AI in Finance – a report by the Alan Turing Institute report

A recent report by the Alan Turing Institute is worthwhile since it penetrates into the hype of AI (Artificial Intelligence) in financial services and pulls out four tangible areas that will see significant change in coming years; (1) fraud detection (2) use of chatbots (3) regulation and policy making and (4) algorithmic trading. The report makes for a good read and while I agree on the first three use cases I am skeptical about the impact AI/ML will have on algorithmic trading. Below is an assessment and critique of the 4 areas with supporting links to relevant articles;

(1) Fraud Detection – Banks and credit card companies have already been using ML to detect credit card fraud for some time now. This use case will only grow as more online data is available due to continued increase in online e-commerce activity (15% of retail sales are now online as per the 2019 report produced by celebrated internet analyst Mary Meeker) which allows for more training data to be available for the algorithms learn and grow smarter. However, this usage will be similar to an arms race as online fraud activity also continues to increase; most recent reports note that 46% of Americans have dealt with credit card fraud in last 5 years. AI and ML are well suited for this detecting online fraud due to the large volume of both fraudulent and non-fraudulent activity available which cannot be efficiently processed by humans. The one downside is that frequency of ‘false positives’ will also increase with prevalence of ML in fraud detection which could result in some loss of customer royalty. This high incidence of false positives has also been one of the challenges of using ML to detect AML (Anti-Money Laundering) where there is very little data for true positives. However the use of AI/Ml in AML is expected to increase due to risings costs of AML compliance (around $25B in 2018) using humans and a rules-based approach which have a very low detection/success rate. One of the global big banks, HSBC with a lot at stake with respect to AML is already using an AI/ML start-up, Quantexra AI for detecting money laundering.

(2) Chatbots – chatbots are in use with some firms (e.g. Erica by Bank of America and Amelia by Allstate) and in pilot stages with other firms and this usage is expected to gain traction in coming years for a few reasons. One is that customers are looking for personalized advice but want lower costs and this is where the algorithms aka “robo advisors” come in handy because they can provide tailored customer interaction at a much lower cost than hiring additional customer service representatives. Additionally, millennial customers feel more comfortable interacting through an app rather than talking in person to a financial advisor. Similar to above online fraud detection, this usage will also increase as customers do more transactions online and generate data which can be harnessed by the AI/ML algorithms to provide rich, personalized interactive transactions.

(3) Regulatory and Policy aspects – given the increasing costs of regulatory compliance and the focus on profits margins, financial firms are experimenting with using AI and ML for regulatory compliance as noted for online fraud/AML and this usage will continue to grow. However, it is interesting to learn from the report that in addition to financial firms, increasingly the regulators are using AI and ML for monitoring the activities of financial firms. AI and ML algorithms have an inherent advantage in processing the enormous volume of transactional data and the reams of reports submitted that is needed to be analyzed by the regulators to identify anomalies and suspicious transactions. The SEC, the Monetary Authority of Singapore and the UK Financial Conduct Authority are 3 regulators at the forefront of using AL and ML for such surveillance. 

(4) Algorithmic Trading –   The four common algorithmic trading (AT) categories cited in the report are (1) signal processing (2) market sentiment analysis (3) news reader and (4) pattern recognition. The report mentions that 50-70% of equity trades are being done through AT however use of AT to execute rule-based trades is vastly different from executing trades based on ‘learning’ using AI or even ML. Using AI and ML to learn from market sentiment or market signals and news and patterns and make trades is not forseeable in the future. One of the main reasons for this and why  AI and ML are not having a significant impact on algorithmic trading is due to lack of financial data which is labelled and without bias and errors. Given that financial firms have not historically viewed data as an asset and not spent enough resources on having quality, front to back data making it available for AT is challenge. Other challenges present in wide-spread adoption of AI and ML in trading besides data are noted in an article in International Banker which mentions other hurdles of adopting AI and ML in trading such as black-box effect; narrow focus and lack of responsibility.  In addition, the well known AI and ML expert Lopez De Prado who was at AQR Capital until recently has a well known presentation on the 7 common reasons why most ML algorithms fail in Finance.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s