RESEARCH DOSSIER
Time Inconsistency and Commitment in Consumer Choice
Abstract. This Research Dossier examines the microeconomic foundations of time inconsistency, preference reversal, and consumer commitment. It develops the exponential benchmark, quasi-hyperbolic preferences, sequential choice under sophistication and naivete, and alternative accounts of menu demand. Transparent analytical examples distinguish willingness to pay for commitment from the deterrent penalty and demonstrate the option value of liquidity under uncertainty. The study compares evidence from monetary and real-effort experiments, household saving, credit, subscriptions, smoking cessation, and exercise interventions. It identifies the distinct inferential requirements for preference estimation, commitment demand, causal intervention effects, and welfare evaluation. The evidence supports a mechanism-specific interpretation of consumer self-regulation and a conditional case for voluntary, credible commitment that accounts for income risk and flexibility. The analysis is a structured synthesis of established models and selected published research rather than original statistical estimation.
Keywords: intertemporal choice; dynamic inconsistency; present bias; commitment devices; quasi-hyperbolic discounting; household liquidity; consumer contracts; behavioral microeconomics; identification; welfare
The economic object of intertemporal conflict
Consumer choice unfolds through time. An individual determines a feasible course of consumption, saving, work, and expenditure, and at subsequent dates decides whether that course will be implemented. Time inconsistency arises when the relative ranking of two unchanged continuation plans depends on the date at which they are evaluated. Commitment is a deliberate alteration of future feasible actions or payoffs intended to influence implementation. These definitions separate a property of preference from a property of institutions.
The economic problem is visible in a household that wishes to preserve money for rent next month yet spends it on an immediately available purchase, or in a consumer who plans to cancel an expensive subscription but repeatedly delays cancellation. Such observations initiate investigation. They cannot by themselves identify a preference reversal: the household may discover new needs, face a change in income, revise expectations, or encounter a transaction cost. A scientifically useful diagnosis preserves the original information set and reconstructs the alternatives actually available at both dates.
Samuelson's (1937) note on utility measurement preceded the development of dynamic inconsistency models. The discounted utility framework became a way to represent dated consumption with an explicit intertemporal weighting rule. Strotz's foundational dynamic utility maximization paper (1955–1956) established the historical question of intertemporal inconsistency. The literature subsequently moved from abstract plans toward markets in illiquid assets, contract design, and experiments. The present dossier takes household-level intertemporal behavior as its unit of analysis. It treats national saving and monetary policy as contextual conditions rather than autonomous research objects. The comparative evidence is deliberately local: an experiment in a particular population supports a result in that setting, subject to its design and measurement, before any argument for transfer.
The analytical objective is to establish when present bias produces a reversal, when a commitment instrument changes the outcome, and when its cost can exceed its benefit. This involves three distinct judgments: the consistency of the model, the identification of the mechanism, and the normative choice of welfare criterion.
The exponential benchmark and conditional planning
Let dates be integers t through T; c_s is consumption at date s, u is increasing period utility, and delta lies in (0,1]. The exponential benchmark evaluates a deterministic consumption plan by U_t = sum from s=t to T of delta^(s-t) u(c_s). Preferences compare complete feasible streams rather than isolated monetary outcomes. A household also faces an intertemporal resource constraint, such as a_(s+1) = R_s(a_s + y_s - c_s), where a_s is beginning wealth, y_s is income, and R_s is the gross return on saving between dates. Borrowing constraints and nonnegative consumption complete the feasible set.
Suppose plans A and B agree through date k-1. Their utility difference at t before k equals delta^(k-t) times their utility difference as evaluated at k. Since this multiplicative factor is positive, the ordering is unchanged. This is the algebraic content of dynamic consistency in the benchmark. It requires unchanged preferences, state, and feasible continuations; a realized shock may change a decision while remaining fully consistent with the original contingent plan.
