used here attempts to arrange traded financial derivatives into product classes or groups based on similar or related properties; properties as identified within a defined scheme of taxonomies; and similarity of properties as meaningful within some context.

The motivation for classification here is not much different to classification in biology in that the focus is not so much on the naming of things but on coming up with the best possible ordering of our knowledge base about the properties of the objects being classified such that the ordering gives us the greatest contextual command of the knowledge already acquired about the objects, and also leads us in the most direct way to the acquisition of more.

In plain English and as an example, within the context of classification for risk based P&L attribution policy as an example, we want to think of how to order the properties of financial derivative contracts in such a way that we can group them around the types of risk sensitive behavior they are likely to exhibit and thus how their P&L behavior may be best explained. Additionally, a fundamentally intuitive grouping helps shed light on more risk-sensitive properties that may be applicable within groups.

(more…)