Of Bank Accounts and Behavioural Economics: Will adaptive learning improve money management interfaces for the poor?
One of the biggest financial challenges anyone faces is the interrelated tasks of budgeting, paying, and saving from income. The rich and poor alike need help in this regard and there has been plenty of work by behavioural economists showing how reminders to follow through on savings behaviours, commitment accounts, account labelling, automatic deposits, formal commitments to friends, and other features can help with figuring out the right strategy, or maintaining the course once it has been set. This challenge is exacerbated for microentrepreneurs who have the added complexity and exigency of running an informal business.
A single financial interface that integrates savings, budgeting, paying and supports goal achievement
A year or so ago, Ignacio Mas and Colin Meyer came out with a paper that presents a new approach to how people could manage their diverse payment, cash-flow management and commitment savings needs simply and intuitively, from a single account and through a mobile interface. It builds on the logic of mobile money platforms, which provide customers with the ability to initiate real-time electronic payments from their mobile phone. By introducing the notion of forward (or deferred) payments, it is possible to create a much richer set of uses for the basic transactional account which caters to people’s need for commitments and earmarking of funds for specific goals.
Mas took the concept further in a second piece where he expands upon the concept and attempts to build a conceptual framework that could be used to organize the features that such an interface should have. The goal of the interface is to give people a simple tool that helps them structure their budgeting and their payments to meet their financial goals, rather than a series of one-off products that individually target some specific need, but can be a confusing and inefficient mess when clients are presented with a whole catalogue of them.
Mas has recently teamed up with MicroSave (with some funding from the Gates Foundation) to explore this concept further and take it to the field to test with actual clients in Bangladesh and India (see Metamon concept paper here).
A parallel concept for micro-entrepreneurs
A recent research study by Alejandro Drexler (University of Texas-Austin), Greg Fischer (London School of Economics), and Antoinette Schoar (MIT) shows that that teaching microentrepreneurs basic heuristics for managing finances (“rules of thumb accounting”) can improve efficiency and profitability outcomes.
The authors’ description of the rule of thumb training:
“In contrast to standard accounting training, rule-of-thumb training, focuses on very simple rules of thumb or routines for financial decision making without aiming to provide comprehensive accounting knowledge. For example, both the standard accounting and rule-of-thumb trainings taught participants to separate their business and person accounts. In the standard accounting training, this was followed by instruction for how to calculate business profits in accordance with a typical accounting curriculum for micro-entrepreneurs. In contrast, the rule-of-thumb training gave them a physical heuristic of how to keep money in two separate drawers (or purses) and to only transfer money from one drawer to the other with an explicit IOU note between the business and the household. This gave entrepreneurs a simple way to figure out the profits of their businesses: at the end of the month, count how much money was in the business drawer.”
The authors conduct a randomized control trial that pits “rule of thumb” training program against a formal accounting training program and finds that entrepreneurs in the rule of thumb class were more likely to keep records, know their monthly revenues, and separate their money for the business and the home than the entrepreneurs who were formally trained. They also had slightly better sales on average and markedly better sales in bad weeks (when they were more likely to have borrowed from the business to deal with problems.)
The “rule of thumb” procedures work probably because (i) they help overcome behavioural biases and/or limited cognitive capacity and (ii) are simpler than full accounting principles (which are so complex they might even exacerbate cognitive capacity limits by overwhelming people.)
Imagine rules-of-thumb as an interface…
Stretching the results of the research a little, one can imagine a smartphone based interface – along the lines of the Metamon project above – which could implement the simplified accounting framework. It could expand on rules of thumb functionality to keep records, set reminders, and have other useful features to manage money – without the full complexity of modern accounting rules. In fact, one could imagine the interface wouldn’t be very different in concept from the one proposed by Mas – what is accounting but a sophisticated way of managing payments?
Both of the approaches outlined above reflect new approaches to financial management for households and microenterprises. However, as described, they are essentially static interfaces and don’t adapt to the individuals who own them. A number of the behavioural economics studies have showed that different client groups react systematically differently in the ways they respond to product features. For instance hyperbolic discounters (essentially, people who have a hard time sticking to a savings plan) receive more benefit from SMS reminders to save in some studies. Similarly, women who report having lower levels of bargaining power receive more benefit from commitment savings accounts in another study. Some clients need help figuring out budgeting; others need discipline or reminders; others need other features.
The question then is: Could we employ adaptive learning algorithms (which are currently being used in education to adapt lessons to student progress) to guide clients toward better goal setting and financial management behaviour? Could clients be pushed to reach increasingly complex financial behaviours and use cases – say, moving from effective budgeting within the month to longer-term savings or paying down debt?
What’s the yardstick?
What would be needed would be a clear and measureable yardstick of performance against which to judge outcomes that would guide the learning algorithm. In education, homework problems and quizzes have answers, when a student gets more correct answers more quickly the algorithm knows it has succeeded and will adapt accordingly. With financial behaviours, it can be more difficult to measure success. When I set a savings goal, not attaining it can mean I lacked the will, or it could mean I changed my mind for very good reasons, e.g. I had an emergency that was more urgent, or an opportunity to invest that was even better. Is there a viable yardstick against which to pit financial behaviour learning algorithms?
It turns out behavioural economists have already wrestled with this topic. Behavioural economics rests on the notion that people sometimes act irrationally, against their own interests. To prove this, behavioural economists had to come up with objective criteria by which they could measure someone’s best interest – and this isn’t easy since personal preferences determine what I want and therefore determine what is in my best interest (hey, if I want to blow all my money on my Hello Kitty sleeping bag collection, who are you to say I am acting against my own interests!)
Usually, the best behavioural economists can do is show that two choices/actions are inconsistent, thus that at least one of them was irrational. For example, when employees are automatically enrolled in 401(k) programs, a lot more of them contribute than when they are asked to voluntarily sign up. Think about a person who would participate under automatic enrollment, but would not participate under self-enrollment – we can’t know from observing these two behaviours which outcome is optimal for them, but we do know that enrolling and not enrolling can’t both be the best choice.
If we are being purists, we would have to stop there. However, since we aren’t all academic economists, we can be practical. For example, we can often figure out which outcome the optimal one by asking the person, or observing their behaviour over the long run (most people who fail to self-enrol will likely say they did intend to sign up but got distracted, and people who defaulted into the program will tell you they are happy with the choice.)
So (and thanks for hanging in this long) the question is, what would we want to optimize against? What simple, quantifiable, easily measureable indicators of financial behaviour should we try to increase or decrease? Would it be sufficient to allow attainment of savings or budgeting goals to be the metric? Or would there be better ways guide a learning algorithm to measure success?