Software Developer in London, United Kingdom

The project involved creating an algorithmic trading platform expressly designed to handle machine learning-based trading strategies. The platform incorporates multi-threading, multi-processing and asynchronous architecture, and is capable of running large numbers of independent strategies, and handling multiple data sources and trading venues simultaneously. The platform is accessible by front-end applications via a REST API (using Flask).

The project involved replication of existing vanilla derivatives pricing models with densely-connected neural networks (DNNs). The machine learning stack used was Scikit-Learn and Tensorflow, with hyperparameter optimization using Scikit-Optimize.

The project involved designing a single standard REST API specification to cover a range of market data sources, with varying requests and return formats: Bloomberg, Reuters Eikon, IBKR, Quandl, Oanda, and FXCM. Wrappers were built around the native APIs for these data sources, allowing significant removing complexity to be removed from downstream systems.

The objective of this project was to use unsupervised learning techniques (using Scikit-Learn) to allow identification and removal of both general market beta, and automated identification of equity clusters (groups of highly correlated stocks). This work was undertaken for an equity long-short hedge fund.

This project was undertaken for a swaptions trading desk at an investment bank.

The tool allowed a trading desk to identify the best maturity/expiries to trade to minimize P&L volatility, given an input vega grid, and taking into account historical correlations across the volatility surface.

This was created for a macro hedge fund. The SignalMonitor tool allowed realtime monitoring of market data levels (or any arbitrarily complex function of those levels) for specific thresholds e.g. for trade entry/exit signals. The tool generated email/desktop/text alerts and was integrated with a hashtag-driven knowledge base such that appropriate research would be automatically sent with the alert.

I was a quant trader on the new "Strategic Hedging" trading desk, whose purpose was to reduce bank-wide RWA/SVAR using vanilla options to hedge tail risk. I built the technology platform from the ground up, including designing the algorithms to solve for optimal RWA-reducing vanilla options hedges.

This project involved working with scenario generation results from multiple risk engines, delivered unreliably, and in an inconsistent format, to a risk database. The toolkit had to deal with missing or incorrect scenario results, the transformation of scenario results in a consistent format, and aggregation for display purposes. The cross-gamma coverage was 6 rate curves and SABR volatility surfaces, and 10 FX forward curves and SABR volatility surfaces.

Created an Excel VBA-based risk aggregation framework, used by six trading desks across the bank, which provided PV/risk/scenario aggregation or drill down to individual trade level. It also allowed design and sharing of standardised reports, and batch delivery of reports to risk management sheets.