Deep learning-based semi-automatic trading system for US stock market. The system automatically extracts a sensible information (in form of features for a deep learning model) from both publicly-available and subscription-based sources of daily and high-frequency market data. After a second stage of dimensionality reduction and features re-combination (via back-testing & cross-validation) the most relevant features are then used to train a complex non-linear model on GPU. The portfolio optimization-driven trading strategy uses probabilistic forecasts made by the model and current positions on the market to generate specific instructions for the traders.
ML-based semi-automatic market making and position trading system utilizing statistical arbitrage opportunities in volatility index – equity index future spreads on US market.
Broker-side stress-testing and optimization system for both aggregating and scaling multiple traders/strategies operating with options, futures, ETFs and stocks into a single portfolio for a better profitability/risk ratio for a broker.
Semi-automatic trading system built for US options market making and position trading using a relative value implied volatility modeling based on a statistical forecasting of co-movements of implied volatility surfaces.
Simulation platform for trading finance. It can simulate exchange functionality, exchange behavior, smart order routers, client behavior, market data sources and all interactions between them with a high degree of realism and consistency. A high-frequency exchange simulator uses a set of ML-based models to replicate a realistic behavior of the market. An integrated order-matching engine allows for tested trading strategies be surrounded by a realistic trading environment which can simulate various stress-testing scenarios (including exotic ones) while still keeping things realistic.
Exchange Simulator built for major European investment bank program trading desk which models reaction of market participants on user intrusions (activity that is absent in historical data files) by replaying historical orders flow, processing orders submitted by users according to exchange rules and modeling the market response on these user orders using probability-driven empirically justified models. The simulator uses a set of ML-based models to replicate a realistic behavior of the market. Having a wide range of behavioral settings allows to simulate various (including exotic) scenarios while still keeping things realistic due to an integrated order matching logic and fine-tuned ML models.
Infrastructure built for US proprietary trading firm for storing and analyzing tick data and order entry data in backtesting and real-time modes, providing an environment for defining and operating value-added metrics calculations in real-time and on historical time spans.
Energy StatArb Portfolio Management System. Semi-automatic trading system built for US energy hedge fund for generating trading recommendations on the market of energy/energy resources futures based on methods of Theory of Probability, Mathematical Statistics and Theory of Graphs, Time Series Analysis.
Model built for major European investment bank program trading desk provides an approach towards calculating optimal order execution schedule based on "implementation shortfall" (ISF) minimization.
Model built for major European investment bank program trading desk provides an approach towards making a quantitative estimates of influence of a market participant on market itself.