06/02/2019
The Lioness, the Herd of Deer, and the Behavioral Algorithm
ARTICLE BY RONEN CHEN
****
How does behavior affect the size of the loan I can request? It’s all about Game Theory in the marketplace.
In the financial world we all know, banks, credit companies and other financial institutions provide credit to businesses or individuals … by leveraging fiscal models developed in the previous century.
These models guide the credit provider in deciding the size of the loan and the interest rate (derived from a risk assessment of the loan applicant), all in accordance with the client's consumer history. After all, who among us has sat facing a bank clerk to ask for a loan (for studies, a car, a wedding…) and without insight into to the bank's data, can predict the total sum we can expect, or the terms of the loan we would receive.
But in recent years, credit institutions have begun to recognize that they cannot safely give credit to a “casual,” unknown customer (perhaps an account holder of another bank) when they have no access to financial information. This case creates a situation in which the credit provider has no financial picture of the customer; the possibility of receiving credit is almost non-existent, and both parties lose out.
The need to increase the customer base on the one hand, and to be cautious about risk on the other hand, has led institutions to seek new and varied solutions to deploy as they carry out risk assessments for new, unknown customers – customers who do not necessarily have financial data to inspect and by which to conduct a risk assessment.
You may have read about, heard of, and perhaps studied the concept of game theory without diving deep into the layers that make up this world. Game theory allows us to make assessments, and analyze varied and extreme scenarios against any behavioral process in a game. Indeed, in the situation discussed above, there are two “players”: the loan applicant (“Player A”) who faces the loan provider (“Player B”). As has been said: "The whole world is a stage."
The merging of behavioral economics – used for academic studies on a variety of situations – with game theory, has introduced to the financial world a new tool: the behavioral algorithm.
A behavioral algorithm is born of the combination of methodologies from behavioral economics and game theory. It tackles questions like: What motivates us to spend more money? What are my key considerations when I purchase a new vehicle? These are in contrast to the other facets: What am I saying to the world when I purchase that vehicle? What is my favorite color and why? The range of such questions is obviously dependent on the sector that we wish to encompass.
Let’s illustrate the dynamic through a scene can be found on the nature channel, in a clip where, on one side of the equation, we watch a lioness marking its prey: a herd of deer. The herd, which until a few minutes ago enjoyed the sun's rays and the abundance of food, spots the lioness and begins to flee in order to escape. But suddenly, curiously, part of the herd begins to leap and bound to dramatic heights. It may look odd to us, but this tactic is a signal to the lioness: “Look, we are strong, agile and fast; your chances of catching us get smaller with every second.” This approach often discourages the attacker, who leaves to find an easier, more accessible target. The lioness has decided on this action by observing clues.
How does it work?
The behavioral algorithm consists of several layers of information that it collects, analyzes and learns from. The process involves collecting information about the customer from a variety of data sources ranging from social networks, forums, open databases, and even databases in Darknet (online data that is technically available, but very, very hard to discover and retrieve). When the information is cross-referenced, a digital profile emerges for this customer. This digital summary allows the algorithm to create a kind of “player” on the “business map” and as such, conduct an in-depth breakdown of its most likely financial preferences. Here the scope of implications is enormous from, say, an analysis of the behavior of the loan applicant: the adjustment of the upper limit of the loan requested in an application, combined with additional information data, allows us to analyze the level of “urgency” of the loan applicant.
Moreover, the behavioral algorithm enables the analysis of information at any given moment, all the while adjusting and learning. It is similar to a living
organism that knows how to find food, digest it, and to ignore or expel what it doesn’t need. It learns how it can “improve its position” the next time, introducing more rich and nutritious food to yield a better final result. In our case, that means achieving more accurate results, providing decision-makers in the organization with reliable, effective and useful information in a very short time, and at any given moment.
The creation of what’s called a “data matrix,” in addition to creating a digital picture that includes a behavioral basis of the loan applicant, essentially creates a representation in which the algorithm can give a score (for the purpose of this discussion, a percentage level) of the likelihood of the loan recipient to return the money, and the risk level from which the interest rate will be empirically and specifically derived.
The collection of all this data occurs automatically and completely autonomously, – but with the approval of the loan applicant – enabling the system to assess the situation at any given moment. For example, we’d have the ability to know whether the borrower’s day is structured from set, consistent processes (travel to the workplace, departure for meetings, etc.), which is an indication of predictive, balanced activity.
This approach not only allows the algorithm to receive information in real-time and for a predetermined period of time; it allows you to collect data and analyze various extreme or unusual situations that signal back to the algorithm (remember the deer!) that there is a problem in which indicate that the borrower will not be able to pay the loan. In other words, the power of the algorithm is not only the actual initial underwriting (before the customer receives the money), but also afterward as an independent, ongoing efficient system tracking the borrower, ensuring that he will be able to make his payments.
The combination of established models from financial theory, together with the behavioral algorithm, can maximize profits of credit institutions, reduce exposure to financially unattractive customers, and minimize the risk of loans that have been granted and will not, in all probability, be returned.
Ronen Chen is the CTO at BizLend ביזלנד, leading the development team in the field of behavioral algorithms.
