Time Inconsistency and Present Bias: Models That Fit Life

The Promise You Made Last Night

You set an alarm for 6 a.m. to go running. At 10 p.m., when you set it, the plan felt easy — obvious, even. The morning version of you would be grateful. But at 6 a.m., when the alarm sounds, the calculation has shifted. The bed is warm. The run is cold. You hit snooze. By 7:30, the plan is dead and you feel vaguely guilty, aware that tonight you will set the alarm again.

This is not laziness in any simple sense. At 10 p.m., you genuinely preferred the plan that included running. At 6 a.m., you genuinely preferred to stay in bed. Your preferences changed — not because new information arrived, but because the moment of action got closer. The future costs of skipping exercise remained the same; what changed was that the immediate comfort of the warm bed became now, and now exerts a gravitational pull that the evening version of you underestimated.

Economists have a name for this pattern: time inconsistency. It is the phenomenon in which a person’s plan for the future is not the plan they execute when the future arrives. Not because circumstances changed, but because the way they weigh present versus future changes with the passage of time. Understanding this phenomenon — modeling it, measuring it, designing institutions around it — has been one of the most productive research programs in behavioral economics over the past half-century.

Exponential Discounting and the Standard Model

Before present bias could be identified as a problem, economics needed a baseline model of how people should discount the future. The standard approach, still dominant in most economic models, is exponential discounting. Under this model, a person discounts future utility at a constant rate per period. If you discount at 5% per year, then a dollar of utility next year is worth 95 cents today, a dollar in two years is worth about 90.25 cents, a dollar in three years about 85.74 cents, and so on. The discount factor shrinks geometrically.

The crucial property of exponential discounting is time consistency. If you prefer Plan A to Plan B when evaluating them from today, you will still prefer Plan A to Plan B when evaluating them from any future date, as long as no new information has arrived. This is because the ratio of discount factors between any two future periods remains constant regardless of when you evaluate them. A person with exponential discounting might be impatient — they might heavily discount the future — but they are consistently impatient. Their plans hold up.

Paul Samuelson introduced the discounted utility model in a brief 1937 paper, almost as a convenient simplification rather than a serious psychological claim. He was explicit that the model might not match actual behavior. But convenience has a way of becoming orthodoxy, and by mid-century, exponential discounting was treated as both descriptively accurate and normatively correct — the way rational people do and should evaluate tradeoffs over time.

Strotz and the Discovery of Changing Preferences

The first serious challenge came from Robert Strotz in 1955-56, in a pair of papers that asked an uncomfortable question: what if people’s discount functions are not exponential? What if, instead of discounting the future at a constant rate, people discount the near future much more steeply than the distant future?

Strotz showed that any discount function that is not exponential will generate time inconsistency. The person’s plan at time zero for what to do at time two will differ from their plan at time one for what to do at time two. The passage of time itself changes the optimal plan — not because the world changed, but because the weights shifted.

He proposed two strategies a person might use to cope with changing preferences. The first was precommitment: binding your future self to the plan your current self prefers. This is Odysseus lashing himself to the mast so he can hear the Sirens’ song without steering the ship onto the rocks. The second was consistent planning: recognizing that your future self will deviate, and choosing only plans that your future self will actually follow through on. This is the sophisticated agent who works backward from what they know they will actually do.

Strotz’s insight was ahead of its time. The economics profession in the 1950s was not ready to abandon exponential discounting, partly because the alternative seemed to open a Pandora’s box: if preferences change, which preferences are the “real” ones? How do you do welfare analysis if the person at time one disagrees with the person at time zero? These questions remain live today.

Hyperbolic Discounting: The Shape of Impatience

The empirical case against exponential discounting accumulated slowly and then all at once. Psychologists, particularly George Ainslie in the 1970s and 1980s, documented a robust pattern in both humans and animals: steep discounting of the near future combined with relatively flat discounting of outcomes further away. A pigeon that prefers a small, immediate food reward over a larger, delayed one will reverse its preference and choose the larger reward if both options are pushed into the future by the same amount. A person who prefers $100 today over $110 tomorrow will often prefer $110 in 31 days over $100 in 30 days — even though the tradeoff is identical.

