FinCad, in a recent white paper, maintains that many companies are leaving cost-effectiveness gains on the table. They would be well-advised to pick up these gains by outsourcing their portfolio level risk analysis.
There are a variety of distinct analyses that turn on different issues at the portfolio (as distinct from the trade) level. There is, for example, the issue of aggregated cash flows, which simply requires the sorting of the cash flows in a chronological order. Although this is important, it is fairly straightforward.
On a somewhat more sophisticated plain, there is stress testing, the revaluation of a portfolio “under a number of different scenarios for the market data to quantify the exposure to more extreme market movements.” There are related analyses directed at counterparty credit risk, such as credit value adjustment or potential future exposure.
System Components
So: if you’re a fund manager and you want to create an in-house system for such analyses, what do you need? An adequate system for portfolio analysis requires many components: data repositories, user interface software, servers, grids, and an analytics library.
The analytics library is the heart of the matter. This will have to: offer powerful portfolio specification; provide the analysis; decouple the trades from the models; re-use objects; meet IT criteria. FinCad has a point more to say about each of these five requisites:
- Portfolio specification requires that the analytics have access to everything that is happening at the trade level, with templates for vanilla trades but without restriction on the complexity of the trades the system can handle;
- The analytical ability should, for example, be such as will calculate the sensitivity of assets to the five-year swap rate and tol do the mathematics involved in pricing continuously, that is by calculus, rather than by the discrete levels involved in “bumping”;
- The decoupling of trades from models is necessary so that model risk can be assessed at will. Perhaps your risk manager is concerned that valuation of stock options in the portfolio should not be wedded to Black-Scholes. He can substitute a different model (Monte Carlo, most likely) and get a quantitative fix on the difference that makes;
- The available models with their calibrated parameters and market data can be cached and reused for various different portfolios or requests on an appropriate system. Finally;
- The analytics library has to fit seamlessly into the rest of the production environment.
FinCad quotes a Celent report to the effect that firms trying to set up this architecture in-house for derivatives analytics will have to make an upfront investment of at minimum $9 million, and then anywhere between 25 and 50 percent of that initial investment each year to keep the pricing and risk data relevant.
Celent also considers inefficiencies due to “drag,” to discrepancies among silos where applicable, and the integration costs between internal and external systems. They estimate that the total cost of derivatives analytics over a five-year product cycle can exceed $45 million.
FinCad concurs with Celent and contends that “it makes more economic sense for this type of development effort to benefit many systems developers, not just one.”