Lessons from 7 Years of Algorithmic Trading Research and Development

I have been on a journey since mid 2016 to learn how to trade algorithmically which is a data-driven method of using a set of rules to define buy/sell decisions on financial instruments. This is also referred to as quantitative and systems trading. Please note that I do not come from a background of finance, trading, math, or statistics but I do have an insatiable drive to learn and a whole lot of “never give up”. I could write volumes on all my experience and failed attempts at creating trading systems over the past 7 years but will spare you details. This journey began when I discovered a simplistic online tool that helped users apply rules to financial data to run a backtest and continued with purchasing some relatively pricey but much more powerful software that ran on my local PC to further develop and test trading system ideas. My mission at this time was to nail down a method of identifying individual Technical Analysis trading signals that had 1) predictive power and 2) a very high likelihood of “working” on unseen “Out of Sample” data. I spent about 3 years doing analysis of individual signals where I would analyze one trading signal, its “In Sample” metrics, statistical properties, noise sensitivity, and value on shifted data to classify it as “likely to work” or “not likely to work”. What I found is that no matter how much analysis I did on a given signal, the random nature of the market and regime changes still played havoc on their forward performance. A given signal may be quite “good” but can still experience large drawdown, periods of sideways non-performance, or stop working soon after going live. <a href="https://medium.com/@josh.malizzi/lessons-from-7-years-of-algorithmic-trading-research-and-development-c63f1d319831">Website</a>