The archaic, paper-based world of trade finance looks set to take a further leap into the digital future, as trade finance distribution platform Tradeteq begins a collaboration with the Singapore Management University (SMU) to explore quantum computing-based solutions for the industry.

Supported by the Monetary Authority of Singapore (MAS) under its artificial intelligence and data analytics (AIDA) grant scheme, the research project, titled Exploring the advantages of a quantum system for machine learning applied to credit scoring, aims to build a predictive machine learning model implemented on a quantum computer in order to tackle inefficiencies in approving trade finance.

“Currently, many small and-medium-sized businesses are unable to grow their companies due to a lack of funding as they are deemed ‘too risky’ by current credit rating models,” says Pang Hwee Hwa, dean of the SMU School of Information Systems. “With shorter processing time, more businesses could be scored and with greater accuracy thereby creating more trust and providing greater access to finance for companies than ever before.”

Quantum computing is still very much in its infancy, and the technology doesn’t yet exist to build a large-scale quantum computer. However, SMU and Tradeteq believe their work may be the first to show a practical advantage for a financial application for the technology as quantum computing continues to improve.

Unlike classical computing, which uses bits that can either be in an on position – represented by a one, or an off position – represented by a zero, quantum computing allows for uncertainty, or a spectrum of states between on and off. A qubit – a basic unit of quantum information – can be on, off, both on and off at the same time, or somewhere between the two.

To find out why this would be useful in the trade finance setting, GTR speaks to Michael Boguslavsky, head of artificial intelligence (AI) at Tradeteq, who is part of the research team on this project.


GTR: Why is Tradeteq embarking upon this project?

Boguslavsky: Tradeteq’s artificial intelligence (AI) credit scoring capabilities are already industry leading and this project we are embarking on with SMU is going to further develop our technology. We are exploring the development of quantum-based neural networks to more quickly and more accurately give credit scores to SMEs and transactions, allowing them access to trade finance which, under normal credit reporting, would not have been possible. Quantum computing is set to be a gamechanger for many sectors, and this project will hopefully lead the charge for trade finance.


GTR: What could quantum computing bring to the table for trade finance?

Boguslavsky: We have done experiments with simulated quantum algorithms for portfolio optimisations, and this has proven to be quite promising. I would not be surprised if we start to see simulated quantum algorithms used for portfolio optimisation relatively soon.

Trade finance portfolio optimisation is different from equity portfolio optimisation. In an equity fund, if you have different opinions about different stocks and you want to optimise your portfolio, for each stock you can make a lot of different decisions.

I may want to buy five shares of a stock, or maybe 20 shares. I may want to be short or long on a position. There is almost a continuous spectrum of decisions, which means that you can smoothly vary the number of shares of each stock in order to reach an optimal level. That makes it a very simple use case for an algorithm.

In trade finance, however, either you take a receivable in your portfolio or you don’t. You can’t take half of it; you can’t take minus two. Therefore, finding the right portfolio by brute force would need computing power comparable to breaking enterprise grade encryption. That is not feasible because it would take too long.

However, with quantum computing, it would be possible to speed this up sufficiently so that it works for portfolios of several thousand transactions.


GTR: Should portfolio optimisation for trade finance origination and distribution therefore be the main focus for quantum computing efforts?

Boguslavsky: Our research work will focus on another area. While we are working on portfolio optimisation, we find that we spend much more time on assessing the risk of each individual receivable and on credit scoring for companies or for transactions. We want to explore the extent to which simulated quantum or actual quantum computing can help with credit scoring.


GTR: How can quantum computing be applied to credit scoring in trade finance?

Boguslavsky: We do two types of credit scoring: company credit scoring and transaction credit scoring.

Company credit scoring is a relatively long-standing area of credit analysis. There are many people doing that and there are many ways to do it. The amount of data available is limited, particularly among small companies. That limits the complexity of the model one can use for this application, and there is no point in over-engineering a solution.

Where the real bottleneck exists is in the transaction scoring. For a trade finance transaction, one potentially has a lot of data, from real-time ship tracking to data on other similar transactions. You can look at transactions involving the same buyer or seller, or transactions of similar goods or between a similar pair of countries.

All of that can be potentially relevant or indeed not relevant at all. Therefore, with transaction scoring, very quickly one goes from having too little data like one has if one analyses the credit of a small company in an emerging market, to having potentially too much data with daily information flows which may or may not be relevant for this transaction.

Here, the speed of processing and the speed of understanding what is relevant and what is not becomes much more difficult. That is where we see very clear inadequacies in the current computing systems, which we believe quantum computing may be able to solve in the future.


GTR: Given the low default rate and risk profile of trade finance, does the business case exist for pushing the needle on greater accuracy in transaction scoring? Is the return going to be worth the effort?

Boguslavsky: First of all, I am not convinced that the non-payment ratios in trade finance will remain low once we get data for March and April this year. There is currently a lot of disruption in the market.

Secondly, it really depends whether you weight the data by volume or by transaction count. A big chunk of trade by volume is done by large multinational corporations with public ratings that are what one uses to assess their risk. However, there are lots of small companies in the world, and in many cases, they lack access to trade finance. One of the main reasons is that they cannot find anyone who is comfortable with taking the risk, or who is able to spend so much time on due diligence that the risk becomes uneconomical.

Moving the needle here means solving the trade finance gap. There are lots of perfectly healthy, diligently run small businesses which deserve funding. But they are not transparent, they don’t have public ratings and they will probably not get them while they are still small companies.

So, as long as they are small companies, if they are to get funding, somebody needs to analyse their risk differently from how it is done currently, and that means transaction level analysis, that means looking at more data.


GTR: When can we expect to see quantum computing used in trade finance?

Boguslavsky: This is not something I expect to be in our production system this year or even when the project finishes in two years. That is more about us exploring possibilities for where our system could be in five or 10 years. That is not something we plan to deploy imminently to any clients.

Quantum computing is still at a very early stage of development. It is not yet suited for anything other than very small pilot problems or problems which were specifically designed to be solved by quantum computing.

If we look at company scoring, for example, Tradeteq currently analyses about 300 features for each company, upon which the credit score is based. On the quantum systems available today, there is no practical way to get near 300 features. We would be lucky to squeeze 15 features in. Getting to where the classical systems are will take several years.

However, theoretically, quantum systems offer large advantages. They can look almost simultaneously through a very large number of possibilities.

At some point in the future, be that three, five or even 10 years from now, quantum systems will reach a scale where they will be useful for these problems.

I am a practical man. I am not doing this research for the sake of finding a problem for quantum computing. I am interested in practical problems arising in credit and finance. And there, I think problems such as transaction scoring can be reformulated in such a way that they will be able to be solved by quantum computing.