So what can NLP do for financial firms?

There’s more than one meaning for NLP, which you might encounter. If you Google it, one of these is Neuro Linguistic Programming, and techniques of communicating with other folks. Well this post is absolutely nothing to do that! Instead, it’s to do with the NLP, which occupies more of my time, natural language processing. In a nutshell, it’s about trying to do the types of tasks that humans undertake when dealing with a “real” language, with a computer.


Sometimes this can involve writing, ie. natural language generation. This could be related to tasks such as creating chatbots. For a financial firm, these chatbots could be programmed to interact with clients, such as on a market making desk, whether that’s answering questions about markets or executing orders. In practice, for investors, natural language generation is probably less important than it would be for the sell side. Instead, for those seeking to generate “alpha”, natural language understanding is more important, in other words making sense of text.


The idea of using text to trade markets isn’t really new. News for one has always been a big mover of markets, and I doubt many would disagree with that. The difficulty is that there is so much text being generated that could impact markets, is that it is impossible for a human to read it all. With machines to help us read the news, we can read text from many sources, be it from newswires, social media and the web more broadly.


The next question is what can NLP help us do? It can help us to structure text, to add tags to texts to make them easier for us to understand. These can range from the timestamp, to understanding which topics a text is about and pretty key, what tradable ticker could be impacted by text. Other important tags include understanding the sentiment associated with a text, how novel that text could, the amount of readership a text has got etc. Once we have structured a text, we can then construct all sorts of things, such as indicators or decipher which types of topics are being discussed the most etc. 


The key question with NLP is how much work you want to do versus a vendor? I would argue that particularly when you begin to tackle NLP, having a vendor to do a lot of the heavy lifting is helpful. Yes you can try to do a lot of the “tagging” yourself, but in many cases, a vendor can do it quicker and cheaper. This is particularly the case if you don’t have access to a massive quant team with NLP experts. There are also open source tools too for NLP. I’d also say that in some areas, such as macro, NLP hasn’t really been used quite so much, and it isn’t too late to start if you’re an investor who wants to use it. If you’re interested in NLP, take a look at The Book of Alternative Data, which Alexander Denev and I wrote as well!