Today, Paul Bilokon (Thalesians Ltd) interviewed Claudio Albanese, a leading expert on quantitative and computational finance and Founder of Global Valuation. Claudio is also Faculty at Machine Learning Institute (MLI).
PB:In 2006, you transitioned from academia to launch Global Valuation. What was the initial business plan?
CA: The central idea was that GPU computing would revolutionize Computational Finance and Mathematical Finance. I believed the industry needed to be fundamentally rebuilt with matrix multiplication as the central performance bottleneck. At that time, the significance of this wasn’t as clear-cut as it is now. Today, we have seen Machine Learning reshape Classical Statistics, anchored heavily on matrix multiplication, a principle that is true for Large Language Models and other AI algorithms as well.
So, I embarked on reformulating the Financial Mathematics framework, drawing parallels with Quantum Mechanics, and casting Stochastic Calculus into an operator formalism. It turned out that the mathematical work was the easy part. A bigger challenge was on the software engineering side, particularly broadening the scope to encapsulate all aspects of risk-neutral valuation theory.
PB:Your product, Esther, has evolved since its first release. Can you walk us through the journey and the initial use cases it addressed?
CA: The first iteration of Esther was developed during my time on a CVA project at Credit Suisse. The software was later adopted and marketed by TriOptima. In 2013, we introduced it as a central XVA service, collaborating with them and the CME for half a decade. Following TriOptima’s acquisition by the CME, Esther powered OSTTRA’s portfolio simulation engine, providing services like XVA metrics and optimization strategies for capital and collateral.
This deployment validated that our mathematical framework could be transformed into robust, high-performance software. We were at the forefront, pioneering next-generation XVA metrics like funding-set FVA, KVA based on nested Monte Carlo simulations, and VaR calculators with Wrong-Way-Risk adjustments.
However, the monolithic software architecture made it cumbersome for developers who had to juggle between low-level programming and high-level business logic, hindering the technology’s industrial scalability.
PB:What enhancements does the second release bring?
CA: We’ve entirely overhauled Esther for its second release, separating the mathematical and concurrency logic from the business layer.
It now functions as a compiler for risk and pricing problems, providing a universal solver. Users can define problems by providing semantically complete specification using a protocol we developed called STEM (smart trade Esther message), STEMs are serialized and processed by the Esther solver, with results communicated back to the compiler via callback calls. The user can program in just about any language, as the only requirement is support for Google protobuf serialization and callbacks.
PB: A universal solver for risk and pricing is quite groundbreaking. What specific use cases does it target?
CA: Initially, Esther aims to tackle Wrong-Way-Risk management for Non-Bank Financial Institutions—a regulatory priority in light of significant losses faced by firms like Credit Suisse and the London Metal Exchange as a consequence of the blow-ups of Archegos and Tsingshan.
Dealing with information asymmetry is a key challenge here. We propose a “zero-trust” solution wherein counterparties can execute risk analytics on their portfolios privately and share results at their discretion. Esther is uniquely equipped to support this client-side paradigm, covering all angles of Counterparty Credit Risk analytics. What is required is that the dealer send a STEM message to the counterparty and the counterparty then decides whether or not to run the analytics on their aggregate portfolio, and ultimately whether or not to share the results with the dealer.
PB:Can you expand on other applications of client-side analytics?
This applies to various domains, including capital and collateral optimization strategies among banks and client-facing product structuring applications, all while maintaining data privacy.
PB:Considering server-side applications, how does Esther stack up? What are its best-suited use cases?
There is a significant cost advantage when switching to Esther, from hardware to energy savings. The simplicity of programming Esther models lends itself well to automation by Large Language Models. Plus, its seamless integration with Machine Learning on AI-enabled devices and its reliance on matrix algebras makes it a compelling choice across any risk neutral valuation application.