With Markov states x and exponential preferences the Bellman equation is V_t(x) = max over feasible a of {u(x,a) + delta E[V_(t+1)(x') | x,a]}. The recursion works because every continuation is discounted at the same constant relative rate. Under interior saving, differentiability, and deterministic return, the Euler condition is u'(c_t) = delta R_t u'(c_(t+1)). At a binding borrowing limit, with a lower bound on next-period assets, u'(c_t) >= delta R_t u'(c_(t+1)), so apparently impatient spending can emerge without a high subjective discount rate.
The benchmark has scientific value because it distinguishes several mechanisms that appear identical in simple choice records. A low delta can explain persistent impatience, and a low cash balance can explain urgent expenditure. Neither entails a reversal over a fixed pair of dated outcomes. The historical review by Frederick, Loewenstein, and O’Donoghue (2002), Time Discounting and Time Preference, documents the difficulty of interpreting diverse measured rates as one structural psychological parameter. The benchmark must therefore be stated with its assumptions rather than inferred merely from a market outcome.
Quasi-hyperbolic preferences and preference reversal
The discrete quasi-hyperbolic specification assigns date-t utility U_t = u(c_t) + beta sum from s=t+1 to T of delta^(s-t) u(c_s), with 0 < beta <= 1. Delta represents persistent discounting across future dates. Beta represents the additional relative weight of the current date. This is an established model in behavioral economics; the following derivations are illustrative applications.
Consider a payoff a at date 1 against payoff b at date 2, where utility is linear over these payoffs, both are certain, and neither changes any other consumption. At date 0 the consumer chooses b at date 2 if delta b > a, because beta delta multiplies both sides. At date 1 the consumer instead selects a if a > beta delta b. A strict reversal occurs precisely when beta delta b < a < delta b. The interval has positive width when beta < 1 and b > 0. If beta = 1, no such interval exists.
Take a = 85, b = 100, beta = 0.7, and delta = 0.95. At date 0, the two discounted utilities are 56.525 and 63.175, respectively. At date 1 the options are worth 85 and 66.5. The same pair changes rank without changes in resources or information. The example is a mathematical implication of the specification, not a measured household discount rate.
Continuous hyperbolic discounting, such as D(h) = 1/(1+kh) for k > 0, also produces decreasing impatience: D(h+1)/D(h) increases with h. Quasi-hyperbolic discounting concentrates the discontinuity between the present and all future dates, while the continuous function permits changing relative impatience over a broader horizon. Studies that use widely separated horizons may therefore distinguish the models only weakly. Laibson (1997), in Golden Eggs and Hyperbolic Discounting, connects a hyperbolic consumption problem to an illiquid asset that requires advance action before cash becomes available.
A calendar-based test must align dates with actual decision opportunities. If immediate payment means electronic delivery tonight in one arm but in-person payment at the laboratory in another, a change in convenience is bundled with the intended present effect. The theoretical reversal inequality is exact; empirical classification is conditional on the equivalence of the alternatives.
Sophistication, naivete, and sequential equilibrium
A present-biased decision maker can have different beliefs about future present bias. A sophisticated individual correctly predicts that future decisions will be governed by beta. A fully naive individual expects future decisions to respect the long-term plan, which can be represented in the simple specification by a perceived future beta of 1. Partial naivete permits intermediate perceived values. These are behavioral assumptions concerning prediction, not classifications of a person's intelligence.
Sophistication generates a game among successive temporal decision makers. At the final date the feasible action maximizes current utility. At an earlier date the consumer chooses the current action taking subsequent strategies as given; backward induction resolves a finite sequential problem with known states, actions, and payoffs. A dynamic choice policy then changes tomorrow's state as well as the current payoff. The date-zero optimum under perfect enforcement can differ from the equilibrium path even when every dated choice is individually optimal for its evaluator.
The central distinction in O’Donoghue and Rabin's (1999) Doing It Now or Later concerns the timing of a single activity with current costs or current rewards. Naivete can delay costly activities and accelerate rewarding ones; sophistication alters these patterns because an individual accounts for the subsequent choice. The result explains why self-knowledge is economically consequential but is not a guarantee of efficient timing.