This pattern is captured by hyperbolic discounting, in which the discount function declines at a rate that decreases over time rather than remaining constant. The simplest hyperbolic form discounts a reward arriving at time t by the factor 1/(1 + kt), where k is a parameter governing impatience. Unlike the exponential function, which maintains constant proportional discounting, the hyperbolic function discounts heavily between now and soon, but much less between later and even later.

The practical implication is vivid. With hyperbolic discounting, you will always prefer immediate gratification when the choice is between now and later, but you will make patient, far-sighted plans when both options are in the future. You will sincerely plan to start your diet on Monday. But when Monday arrives and becomes now, you will eat the cake. This is not hypocrisy; it is the mathematical consequence of a discount function that weights immediacy disproportionately.

Richard Thaler, in a 1981 paper, provided clean experimental evidence. He asked people to state the amount of money that would make them indifferent between receiving it immediately versus receiving a fixed sum at various delays. The implied discount rates were enormous for short delays and declined sharply for longer ones — exactly the pattern hyperbolic discounting predicts and exponential discounting rules out.

Laibson’s Beta-Delta Model

The full hyperbolic discount function, while empirically motivated, is mathematically unwieldy. It generates infinite-horizon optimization problems that are difficult to solve because the person is effectively playing a game against their future selves at every moment. In 1997, David Laibson proposed an elegant simplification that captured the essential feature of present bias while remaining tractable: the quasi-hyperbolic or beta-delta model.

The idea is simple. Take the standard exponential discount factor, delta, which operates between all future periods. Then introduce a second parameter, beta (between 0 and 1), which applies a one-time extra discount to all future periods relative to the present. If beta equals 1, you get standard exponential discounting. If beta is less than 1, you get present bias: the present is disproportionately valued relative to all future periods, but the relationship among future periods is exponential and thus time-consistent.

A person with beta = 0.7 and delta = 0.95 evaluates outcomes this way: the present gets full weight, next period gets weight 0.7 x 0.95 = 0.665, two periods ahead gets 0.7 x 0.95^2 = 0.632, and so on. The jump from “now” to “next period” is steep (a 33.5% discount), but the jump from period 5 to period 6 is gentle (only 5%). This person will make patient long-run plans and violate them when the moment of action arrives.

Laibson applied the model to savings behavior and showed that it could explain several puzzles: why people simultaneously carry high-interest credit card debt and hold low-return savings (the debt is easy to access and spend, the savings provide commitment); why increases in credit availability reduce savings rates; why people express a desire to save more but fail to do so. The model also predicted that people would value illiquid assets — retirement accounts with early withdrawal penalties, for example — more than their purely financial characteristics warranted, because the illiquidity serves as a commitment device against their own future selves.

Empirical estimates of beta have converged around 0.7 to 0.8 for many populations, though there is significant individual variation. Some people show little present bias; others show a great deal. The distribution matters, because policies designed for the median present-biased person may be unnecessary for the patient and insufficient for the severely biased.

Commitment Devices: Odysseus, Savings Accounts, and StickK

If present bias is the disease, commitment devices are the medicine — or at least the attempt at one. A commitment device is any arrangement that restricts your future choices in order to align your future behavior with your current preferences. Strotz identified the logic; modern behavioral economists have cataloged the examples.

The archetypal commitment device is Odysseus tying himself to the mast. He wanted to hear the Sirens but knew that in the moment, their song would overpower his reason. His solution was to remove the option of acting on his future preferences by physically constraining himself. The key insight is that he wanted to be constrained. A rational agent with exponential discounting would never voluntarily restrict their own future choice set — more options are always (weakly) better. But a present-biased agent may pay to limit their options, recognizing that their future self will make choices their current self regrets.

