Risk magazine recently ran an interesting article. To extract a few choice phrases:
- Before, analysts could be abstract, and mathematical ability was prized; now they have to be pragmatic, and computer programming is the essential attribute
- the marginal impact on the capital position of any proposed trade needs to be known in advance, which means a complete recalculation at portfolio level and a huge, ongoing drain on computing resources
- For practising quants, what’s more important, for example, is the ability to compute the impact of every trade on the capital position, and these kinds of problems will probably be solved by engineering means, rather than mathematical
To paraphrase, architecture matters. By architecture, I mean the overall design of a system informed by a deep understanding of the fundamental concepts which underpin the domain and the nature of the information that flows and interacts when problems are solved in the domain.
Without a well-chosen architecture, you cannot manage the complexity of today's financial world. A portfolio with a range of collateral agreements requires a large collection of carefully constructed curves. Exposure-based calculations such as CVA on such a portfolio require a modelling approach that treats a vast number of risk factors consistently. An organisation faces considerable operational risk without coherent management of its valuations. In short, you can't solve large-scale problems without architecture and today's problems are large in scale.
Over the next few weeks, I will explore this topic with a series of posts on the fundamental concepts and design ideas that must underpin the architecture of a modern analytics platform if it is to scale to an enterprise level.