For a numerical illustration, suppose task completion yields a benefit of 15 at date 3 if undertaken by date 2, with current disutility costs 6, 7, and 10 at dates 0, 1, and 2. Set beta = 0.6 and delta = 1. At date 2, completing has utility 0.6 × 15 − 10 = −1, so the consumer abandons the task. At date 1 a sophisticated consumer anticipates abandonment after delay, and immediate completion yields 2 rather than zero. At date 0, doing it now yields 3, whereas letting the date-1 self complete yields 0.6 × (15 − 7) = 4.8. The sophisticated consumer waits until date 1 and then completes. A fully naive consumer at date 1 expects date-2 completion; the anticipated utility of delay is 0.6 × (15 − 10) = 3, exceeding the immediate-completion utility of 2. At date 0 the naive consumer expects completion at date 1 and waits; at date 1 the consumer delays again, and at date 2 abandons. The same preferences and feasible actions thus produce different equilibrium paths through beliefs about future choice. The example is an original illustrative computation based on established theory, not a fitted empirical model.
Measurement should distinguish the date at which future behavior is predicted, the eventual decision opportunity, and the realized action. A consumer may accurately predict future impatience yet choose no commitment because protection is costly or because liquidity is essential.
Alternative mechanisms and competing explanations
Dynamic inconsistency is one mechanism within the broader economics of self-regulation. Gul and Pesendorfer's (2001) Temptation and Self-Control studies preferences over menus. A representation of their approach can be written W(A) = max_(x in A){u(x)+v(x)} - max_(y in A)v(y), where u describes commitment utility and v temptation utility. A menu containing a tempting alternative can reduce evaluation even if the consumer ultimately chooses the intended item, because resisting temptation itself matters. Observing a preference for a narrower menu therefore does not uniquely establish quasi-hyperbolic discounting.
State-dependent utility is another alternative. An individual deciding after work may experience fatigue, hunger, or an altered health condition. If a prior plan did not condition on those states, its revision cannot easily separate present bias from changes in instantaneous utility. Learning matters similarly: a household may revise a purchase plan when prices or expected income change. A model with stable preferences and updated beliefs can generate substantial variation in implemented choices.
Payment timing involves objective and subjective risk. Suppose a future reward has a probability q of delivery, and the utility of nondelivery is zero under linear utility. Its effective weight is qD(h). A consumer who trusts immediate payment more than delayed payment can display an apparent break at the present without beta below one. Andreoni and Sprenger (2012), in Risk Preferences Are Not Time Preferences, investigate the implications of uncontrolled risk for apparent present bias. An experimental payment guarantee must be credible to the participant, not merely to the experimenter.
The handbook chapter by Ericson and Laibson (2019), Intertemporal Choice, places discounting beside temptation, forecasting and time perception. These alternatives have different intervention implications. Credible payment eliminates one form of uncertainty; reminders address attention; insurance addresses income risk; a binding restriction addresses a strategic conflict with anticipated future choice. A universal commitment intervention would obscure these distinctions.
Counterexamples are central to inference. A consumer who rationally buys medicine today rather than waiting for a discount may value urgent relief. A household may hold cash despite higher returns on locked savings because an uninsured emergency would be costly. Neither decision is anomalous under appropriately specified preferences and constraints. Present bias becomes an explanatory claim only when alternative causal pathways have been constrained or separately measured.
Commitment as an economic institution
A commitment device modifies a later choice set, a relative price, or the information environment to make an earlier intention easier to execute. A hard commitment either removes an action or imposes a material penalty for taking it. A softer commitment relies on reminders, social accountability, salience, or reputational consequences. The boundary is economic rather than verbal: the device must alter expected behavior through a credible mechanism.
Bryan, Karlan, and Nelson (2010), in Commitment Devices, review informal and formal interventions and their policy design. The category includes institutions with quite different enforcement arrangements. A locked bank account is a contract over withdrawals. A self-funded deposit contract makes a payment contingent on an observable outcome. A publicly announced intention depends on social consequences. A reminder is ordinarily an attention intervention; it becomes a commitment instrument only insofar as it realizes a prior deliberate constraint or consequence.