In the domain of savings, illiquid retirement accounts function as commitment devices. The penalties for early withdrawal from a 401(k) or IRA are not bugs; for the present-biased saver, they are features. Laibson’s “golden eggs” paper showed that people with present bias may rationally choose assets with lower returns if those assets are harder to liquidate, because the illiquidity protects against their own future impulsiveness. This helps explain why defined-benefit pension plans — which are essentially impossible to raid before retirement — produced higher effective savings rates than defined-contribution plans that allow loans and hardship withdrawals.

In developing countries, Pascaline Dupas and Jonathan Robinson found that giving Kenyan market vendors a simple locked savings box — no key, just a slot — increased savings significantly, even though the box paid no interest and was not truly secure. The box worked because it introduced friction between the impulse to spend and the act of spending. Ashraf, Karlan, and Yin found that a commitment savings product in the Philippines, which restricted withdrawals until a self-chosen date or savings goal, attracted substantial demand and increased savings by 81% after one year.

StickK.com, founded by Yale economists Dean Karlan and Ian Ayres, takes the logic further. Users commit to a goal — lose weight, quit smoking, finish a thesis — and put money at stake. If they fail, the money goes to a charity, a friend, or (most effectively) an “anti-charity,” an organization whose mission the user opposes. The site works by converting a future cost (health consequences, regret) into a present cost (losing money to a cause you dislike). Early evidence suggests it increases follow-through, though the people who use it are presumably those who are already somewhat aware of their self-control problems.

Time Inconsistency in Monetary Policy

The same logic that explains individual self-control problems also illuminates one of the central problems in macroeconomics: the time inconsistency of optimal policy. Finn Kydland and Edward Prescott, in their landmark 1977 paper “Rules Rather than Discretion,” showed that a government with discretionary control over monetary policy faces a temptation structure remarkably similar to the person with the alarm clock.

Here is the setup. A central bank announces that it will maintain low inflation. If the public believes this, they set wages and prices accordingly, expecting stable prices. But once expectations are anchored, the central bank faces a temptation: a surprise burst of inflation would reduce real wages, boost employment, and generate short-term economic gains. The bank knows it promised low inflation, but right now, cheating on that promise has immediate benefits and only gradual costs.

The problem is that the public, if rational, anticipates this temptation. They do not believe the low-inflation promise because they know the bank will deviate once expectations are set. The result is an equilibrium with higher inflation than anyone wanted — not because the central bank is malicious, but because it cannot credibly commit to its own future behavior. This is exactly Strotz’s problem of changing preferences, transposed from the individual to the institution.

Kydland and Prescott’s solution paralleled Strotz’s: rules rather than discretion. If the central bank is bound by a rule (an inflation target, a fixed exchange rate, a constitutional constraint), it cannot deviate even when deviation is temporarily attractive. The costs of commitment — reduced flexibility to respond to shocks — are outweighed by the benefits of credibility. This insight shaped a generation of central bank design: independent central banks, inflation targeting frameworks, and the appointment of “conservative” central bankers (in Kenneth Rogoff’s formulation) who are personally averse to inflation all serve as institutional commitment devices.

Kydland and Prescott shared the 2004 Nobel Prize for this work, and the parallel to behavioral economics is instructive. In both cases, the core problem is the same: a decision-maker whose preferences at the moment of action differ from their preferences when planning. In both cases, the solution involves voluntarily surrendering future flexibility. The difference is one of scale — the individual locks away their savings; the institution locks away its policy discretion — but the logic is identical.

Implications for Retirement, Addiction, and Climate

Present bias and time inconsistency have concrete implications for three of the largest policy domains.

Retirement savings. The beta-delta model predicts that people will under-save relative to their own long-run preferences. They intend to save more, but when each paycheck arrives, the present looms large and consumption wins. This is not an abstract prediction; it matches decades of survey evidence showing that most Americans report saving less than they think they should. Thaler and Benartzi’s “Save More Tomorrow” program exploits the structure of present bias: by asking people to commit to saving a fraction of future raises (not current income), it aligns the commitment with the time when people are patient (the future) and avoids the immediate pain of reduced consumption.

