Data Science and Machine Learning In Finance

Introduction

Data Science and Machine Learning have expanded to many industries and sectors. These concepts are very prevalent anywhere you look from medicine to manufacturing, to insurance, to natural resources, and so many more. But one of the biggest industries that machine learning and data science have taken over is the finance industry. This industry has a massive place for machine learning and data science as its core is analyzing historical data, finding trends, and acting on those insights – just the perfect thing for machine learning and data science.

Article Summary

In an article written by Moez Ali (the article can be found here), he talks about 4 major use cases of data science and machine learning in finance and gives some examples of them. They are:

1. Risk assessment and fraud detection – This is where machine learning algorithms try to identify patterns that are associated with high-risk transactions or are possibly fraud by analyzing historical data and applying the learned trends to now
2. Customer segmentation and personalization – This is where customer data is analyzed, and customers are placed into segments and targeted with personalized ads, plans, offers, etc.
3. Trading algorithms – this is where algorithms are trained to tell the person whether to buy or sell an asset via historical data
4. Regulatory compliance – There are two parts to this:
   • The automation of reporting the financial activities of these firms to the government
   • Fraud detection

One way that these algorithms could be tested before being put into use in the current markets is by backtesting. Backtesting is where an algorithm is tested on historical data to find any errors or flaws to be corrected. Here is one path that an algorithm could take on its way to be implemented:

1. A need for the algorithm will be determined
2. The algorithm is written
3. The algorithm is trained on historical data
4. The algorithm is backtested (tested on historical data) and then troubleshot to ensure no flaws remain
5. The algorithm is put into use in the current markets

My Take

Overall, I found this article very fascinating, its combination of technical jargon with non-technical explanation and context really made this an interesting read. I also liked how the author gave real-world examples of the algorithm types. I found it really interesting how these algorithms are developed and how they are implemented with backtesting. I found it really smart how they figured out a way to test the algorithm on real-world markets and see if the algorithm would work instantaneously. I have a question though – how far back of market data is testable data? I had this question because the world evolves quite rapidly, so a market reaction to something 100 years ago might not be the same as now due to the advent of new technology and much more.

Conclusion

All in all, this was a really great article. It mixed both technical jargons with context and non-technical content. I recommend you guys give it a read (the article can be found here)!