Preface
The recent change in OTC trading practices has resulted in the demand for accurate credit risk fees to be made available near instantaneously for the traders. For the technology people in many institutes from Tier 1 banks to hedge funds, this is proving to be something of a black swan. At EM Risk we have a collection of experienced personal that can help in the whole spectrum of CVA, from vendor advisory through to developing in house implementation.
CVA by its very nature has created a unique collection of problems so esoteric even experienced personal will frequently not have seen the likes of before. At EM Risk we have been working closely with a number of vendors to help them better understand the requirements of the clients, and fully leverage advances in hardware technology.
To help understand the classic problems that are commonly faced by institutes of all sizes, we have outlined them below:
Market Data
In order to provide consistency between the price of the trade, and the price of the credit charge, it is required that the same front office market data be used. Whilst processes such as CVA might at first appear like traditional functions performed by the risk department, many institutes upon closer inspection require such functions to be owned by the front office space.
Model Integration
So that accurate CVA charges can be derived it is required that any CVA implementation has access to models identical to those of the front office pricing system. To try to replicate them exactly in a different architecture is an un-reliable, time-consuming, expensive act of code duplication, as such it is recommended to make sure that whatever system architecture is used for CVA calculation it is compatible with the front office models.
The constraint of the pricing models is of paramount importance, as any limitations such as threading model (ie COM Apartment) or CPU architecture (ie 32bit) can be in stark contrast with the best practice for a CVA calculation engine.
Calculation
Due to the sheer volume of calculations required for a Monte Carlo CVA evaluation, and the linear separation nature therein it is common practice to distribute calculations on a Farm or Grid Cluster. This should always be considered with the constraints posed by the existing Front Office models, whilst balancing the costs from vendors licensing their products per CPU Core combined with the hardware costs. It is therefore desirable to minimize the size and complexity of any such calculation engine, whilst allowing for future scalability. One cost effective solution can often be achieved by efficient use of optimized storage techniques, EM Risk are available for our proven advisory in this field, which has previously allowed our clients to cut their Grid expenditure by over 50%.
Result Aggregation & Display
One of the common issues presented during and after a CVA implementation scheme is the ability to aggregate the results, facilitate drill down diagnostics, justification of results and execute What If? Scenarios. Given the time constraints many CVA projects are under, these requirements are often prioritised with many neglected and a poor overall return on investment is achieved.
As such is it vital that there is sufficient understanding to allow for a simultaneous Bottom-Up development, this can dramatically reduce the time to market and provide a swifter and often larger return on the investment. An easy way to help provide the required separation is by having flexibility and extensibility in the output of the engine as well as choosing a presentation technology which is inherently elastic in nature. EM Risk can provide tools to augment Excel to allow for extremely rapid integration whilst allowing low wastage migration to rich GUI technologies like WPF.