Auto-enrollment in retirement plans works for the same reason. The present-biased person who is auto-enrolled faces a choice between doing nothing (staying enrolled, which serves their long-run interest) and taking action (opting out, which requires present effort). Present bias, for once, works in the right direction: the effort of opting out is immediate, so the person does nothing. The asymmetry of present bias is channeled by the design of the default.

Addiction. Addiction is the extreme case of present bias. The immediate pleasure of the substance overwhelms the future costs of dependence, health deterioration, and social harm — and the gap between present and future valuation may be especially large for addictive goods because the substance itself alters the discount function. Models of rational addiction (Becker and Murphy, 1988) show that forward-looking agents might still become addicted, but they add that agents with present bias will become addicted more easily and quit less readily. Gruber and Koszegi (2001) showed that present-biased smokers may welcome higher cigarette taxes — a commitment device imposed by the state — because the tax helps their long-run self resist the short-run temptation. This turns the standard welfare analysis of sin taxes on its head: the tax is not a pure deadweight loss but a partially welcome constraint.

Climate policy. Climate change is the time inconsistency problem at civilizational scale. The costs of mitigation are immediate (higher energy prices, industrial restructuring, political conflict), while the benefits are decades or centuries away. A present-biased electorate and the politicians who serve them will systematically delay action, not because they deny the science, but because the discount function makes future catastrophe feel abstract and present sacrifice feel acute. International climate agreements face the additional layer of time inconsistency that Kydland and Prescott identified: nations promise to reduce emissions, but when the moment of compliance arrives, the temptation to free-ride is overwhelming and credible commitment mechanisms are weak.

When Present Bias Might Be Rational

Not all apparent present bias is a mistake. There are circumstances in which steep near-term discounting is a reasonable response to the actual structure of the world.

If the future is genuinely uncertain — if you might not be alive next year, if your income might collapse, if the political regime might change and confiscate your savings — then consuming now rather than saving for later is not present bias but rational adaptation to risk. People living in poverty, in conflict zones, or with serious health conditions may exhibit steep discounting not because their psychology is distorted but because their environment makes future consumption genuinely less likely.

Uncertainty about the stability of one’s own preferences also matters. If you are unsure what you will value in ten years, committing to a rigid long-term plan imposes costs. Some apparent present bias may reflect a rational reluctance to lock in choices based on preferences that might change for legitimate reasons.

Evolutionary arguments suggest that present bias may have been adaptive in ancestral environments where resources were scarce and perishable, threats were immediate and frequent, and the planning horizon was short. The discount function that served our ancestors in a world of uncertain foraging may be poorly calibrated for a world of 30-year mortgages and 40-year retirement horizons, but it is not irrational in the environment that shaped it.

These caveats do not invalidate the research on present bias. They sharpen it. The question is not whether people sometimes discount the future steeply — they do — but whether the steepness reflects the actual structure of their risks or an ingrained tendency that persists even when the environment has changed. In many modern contexts — saving for retirement in a stable democracy with a functioning financial system, for example — present bias probably does lead to choices that people themselves would prefer to avoid. But in other contexts, the same behavior may be the best available response to a genuinely hostile world.

The Architecture of Self-Control

The study of time inconsistency and present bias has yielded more than academic models. It has produced a practical insight that cuts across individual behavior, institutional design, and public policy: when people reliably fail to follow through on their own plans, the solution is often not willpower or information but architecture. Build the commitment into the structure. Make the patient choice the default. Put friction between the person and the temptation. Make the future consequences more vivid and the present rewards less accessible.

This is not a minor technical adjustment. It represents a genuine shift in how economics thinks about human agency. The exponential discounter is a unified self — consistent, forward-looking, in command. The present-biased agent is a divided self — a planner who makes promises and a doer who breaks them, a sequence of selves with overlapping but conflicting interests. Designing institutions for the divided self requires a different kind of economics, one that takes seriously the gap between intention and action.

Strotz saw this in 1955. Laibson formalized it in 1997. Thaler and Sunstein operationalized it in 2008. Kydland and Prescott applied the same logic to governments. The thread that connects all of them is the recognition that time changes more than the clock. It changes the person — or the institution — doing the choosing. The models that fit life are the ones that account for this.