Machine Learning in Finance

Kumar Sambhav, Manager, Research Computing, NUS Information Technology

Some of the most effort-intensive tasks within the financial services and applications have been managing assets, evaluating levels of risk, calculating credit scores, and even approving loans. The amount of data that has to be scoured, read and understood is humongous and humanly impossible and even if it is done with extensive care, it might not fetch proper results. Machine Learning models thus come in handy for such tasks as, instead of us humans doing the processing, we let computer programs handle it for us.

The data that is encountered in financial applications are mainly daily transactions, bills, payments, vendor quotations, customer profiles, credit reports etc. The data that is collected can be used to train Machine Learning models and put to the following applications:

Fraud Detection

One of the biggest problems that financial institutions face today is fraud. Fraud happens in many ways and at multiple places. Credit Card fraud, fraudulent transactions on debit accounts, forging customer documents for credit applications etc. are some of the ways in which frauds happen. The monetary impact of financial frauds amounts to billions of dollars being lost every year. Financial institutions in which frauds are frequent also lose customer trust and brand value.

ML models scan through huge datasets and look for unique activities or Anomalies in transactions and flag them out. These flagged transactions can then be investigated further. You can read more about anomaly detection here: https://scikit-learn.org/stable/modules/outlier_detection.html

Algorithmic Trading

Trading decisions based on emotions, which is a common limitation among human traders whose judgment may be affected by emotions or personal aspirations. However, Algorithmic Trading makes trading decisions devoid of emotions and focuses by analysing large volumes of data. Trading models make thousands of trades every day through fast trading decisions which gives traders a significant advantage over conventional traders. The trading models come with a set of instructions on various parameters – such as timing, price, quantity, and other factors – to help in placing trades without the trader’s involvement. More about Algorithmic trading can be found at: https://www.datacamp.com/community/tutorials/finance-python-trading

Portfolio Managers

Portfolio managers are online applications created using ML concepts and act as automated advisers. Based on a certain risk tolerance of an investor, the algorithms help to establish a financial portfolio. Robo-advisers, as they are called, are much cheaper alternative to their human counterparts and like algorithmic trading models, provide inputs for investment, devoid of any emotions or personal ambitions. This blog is an interesting take on Portfolio Managers: https://blog.quantinsti.com/portfolio-management-strategy-python/