Mackenna Dsouza

A CME Natural Gas American Options (ON) market maker required an American Call Option (Bjerksund Stensland 2002 approximation) model for live pricing and greeks. Risk managers, mid-office, and back-office personnel were required to use the same pricing model for P&L reporting, risk analytics, and settlements.

The trader used a Windows desktop-based dynamically linked library ( Written in C# ) loaded by an MS Excel VBA function accessible from an Excel workbook. The risk managers implemented their own version of the pricing model for risk reporting and analytics. Mid-office personnel used a model provided by the ETRM vendor. Back office personnel used an implementation within their derivative accounting software to settle these options. A quant team introduced recommended standardized validated models which needed to be rolled out the all the stakeholders. Frequent updates as a result of research, audit requirements, and methodology changes needed to be instantaneously deployed to all the stakeholders across four global trading locations ( Houston, New York, London, Singapore ).

For every change in the model, the quant team, supported by the companies IT department, had to deliver

• a C++-based derivative pricing model library as a Unix shared object (.so) for server-based applications compiled and deployed on a Sybase platform.
• a Windows statically linked library, packaging the same C++ code to be integrated into the Windows-based ETRM application.
• A .xll incorporating the C++ code into an Excel extension for use by the traders on their desktops
• And a dynamically linked library to be used by the derivative accounting software used by the back office.

Roll out of an enhanced pricing algorithm to price tail risk better took three months to be deployed globally. In the interim period, market makers, risk managers and quants were using the new model. Mid office was reporting off a non validated model provided by the ETRM vendor. Back office was generating Accounts Payable and Accounts Receivable records in the derivative accounting system using a homegrown system that was three model releases behind as a system upgrade was the only way to incorporate the new model when there was a change.

This led to recurring reconciliation efforts between traders, mid and back office personnel at the end of every accounting period . It also led to invoicing disputes with trading counterparties external to the firm.

The desired solution required use of a single globally deployed pricing model, updated regularly, yet accessible seamlessly to the trader desktop, ETRM, Server based pricing platform, accounting software, auditors and any future system with minimal integration efforts and time to market.

These objectives were met using a low code, highly available, planet scale, containerized, kubernetes deployed, functions as a service approach, that allowed the quant and IT teams to roll out models using a Jenkins based CI/CD process several times a day. Models were deployed as stateless functions called using simple http get requests accessible from software clients such as spreadsheets, javascript , python, Java, C++ , C# and Golang programs. Several research quant groups could leverage these models from tools like R , SAS and Matlab.

This enabled the quant team to continuously deploy updates. Due the cohesive yet decoupled nature of http request based integration, standardized model rollout and trading compliance policy enforcement became a non-issue.

Pricing discrepancy reconciliation efforts were almost non existent. Auditors were assured that every function within the trading firm was using the same validated models globally. The quants were assured that they could continuously improve derivative pricing infrastructure to capture and price risks and roll out those changes globally within minutes.

The FAAS based model deployment reduced the artifacts needed for a roll out from 7 complex efforts in conjunction with IT, to a unobtrusive automated CI/CD based deployment using a GIT workflow derived devops process.

Focus could now be on building better pricing models and agile deployment of model updates with no significant software engineering effort.