Acadia’s Open-Source Risk Engine (ORE) - How its expanded functionality provides a real choice for firms

What is ORE?

Acadia has successfully supported the non-cleared derivatives market through regulatory compliance of all six phases of the global Uncleared Margin Rules and emerged as a key industry utility for initial margin calculation, reconciliation, and exchange. In order to accomplish this feat for the world’s 1,000+ largest derivative trading institutions, itd eveloped a standardized pricing and risk framework for calculating risk sensitivities (e.g. “greeks” formatted to ISDA’s CRIF standard), backed by an industry-leading centralized market data service, covering the wide range of vanilla to exotic financial derivative products traded OTC and supporting underlying reference data. At the heart of Acadia’s hosted Risk Services is the Open-Source Risk Engine (ORE).

‍
ORE has a long history of open source support going back to its authorship by Quaternion Risk Management in the early 2010s (Quaternion was acquired by Acadia in 2021), and has historically maintained a delineation between the commercial version of the software (ORE+), licensed by several clients for local production-level pricing and risk calculations, and the free open source version (expanded product coverage and simulation methodologies built on another open source project, QuantLib).
‍

With quarterly open source releases running from 3Q22 through 4Q23, Acadia will be releasing the vast majority of ORE’s commercial functionality into the open source domain for the very first time; from exotic pricing model coverage across all asset classes, par market risk sensitivities and several flavors of Value-at-Risk (including ISDA SIMM™), a credit exposure simulation framework (including xVAs) supported by extensions for American Monte Carlo, cross-asset modelling, and multi-threading, to standard regulatory capital metrics like SA-CCR, FRTB-SA, and BA-CVA, and a novel Scripted Trade framework supported by Adjoint Algorithmic Differentiation (AAD) – this is a highly robust sell-side pricing and risk engine, rivaling the capabilities of other large software vendors, all freely available in the open source domain under a modified
BSD license.

ORE - a robust model validation tool

Acadia knows that ORE provides equivalent pricing capabilities to larger software vendors because 150+ clients have independently validated Acadia’s hosted Risk Services output (e.g. mark-to-markets, risk sensitivities, historical P&L calculations feeding VaR) in the context of their regulatory compliance for UMR. Other clients have similarly validated ORE’s performance for use cases outside of UMR, e.g. credit exposure simulation and xVAs. Model performance benchmarking to each clients’ own local proprietary models and/or software vendors is a critical step in the model validation lifecycle, and Acadia has never once failed a model validation or been declined regulatory approval. We continue to make enhancements to our hosted modelling infrastructure in light of client feedback on best practices and new market developments (e.g. LIBOR transition to Alternative Reference Rates), and these community-driven/industry-led best practices all make their way back into subsequent open source ORE releases. As a result, ORE users are assured that the models and methodologies contained therein align with commonly accepted industry best practices.

Regulators have viewed our models favorably because of ORE’s transparency down to the source code level. ORE’s models and underlying pricing/risk methodologies have passed validation muster by numerous banks and non-bank swap dealers subject to strict regulatory model risk management requirements under e.g. Federal Reserve’s SR 11-07 and Prudential Regulation Authority’s CP6-22/PS6-23 model risk management principles for banks. By its very open-source nature, it’s not a “black box” like other vendor systems and can be extensible and customized to precise client requirements, allowing ORE users to demonstrate strong ownership of model calibrations and configurations during regulatory reviews. Acadia can provide “out of the box” configurations to support global reference data coverage across 80+ currencies, 250+ interest and inflation rates, 400+ commodities, 8,000+ single-name and index bond and credit default references, and 30,000+ single-name and index equity references for both linear and option products.

