While stocks tend to have relatively high liquidity and price transparency, there are a multitude of financial products that are illiquid and price-opaque. A boutique investment bank was interested in development of a machine learning-based pricing algorithm for retail structured notes, a highly illiquid class of fixed income securities. Retail structured notes represent a roughly $2.5 trillion global market comprised of hundreds of thousands of individual securities. Due to the wide array of heterogonous structures, complex variable rate coupon formulas and cross-asset class linkages, sparseness of market-wide transactional data on an individual security basis, and dearth of consistent secondary market pricing, determining contemporaneous pricing of these products is normally done by experienced traders one security at a time. Our client wanted to explore automation solutions for this process.
Working extensively with our client’s trading team, Mazzaroth combined financial markets time series data across multiple asset classes with transaction data spanning several years, a time frame which covered more than one market regime. We then developed and tested parametric and non-parametric models, and ultimately selected an ensemble that was capable of providing accurate and consistent pricing across an entire class of structured notes. The robustness and accuracy of the system were tested over a multi-month period by the trading desk prior to deployment.