Just got another example of the fact that when you get an idea, you should seriously consider that someone may already have explored it. Basically something along the lines “Before shouting Eureka and dreaming of your Nobel Prize speech, just Google the concept and find out how many books were already written on the topic...”.
Other than bringing you back to earth regarding your IQ, it can actually save you a lot of time. Other people's research may already have proven that the idea works and explained how, or you may discover without sweating too much time on it that it's actually crap...
Concretely, trying to analyze how I could leverage neural networks algorithms to better forecast volatility, I ended up running a quick Google search only to discover at least 5 solid books on Amazon dealing only with that specific target: applying the well known artificial intelligence development tool to quantitative finance and market data forecasting...
Now it may be time to tell a few high level words on neural networks for those interested. Basically, computer science folks specialized in machine learning / artificial intelligence developed in the 80's an algorithm built around a network of multivariate functions which behave both individually and all together as a piece of human brain tissue: each component has the ability to input multiple “impulses” from other components, and has the capacity to send in response a single “impulse” to one of several other components. The link between the inputs and the output of each component is based on parameters which can be “trained” on a data sample to “learn” how to best predict the final output of the overall network (a bit like you set up a linear regression on a sample to then predict the outcome of events outside of your sample).
The above is a massive shortcut, but it summarizes the main idea. For those who want to dig deeper, I would heavily recommend the online free “Machine Learning” course designed by Professor Andrew Ng, Director of the Stanford University Artificial Intelligence Lab, and his team:
There is an advanced track which requires completing assignments, but you can also sign up simply to watch the video lectures and download the related notes. And lectures 8 and 9 happen to be on neural networks.
Anyway, following this class and looking at the applications of neural networks mentioned there, I could not see any around finance, and for a second I thought I may have put the finger on something... But as mentioned above it was simply the fact that The Google Check Reflex should prevail.
Cheers,
Olivier
No comments:
Post a Comment