Return to the reversal example where a = 85 and b = 100. At date 0 the unconstrained equilibrium path switches at date 1 to a, worth 56.525 in date-zero utility. Enforced waiting is worth 63.175. Under the stated assumptions, the maximum date-zero utility cost for perfect commitment is therefore 6.65. If a fee F is paid immediately and utility is linear, commitment is chosen when F < 6.65, with indifference at equality. If the device instead permits early choice for a penalty p levied at date 1, then a-p <= beta delta b is required to prevent the reversal (with indifference resolved in favor of waiting). Here p >= 18.5. The commitment fee ceiling and the deterrent penalty threshold measure different things. A penalty that is never incurred on the equilibrium path is not an immediate cost, yet its existence can deter a deviation.
The mechanism requires credible enforcement. A participant with unrestricted power to reverse the restriction at no cost has merely expressed an intention. Conversely, stringent irrevocability may become expensive when circumstances change. The economically relevant contract terms are enforcement probability, penalty size, timing, eligible exceptions, and the household's alternatives after a shock.
Liquidity and constrained household saving
The allocation of wealth between liquid and illiquid claims is a choice about future access to resources. Liquid savings are available for consumption, credit obligations, and unexpected costs. Illiquid balances can protect earmarked assets but may create hardship when regular income fails. The same household may therefore demand both forms of saving.
Laibson's (1997) illiquid asset analysis gives a formal foundation for this distinction. In that model, the timing of asset conversion prevents instant consumption financed by the illiquid claim. A market return alone cannot characterize the asset's economic function: its conversion technology also enters the future feasible set. This matters for the interpretation of observed consumption responses because current income and assets need not be interchangeable in the decision problem.
An illustrative household has 300 units available at the start of a month, of which 200 is intended for a future rent payment. It locks 200 and retains 100. With probability 0.2 it faces a necessary expenditure of 160 before the lock expires. If no affordable outside liquidity is available, the locked account creates a shortfall of 60 in that state. Expected missing liquidity is 12 units, but expected utility loss must incorporate the fact that the missing expenditure is necessary and marginal utility is state dependent. Comparison with the flexible allocation requires the probability that the unlocked earmarked funds would be spent prematurely and the resulting future rent shortfall. Account balance alone measures neither quantity.
In the limiting case of perfect insurance and reliable enforcement, the option value of liquidity can be small, leaving more room for commitment. Under high income volatility, weak insurance, or expensive emergency credit, that option value rises. A household may rationally decline an otherwise beneficial product. Aggregate take-up therefore depends on both the value of self-control and the cost of lost flexibility.
Beshears and colleagues (2020), early withdrawal penalty design, directly pose the question of which early withdrawal penalty attracts deposits. The research question is important because penalty severity is an economic variable, rather than a fixed property of all commitment accounts. This dossier makes no numerical claim about that study's results because its accessible publisher record did not permit detailed verification.
Identifying the mechanism in experiments
A credible experiment measures the same underlying trade-off across time while controlling other dimensions of choice. For a fixed dated pair of rewards, repeated decisions can reveal actual changes of ranking. For choices made at one date, a front-end delay compares a near-term pair with a future pair separated by the same interval. The latter identifies an interaction of delay and horizon only under a maintained model; it is not a direct observation of a reversal.
In binary monetary elicitation, an individual selects an amount today or a greater delayed amount. Under linear monetary utility and certain payment, the switch point identifies a discount factor. With concave utility, the switch point also reflects curvature. With consumption smoothing, an experimental payment can affect consumption over several dates; the independence of dated utility assumed by the calculation fails. The measured outcome is a choice over cash transfers, which is not necessarily a choice over dated consumption.
Andreoni and Sprenger's (2012) convex budget experiment jointly estimates curvature and discounting from allocations across dates and reports dynamically consistent time preferences in its setting. A convex budget allows more than binary choices, which offers additional information about marginal substitution under specified utility. Identification still depends on the mapping between experimental rewards and consumption, on background wealth, and on the credibility of delayed delivery. Researchers should report how corner choices, multiple optima, inconsistent responses, and estimated background consumption are treated.
