Sun 24-Mar-2019

In my work and studies I consume – voraciously if not somewhat compulsively – vast (and I mean vast) amounts of content: academic papers, online articles, talks, events, and meetups. While it can get intense and a bit FOMO, I consider myself really fortunate to be exposed to so much truly brilliant, cutting-edge thinking by people so much smarter than I am.

I try to record, save, bookmark and Evernote quality content, but I thought it would be a good idea to also share some of it with my readers. To that effect, I am launching a series titled #articlesworthreading, in which I will present articles which made an impact on me.

In the inaugural post, I will present a 2003 article by Dr. Monica Lam from California State University titled “Neural Network Techniques for Financial Performance Prediction: Integrating Fundamental and Technical Analysis”*.

Dr. Lam’s article is worth reading for a number of reasons.

Firstly, there is the subject matter. Dr. Lam took on a very ambitious task of training her AI (to be more exact: artificial neural networks, ANN’s) to predict which stocks in her selected universe will be in the top third (33%) of performers (defined as % return). There are hundreds of articles in which authors attempt to train their AI’s to predict the performance of entire markets (proxied by their equity indices), but there are far fewer experiments in which AI is actively constructing some sort of bespoke portfolio.

Secondly, it is one of a few articles which isn’t purely focused on AI / neural network architecture and performance. Many similar articles treat market data just like any other dataset and approach their research purely as experiments in pattern recognition and prediction. These articles tend to focus on the technical angle, such as architecture of the neural networks, algorithms, training functions etc. – they are much computer science- and data science-focused. They are of somewhat limited value to investment professionals considering adoption of AI in the investment decision process.

Thirdly, the research was conducted at the time when anything to do with AI was decidedly *not* en vogue. It wasn’t any of the officially recognised “AI winters”, but – after short-lived mini-hype of neural networks in the late 1990s, and, more importantly, shortly after dot-com bubble burst – there wasn’t much interest nor excitement around the topic. In the first decade of the 21st century the tech world and the (Western) world, in general, were focused on exponentially growing Web, e-commerce, online banking, MP3’s, mobile connectivity and social media. So Dr. Lam was definitely going against the tide.

Last but not least, Dr. Lam performed a thorough investment performance analysis – this is what made her article really stand out for me. She compared % returns of AI-selected top performers against actual top 33% of equities in the sample in different neural network configurations. In some of her experiments, ANN’s were working off historical time series only (making it essentially a technical analysis-based approach), in some, she enriched the time series input with several fundamental inputs.

While AI-selected top performers underperformed actual top performers, they *outperformed* the investment universe as a whole. Moreover, this outperformance was consistent and statistically significant. So AI did manage to beat the market – and that is an exciting and promising conclusion for future research (and we need to remember that the article is over 15 years old).

You can find the article on Google Scholar. Please note that it may be behind a paywall.

*full citation: Lam, Monica. “Neural Network Techniques for Financial Performance Prediction: Integrating Fundamental and Technical Analysis.” Decision Support Systems 37.4 (2004): 567-81. Web.