What has AI made possible in the financial industry that wasn’t possible before?

The aim of artificial intelligence is to enable computers to autonomously perform the tasks assigned to them based on algorithms, while reacting adaptively to unknown situations.

Let’s take an example from asset management: today, personalized investment decisions can be made in line with the customer’s individual needs and everything can be calculated automatically by computers. In the past, this was either a very lengthy process or it was replaced by standard solutions. Or another example from investment banking: AI is already able to make automated optimal decisions about how and at what time orders for shares are executed. Until a few years ago, most buy and sell orders were handled manually by phone. The capital asset pricing model developed in the 1960s, which formulates the expected return of shares on the basis of historical capital market data, still assumed a linear relationship between an individual security and the overall market. But not all things are linear in the world. Today, with AI, we’re able to calculate the expected returns and prices of derivatives in such a way that we can take complexity into account. However, this isn’t calculated on the basis of a linear equation, but on a complex mathematical function that can have millions of parameters.

What is the inspiration behind artificial intelligence?

Firstly, the data, secondly the mathematical methods, and thirdly giving today’s computers their capabilities. Never before has there been so much data from so many different sources available to us, from which we can find correlations with AI. It wouldn’t have been possible in this form a few years ago. While today a capacity of 1 billion computing operations per second costs around 4  cents, the same operation in 1950 cost the equivalent of around 1,800 billion dollars.

What kind of added value could the financial industry gain from AI applications?

Banks can harness AI to both reduce costs and increase revenue. For example, banks can apply machine learning to their loan process to optimize credit decisions: AI can discover patterns in data not found by humans to make better lending decisions – even if the process is not fully automated. So, overall, AI contributes to competitiveness. 

In which areas of a bank is AI a topic of interest?

Artificial intelligence affects all areas of a bank – whether it’s retail banking, asset management or investment banking. This covers both the front office, where new products and services are needed for customers, the middle office, where, for example, fraud detection is needed, and the back office, where processes need to be automated.

Does artificial intelligence have any limits?

First and foremost, in the company’s IT infrastructure, which needs to be able to handle the data, connect the various databases with each other and provide the necessary computing capacity. Secondly, it’s important to be aware of data protection and to comply with the regulatory requirements in this regard. Not everything that’s possible may be deployed. In Switzerland and in many other countries, regulations stipulate that it must be clear and comprehensible at all times how a result was achieved. Let’s look at the example of a credit decision again: If I, as a customer, want to know why I didn’t obtain a loan, the bank has to provide me with plausible reasons. If the bank is now using very complex mathematical models with millions of parameters to reach its credit decisions, it’s very difficult to explain to the customer in a comprehensible way how these decisions came about. It’s generally different when using less complex, rule-based machine learning services such as Roboadvisory: here, the results are usually comprehensible.

Is AI becoming a key competitive factor in the financial industry?

AI is already being used successfully by leading banks in many areas. The technology will develop in such a way that all financial institutions will have to start using certain basic functionalities in order to remain competitive. An efficient IT infrastructure is indispensable. But not all banks need to play a pioneering role in this field.

What areas of AI research are you currently working on?

My team and I are involved in a wide variety of research projects funded by the Swiss National Science Foundation, Innosuisse and the European Union, as well as the financial industry. The areas of research range from credit risk models in peer-to-peer lending to fraud detection in blockchain payments and optimal trading decisions. We’re also working on making complicated AI models comprehensible. This is what we refer to as explainable artificial intelligence, which is a set of processes and methods that enable human users to understand the results and output produced by machine learning algorithms. Another exciting field of research is what’s called reinforcement learning. Here, the computer independently makes optimal decisions in a complex environment. The computer can run millions of simulations in order to continuously develop.

And what is the scope of application for this in the financial industry?

For example, investment banking. On the financial market, investment bankers compete with other market players every day to make buy and sell decisions. In our research projects, we look at how to harness reinforcement learning to make optimal, automated trading decisions. Unlike chess, for instance, which is played according to clear rules, financial markets are extremely complex. There are lots of uncertain and random events that play a role – what we call high noise-to-signal ratio.

So how do you teach the computer to be a successful trader?

We give the computer the goal of making as much profit as possible at the end of the day and programme it to analyse an inconceivably large number of variants. It buys and sells randomly at first, and optimizes its behaviour based on the profits made at the end of the day. Since it’s practically impossible to calculate all the options despite the massive computing power available, sophisticated mathematical procedures are needed to find a solution here.