Research & Innovation

Stevens Initiates CRAFT-Sponsored Fintech Research Program

Interdisciplinary Stevens-directed projects address forecasting, randomness, equity in credit decisions, more

The Stevens-led CRAFT (Center for Research toward Advancing Financial Technologies), which launched in fall 2021 as the first-ever NSF-supported center devoted to financial technology research — and will hold its next Industry Advisory Board meeting on the Stevens campus November 3-4, 2022 — recently announced the kickoff of a robust research program.

Backed by its landmark National Science Foundation award, CRAFT recruited a Chief Research Officer and began funding an initial suite of fintech research projects earlier this year.

Two of those initial projects involve Stevens faculty and student teams, working to create technologies that address technical investment, portfolio management and market stress-testing — as well as broader societal issues of equity and fairness.

Using quantum tools to improve financial decisions

For one project, an interdisciplinary Stevens team will leverage the emerging tools of quantum science in an effort to help portfolio managers, investors and even automated systems make better decisions.

“There are strong industry needs for faster computing, stronger security, and better scalability in big-data finance,” explains business professor Zhenyu Cui, the project’s lead investigator and an expert in financial engineering and the application of theoretical and algorithmic methods and models to financial applications.

Dr. Zhenyu Cui with the Manhattan skyline in the backdrop.Business professor Zhenyu Cui leads Stevens' first CRAFT-funded quantum fintech research project“Quantum processes may help provide that.”

For the CRAFT-supported project, Cui’s team — which also includes Stevens quantum physicist Rupak Chatterjee, business professor Chihoon Lee and financial engineering Ph.D. candidate Zhiyang Deng — will work on at least three means of leveraging quantum techniques to address financial prediction, risk management and other complex challenges.

“Our method, at its core, is quantum computing,” Cui explains, “and in different applications we may combine and enhance the method using artificial intelligence (AI) and machine learning methods.”

First the team will design a quantum-based framework that is “model-free”— which simply means that it will not begin with any built-in assumptions about current and historical financial markets, nor their best- or worst- case scenarios. Instead, the new system will introduce observed data (in this case, real-time options prices) directly into its simulations of asset prices. 

“One famous statistics quote says ‘all models are wrong, but some are useful,’ ” says Cui. “The idea of ‘model-free’ method is to make the method more robust and not subject to model-misspecification errors.”

Separately, the team will also develop new quantum-based modeling tools useful for risk analysis and asset valuation of very complex financial products — cliquet options, equity-linked derivatives, barrier options or VIX options, for instance — and also complex insurance products (such as variable annuities with different guarantee structures, flexible-premium variable annuities, and variable annuities with fee structures closely tied to volatility indexes).

New quantum-driven algorithmic tools may prove more accurate than those currently used in industry, Cui says, because complex financial simulations require highly random numbers and random samples as they run — and truly random numbers can now be derived from quantum mechanical processes due to their inherent randomness.  

“Quantum random numbers are demonstrably more random than the ‘random’ numbers we currently generate and use in industry,” he notes. “Our own Stevens researchers, including Dr. Chatterjee, have already proven this in recent experiments generating random numbers using quantum mechanics — a collaboration between our Center for Quantum Science and Engineering and the Hanlon Financial Systems Center.”

The extreme randomness generated by those processes, says Cui, can power ever-better Monte Carlo simulations to (for example) accurately valuate derivatives, Asian options prices, volatility options or other financial products and markets.

“Industry is very excited about the possibility of quantum random numbers,” he points out.

Finally, the Stevens team will attempt to use quantum-driven optimization processes to test two widely used algorithmic techniques — QUBO, or Quadratic Unconstrained Binary Optimization, and QAOA, the Quantum Approximate Optimization Algorithm — for portfolio analysis and real-time, automated investment advising (also known as “robo-advising”) and portfolio rebalancing. 

“The theory is that we can train algorithms driven by quantum processes to update portfolio weights with very fast real-time moves that react to market conditions faster than a human could, or existing financial algorithms can do, during for example a rapid downturn or market crash.”

Opening the hood on credit decisions

For another early CRAFT-funded Stevens project, Stevens computer science professor Jia Xu is examining important questions of fairness and equity — specifically with regard to personal loan extension or denial decisions, which are often made by automated systems based on personal data.

“AI-driven systems are not yet very good at explaining the ‘why’ of their predictions and decisions,” explains Xu. “Most AI is not very transparent at all. It’s often mostly a ‘black box.’ ” 

Jia Xu (jxu70)Professor Jia Xu explores the processes behind AI-driven credit decisionsTo improve this transparency, Xu and CRAFT Center Director Steve Yang are examining issues of fairness and explainability in AI-assisted credit decisions by building a better AI. Their project, “Causal Inference for Fairness and Explainability in Financial Decisions,” will complete and publish findings in mid-2023.

In previous research analyzing a University of California, Irvine default-payment dataset, Xu has found that given an individual’s credit card limit and payment and billing history, eliminating certain potentially biased attributes and variables from credit-decision systems — including gender, age, education level and marriage status — does not significantly reduce the accuracy of those decisions with regard to likeliness to default later.

“The conventional approach, which is silencing some protected attributes in a model, cannot entirely eliminate bias from features like zip codes that act as proxies for other data points,” Xu explains. “It is essential to understand the sources of bias in a model.”

The team will create a novel learning model that deploys certain special types of algorithms within itself to report back as it runs, helping observers understand some of the processes and features of its automated decisions.

Xu hopes the investigations can inform the building of improve systems that generate more equitable decisions across ethnic backgrounds, genders, and geographic areas.

In addition to its usefulness for credit-issuing organizations, she adds, understanding how algorithmic credit decisions are made will also be useful to consumers who wish to improve their credit worthiness.

Stevens computer science doctoral candidate Xuting Tang and postdoctoral research Abdul Rafae Khan will also contribute to the project.