My current research focus on modeling market microstructure model using traditional and deep neural point process. The goal is to build a probabilistic generative model that takes variable-length sequences in some time interval and outputs attention-weighted representations. These latent representations will then be aggregated to summarize market impact functions.

  • An introductory slide is made to give an intense briefing about Variational AutoEncoding, an important method that is used in our project

  • Recent TPP and NeuralTPP research topics are reviewed and an extensive Temporal Point Process framework is implemented based on this paper. I also made a slide to give you a sense of general framework in modeling irregularly spaced event-associated time series data.

AutoEncoding Kernel Intensity

My most recent internship is to collaborate with Nodejs engineers to develop arbitrage strategies in algorithmic and high-frequency trading.

  • Develop reinforcement learning algorithms to solve stock trading problems that allow for easy configuration, parallel training and experiment tracking

  • Advance team understandings of cutting-edge research and white papers in Decentralized Finance and Web3 Project

  • Use data-driven approach with Geometric Brownian Motion model to find near-optimal call-put strangle that hedges against impermanent losses

Python MLflow Web3 Yaml