The commitment device.

In 2004, a team of economists offered customers of a rural Philippine bank a savings account with an unusual feature.

The account worked like a normal savings account in most respects: you deposited money, the money earned interest, and you could eventually withdraw it. The difference was a restriction on early withdrawal. When you opened a SEED account — Save, Earn, Enjoy Deposits — you set a goal: either a target amount you wanted to reach, or a target date you wanted to reach it by. Until then, you couldn't take the money out.

This is, from a standard economic perspective, a worse product than a regular savings account. It has the same interest rate with less flexibility. Rational consumers should prefer the account without restrictions.

Twenty-eight percent of those offered the SEED account opened one. Their savings balances increased by 81% of the control group mean.

The logic of self-binding

The experiment, run by Nava Ashraf, Dean Karlan, and Wesley Yin and published in the Quarterly Journal of Economics in 2006, tested a prediction from behavioral economics: that people who recognize they have self-control problems will sometimes prefer to constrain their own future choices.

The idea has a long history. Homer's Odysseus, sailing past the Sirens, had himself lashed to the mast so he couldn't steer toward the music even when he desperately wanted to. He knew his future self couldn't be trusted in the moment of temptation, so his present self took away his future self's freedom.

The behavioral economics formalization of this is called hyperbolic discounting. People with this preference pattern value the present much more than the future — but they also value the near future much more than the distant future. Someone with this preference might say on Monday that they will save money on Friday. On Friday, however, Friday has become today, and today is much more valuable than Friday was when it was in the future. The savings plan gets deferred again.

The commitment device is a solution to this problem. By restricting access to the money, it prevents the future self from making the choice the present self knows the future self will make.

Who takes up commitment savings

Ashraf, Karlan, and Yin measured hyperbolic discounting before offering the SEED account. What they found was that take-up of the commitment account was significantly predicted by this measure: customers who showed more evidence of present-biased preferences in the baseline elicitation were significantly more likely to open a SEED account.

This is a strong finding. It means people who took up the commitment device were, at some level, aware of their own self-control problem and responding rationally to it by choosing a product designed to overcome it. They were not confused about what they were signing up for. They were doing what Odysseus did — binding the future self before the Sirens came into range.

It also means that commitment savings products are likely self-selected by the people who need them most. This is unusual in financial services, where products designed for financially vulnerable customers often reach people who don't particularly need them while the most vulnerable opt out.

The policy implications

The SEED experiment belongs to a broader literature on defaults and choice architecture, but it makes a different point than the auto-enrollment experiments.

Auto-enrollment in retirement savings works by changing the default: instead of actively choosing to save, employees are enrolled by default and must actively choose not to save. The finding from Save More Tomorrow and similar programs is that inertia is powerful — people tend not to change whatever default they're given. Most auto-enrolled employees never opt out.

The commitment device works differently. It requires active choice — someone must affirmatively open the account, choose their goal, and accept the restriction. It does not rely on inertia. It relies on self-knowledge: the recognition that future-me will make a decision that present-me doesn't want made.

Both mechanisms work. They work for different reasons and they're suited to different implementation contexts.

For public programs, the commitment device insight has several applications.

Health behavior change is an obvious domain. Smokers who want to quit but know they'll weaken; drinkers who want to cut back but know Friday evening will be different from Wednesday morning. Commitment contracts — where people pledge money, or reputation, or some other value, against a behavior change — have been tested with varying results. The evidence suggests they work when the commitment is credible and when the person genuinely wants to achieve the change. They don't work well as external impositions.

Savings for specific purposes — medical expenses, education, housing — are another domain. Labeled accounts and restricted accounts consistently show higher balances than unrestricted accounts even at the same interest rate, because the restriction helps people protect the money from their own impulse spending.

The harder question

The SEED experiment established something important: people will voluntarily accept restrictions on their own future behavior when they recognize those restrictions help them achieve what they actually want.

This creates a policy design question that is less simple than it appears.

If people want to constrain themselves, the obvious move is to offer them tools to do so. Make commitment savings accounts widely available. Create opt-in restricted accounts for health spending or retirement. Let people choose to be tied to the mast.

But the experiment also found that only 28% of those offered the account opened one. The majority didn't — not because they didn't have savings goals, but presumably because they either didn't recognize their self-control problem, didn't believe the product would help, or preferred the flexibility of unrestricted savings.

For the 28% who opened accounts, the intervention worked. For the 72% who declined, it offered nothing. If the goal is to reach the full population of people who would benefit from commitment savings, voluntary opt-in products will only go so far.

The auto-enrollment solution — making commitment the default and opt-out the active choice — has obvious appeal. But it changes the ethical character of the intervention. Voluntary self-binding respects autonomy in a way that mandatory enrollment does not. Odysseus chose the mast. He wasn't strapped there by someone else on the theory that he'd thank them later.

This is where the behavioral economics literature runs into harder political philosophy questions. How much weight do we give to the preferences of the present self relative to the preferences the person expresses when they're not in the moment of temptation? When does a choice architecture that overrides stated preferences serve the person's actual interests, and when does it substitute the designer's judgment for the person's own?

The SEED experiment doesn't answer these questions. It does establish that the self-control problem it addresses is real, that commitment devices help those who seek them, and that a substantial minority of people will actively choose to constrain themselves when given the option. The design question — how far to go beyond that minority — remains genuinely unsettled.

Next issue: The land question.