Algorithmic trading strategies for multi-asset classes

PhD Thesis Proposal Defence


Title: "Algorithmic trading strategies for multi-asset classes"

by

Mr. Feng HAN


Abstract:

In the sales and trading domain, most execution orders are placed by 
computer algorithms. Traders, quants and portfolio managers collaborate 
together to execute a predefined portfolio, seeking to optimize return 
while minimizing risk exposure and market impact. Stocks and futures 
exchanges leverage matching engines to record every order in limit order 
books including trade price and volume, bid or ask orders on each side, 
and levels of the electronic book. We contribute to algorithmic trading 
strategies for multi-asset classes for traditional assets including 
equities, futures contracts, and emerging assets like cryptocurrencies.

The first contribution is algorithms trading strategies for Hong Kong 
stocks investment strategies. We generate trading signals at the minute 
level, implement Stop and Reverse algorithms and MACD agents & strats 
modules to generate buy and sell signals directly.  We backtest 391 stocks 
with returns of around 7% from November 2020 to January 2022. This 
algorithm is scalable to other financial instruments like futures and 
forex.

The second contribution is statistical arbitrage across different futures 
exchanges.  We explore tick-level precious metal futures contracts data to 
analyze Shanghai Futures Exchange, Chicago Mercantile Exchange, and Japan 
Exchange Group. Our results show the futures are price strongly correlated 
and we backtest the strategies and report the observations.

Apart from traditional asset classes, we explore trading emerging assets 
including cryptocurrency spot and perpetual futures contracts. The third 
contribution is cryptocurrency trading with the implementation of 
technical indicators including multi-strategies e.g. shorting. We backtest 
2019-2021 multiple periods. Here we compare 3 assets: equities, futures, 
and cryptocurrency from volatility, liquidity, and transaction cost.

To better understand the trading conditions of the above-mentioned assets, 
we need a data generator to simulate markets. The fourth contribution is 
creating a simulator to generate orders based on WassersteinGAN and 
compare parameters with actual trades recorded from stock markets. We 
input the 1-minute trade price data and compare the generated real and 
fake data using Kolmogorov-Smirnov distance, MSE, MAE scores to quantify 
our model performance. The main goals are to improve transaction 
performance, reduce impact costs, ensure execution efficiency, protect 
transaction intent, capture transaction opportunities, and obtain 
short-term Alpha returns in the whole transaction cycle.


Date:			Tuesday, 9 November 2021

Time:                  	2:30pm - 4:30pm

Zoom Meeting: 		https://hkust.zoom.us/j/2871974170

Committee Members:	Prof. Jiheng Zhang (Supervisor)
 			Dr. Xiaojuan Ma (Supervisor)
  			Prof. Lei Chen (Chairperson)
 			Prof. Qiong Luo


**** ALL are Welcome ****