Technology

Data-light architecture

Many vendors still encourage a legacy approach to fixed income attribution, where user-supplied risk numbers (yield to maturity, modified duration, convexity) are used as a proxy for a full pricing model. Although the underlying algorithms for this approach are conceptually simple, the volumes of data required are typically expensive, time consuming and difficult to supply, especially for OTC securities and complex derivatives.

Flametree takes an alternative approach. Moore’s law has driven down the cost of computer power to negligible levels. By using explicit pricing models for all security types, we no longer require a supply of risk numbers. Instead, Flametree calculates everything from first principles, using only security master data and a yield curve. In addition, Flametree calculates returns, modified duration, convexity, YTM, OAS and many other quantities as a spin-off from attribution analysis.

This data-light approach requires orders of magnitude less information compared to conventional attribution analysis, without compromise on accuracy or coverage. In cases where our results differ from a perturbational analysis, Flametree's approach is actually more accurate, since the program makes no assumptions about cash flow distributions. However, Flametree can still use externally supplied risk numbers if required.

Flametree uses security-level weights and returns as its starting point for attribution analysis. If this data is not already available within your existing workflow, we can provide a returns calculation engine that is deployable in the same way as the core system.

Microservices architecture

Flametree is designed around the idea of microservices: modular, loosely coupled, independently deployable components with clear interfaces that follow industry standards. Each component follows the Unix philosophy of ‘do one thing, do it well’.

Flametree's components communicate via open, documented APIs. This allows our clients to:

- define and use their own pricing models in addition to those provided in our standard library
- define new types of portfolio risk
- call or embed Flametree’s performance and attribution engines seamlessly into other software.

For instance, if you have your own pricing code for specific derivative types, or a preferred third-party library, a simple translation layer allows it to be called from Flametree and used in your attribution reports (see OpenPricing). Our core performance and attribution engines can be also be called as services, using local or cloud-based hosting.

High-speed computing

Fixed income attribution is computationally intensive, since attribution needs every security in the portfolio and benchmark to be repriced many times at each date.

To address this requirement, Flametree offers:

High speed. Flametree has been written from scratch and has no legacy code restrictions. Flametree does not use a database and so is not restricted by hardware bottlenecks, in which large volumes of data are read from and written to disk during the calculation. The result is an attribution engine that is typically orders of magnitude faster than other systems.

Appropriate technology. Flametree is written in highly optimised C and C++ and uses multithreading to spread the computational load over all the processors available on your PC.

GPU (graphical processing) integration. For even faster processing, Flametree can offload most of its computational load onto external GPU processors, leading to massive speed increases. A GPU has thousands of processing cores running simultaneously. Although each core is slower than the CPU in your computer, in parallel they lead to massive speedups. FIA can use a $2000 graphics card to generate a typical speed increase of 30 to 50 times over conventional PC architectures.