Deployment – out of the box

Acadia has listened to client feedback on accessibility restrictions for those users less familiar with ORE’s C++ source code and included new Python integration in the recent 1Q23 release. This allows ORE to be readily installed as a self-contained Python package (e.g. “pip install open-source-risk-engine”) and run via simple Jupyter Notebooks, making it accessible to a wider community beyond C++ developers – this functionality was particularly well-received by the 120+ people in attendance at Acadia’s London Quant Summit in April 2023. For production-level implementation, ORE can be containerized via Docker for integration with REST APIs, which is how Acadia deploys ORE within its IT architecture. Flexible local deployment options range from command-line usage on Windows/PC, Linux, and Mac, to Java and Python programs via SWIG (e.g. Jupyter Notebooks), to HTTP requests into local cloud-based data and analytics services; all have been successfully utilized by various Acadia clients over the last several years.

ORE Use Cases

Acadia has observed three common client use-cases for ORE:

1) as a benchmarking tool to facilitate and prototype propriety internal models at large institutions, informing primary model development without “recreating the wheel” for basic model components like discount curve construction and/or risk factor evolution models in exposure simulations

2) as an independent model utilized by local Model Validation teams at large institutions, saving enormous time in preventing one-time/”offline” model creation and/or replication

3) as the primary pricing and risk model for smaller-to-medium-sized institutions lacking either the budget for larger-scale vendor software and/or large development teams to support proprietary internal model development.

ORE supports multiple pricing methodologies and calibration assumptions for individual instruments, and its output can be automatically linked straight through to either Model Owner and/or Validation documentation required by a firm’s internal model governance framework. Model input/output benchmarking depends on a pipeline of source data transformations, manual results manipulation, and manual table/chart generation and related updates (e.g. within Word and/or LaTex). To bypass enormous manual effort, one-off helper tools can be created (e.g. in Python) and linked automatically to local ORE output, which can then be re-run/re-used regularly on-demand based on a firm’s internal model lifecycle requirements. By organizing and centralizing a benchmarking framework around the industry-standard models in ORE, a large majority of the Model Documentation and validation pipeline can be fully automated. Not only is the ORE software completely free, but reductions in the associated validation pipelines and related headcounts can lead to further cost savings.

Expanding the functionality

The 3Q23 ORE release, contains an innovative Scripted Trade framework inspired by leading-edge risk quants like Jesper Andreasen and Antoine Savine. It is equivalent to next generation, large-scale sell-side pricing engines evolved from early iterations in the late 1990s (e.g. Emmanuel Derman at Goldman Sachs, SynTech at Gen Re, GRFN at BNP Paribas, OptIt at Societe Generale) to more advanced recent implementations at the likes of Bank of America, Nordea, Danske Bank, and Saxo Bank. Other large banks like J.P. Morgan, UBS, Commerzbank, and Citi have all explored similar scripted trade implementations to solve for increasing cost and regulatory pressures on model performance, documentation, and validation. ORE’s Scripted Trade framework allows users to author a custom script that describes the payoff of even the most  complex financial instrument, and value the trade under a common Monte Carlo simulation framework. The model can be parameterized under a multivariate Black Scholes or Gaussian process, and contains support for high-performance computing enhancements via parallelization, multi-threading, and AAD. For many firms, the alternative valuation methodology provided by the Scripted Trade framework allows for a relevant independent benchmark in the validation lifecycle, while for other firms it may allow for business extensions into the exotics space with lighter development and validation requirements for a single model framework, as opposed to the development and support of multiple exotic pricing models.

‍

Please visit Acadia’s ORE website (www.opensourcerisk.org) to stay up-to-date on the upcoming release schedule, and consider registering your interest to join Acadia’s next Quant Summit in New York City on November 1st to hear more about client use-cases and their first-hand experiences using ORE.

About the Author:

Scott Sobolewski is a risk and finance professional specializing in capital planning, stress testing, derivatives pricing, and model development at large US banks. He advises financial institutions on risk management and regulatory compliance matters, helping clients accelerate development timelines and achieve high-value institutional objectives. Scott has extensive experience meeting Dodd Frank deliverables and managing regulatory relationships with the Federal Reserve, OCC, FDIC, and others through prior roles at Citigroup and Santander US. He holds a B.A. in Mathematics and Economics with Honors from Williams College, as well as active Chartered Financial Analyst (CFA) and Certificate in Quantitative Finance (CQF) qualifications.