A Trading Algorithm: theory and implementation of an expansion from a single asset to a portfolio

This work explores the attempt of building a mathematical model aiming to describe and predict a very complex phenomenon: price action in financial markets. Trading has always been a fundamental aspect in the macroeconomic and financial setting, and trying to predict future directions of several assets has been a hot topic for centuries. What makes this framework so complex is the unquantifiable amount of variables affecting the markets at every instant, large part of which are completely unmeasurable. The simple price action curve we see everyday is a result of macroeconomic movements, news on world politics and economics and many other factors. However, the most important influence are investors feelings. This can only give a hint about how complex the “generating processes” for asset prices are. Well aware that there are no secret codes to success, particularly in this field, we tried to implement an automatic algorithm that could try to generate educated guesses in this peculiar environment. Our investigation made use of several tools, involving coding, machine learning, great amount of logic and mathematical reasoning, and most importantly the experience of long-time investors. The main object of this project was initially to study and implement an optimal exit strategy from a financial position, that is what we called a “trade”. This has been done mainly thanks to the study of several ad hoc indicators and measures, coupled with mathematical and logical approaches that were financially sound. At a second stage, when an exit strategy was established, the research proceeded with an expansion from a single asset to many. Indeed as a benchmark for the trading pipeline a single financial index was exploited: the DAX30. As the diversification of investments is a key concept in finance, the interest moved to the exploration of more assets. In order to do so, nonparametric classification algorithms were used to find other assets that could potentially work in our set up. To top up our investigation as not all the generated trades were satisfactory enough, we performed a deep cleaning of the positions entered over time. The main ingredients here have been experience in trading and reading topic-related graphs like candlesticks, coupled with logical, statistical and mathematical reasoning. This research is a result of an intership project developed at XSOR capital Limited, an alternative investment boutique that has been working on this automated process for the past few years.