In a longitudinal effort design, the economic object is more directly tied to the timing of a real burden. Augenblick, Niederle, and Sprenger (2015), Working over Time, compare monetary and real-effort choices. They report substantially more present bias in effort than money, and the effort measure predicts demand for a meaningful commitment device. The two findings can coexist with the convex-budget result because the outcome domain and payment-to-utility mapping differ.
A useful design records both the planned allocation and the revised allocation over an unchanged pair of effort dates, holds remuneration constant, and separately measures demand for a binding commitment. Even such a design requires attention to task learning, fatigue, experimenter demand, and unobserved changes in the outside schedule. The key empirical gain comes from observing distinct implications of the mechanism rather than relying on a single structural parameter.
Randomized offers, selection, and causal estimands
Random assignment to an offer identifies the causal effect of being offered a commitment product under valid assignment, consistency, and an appropriate analysis of missing outcomes. Let Z be randomized offer status, D whether an individual adopts the product, and Y the outcome. The intention-to-treat estimand is E[Y|Z=1] - E[Y|Z=0]. Comparing D=1 with D=0 will generally mix treatment and selection because adoption is a choice.
An instrumental-variable interpretation of the ratio of the intention-to-treat effect to E[D|Z=1]-E[D|Z=0] requires a nonzero first stage, independence, exclusion, monotonicity, and a well-defined treatment. The result is local to compliers. Exclusion can fail when the offer itself teaches saving techniques or changes attention without adoption. A zero or small offer effect can therefore reflect limited adoption or other channels rather than an ineffective product for every individual.
Ashraf, Karlan, and Yin's (2006) Philippine commitment savings experiment reports an offer randomized among clients of a bank. Among 710 clients offered the product, 202 opened accounts, approximately 28.4 percent. The bank-client baseline sample was 1,777. These counts describe product demand among an offered sample, rather than a population estimate of present bias. A baseline time-preference measure predicted take-up among women in the reported analysis; that association is distinct from the causal effect of the randomized offer.
The appropriate outcome vector extends beyond the targeted account balance. Other savings, borrowing, consumption, emergency expenditure, forfeited funds, and financial distress can move in response to a restriction. A treatment may increase earmarked deposits through substitution from another account while leaving total wealth unchanged. Household financial welfare requires broader measurement.
Randomization can also operate at the household, workplace, or group level. Correlated outcomes require standard errors compatible with assignment and interference. A commitment contract may affect family transfers or co-workers' schedules, violating a no-interference assumption. Predefined estimands, attrition analysis, and appropriately clustered inference protect the interpretation more effectively than a larger collection of loosely related outcome tests.
Evidence from contracts, debt, and exercise
Market contracts create settings in which stated intentions, realized use, and renewal decisions can be compared. DellaVigna and Malmendier (2006), Paying Not to Go to the Gym, analyze three US health clubs and 7,752 members over three years. The reported monthly-flat-fee users attended about 4.3 times per month despite a fee above $70 and a ten-visit alternative costing $10 per visit. These quantities describe the studied membership choices, not today's gym market. At exactly $70, arithmetic break-even relative to $10 per visit is seven visits per month; at a higher fee it is higher than seven. Realized average attendance alone does not prove that every subscription was individually disadvantageous, because service differences, stochastic demand, and convenience are relevant.
The authors also examine renewal and cancellation behavior and consider overconfidence about future self-control or efficiency. A subscription can be privately useful as a prepaid incentive, yet automatic continuation can impose costs when a user overestimates future cancellation. A valid contract analysis values all terms and uses pre-purchase forecasts when available.
Meier and Sprenger (2010), Credit Card Borrowing, match elicited time preferences to credit and tax information and report a conditional association between measured present bias and credit card borrowing. The association survives reported controls for income, demographic variables, and credit constraints. This is evidence about correlation, not a randomized causal effect of present bias on debt. Reverse causation, unobserved income shocks, and measurement error retain analytical relevance.
Royer, Stehr, and Sydnor (2015), workplace exercise commitments, study incentives and self-funded commitments in a workplace experiment. The initial incentive alone produced small persistent effects; the additional offer of a commitment contract was associated with longer-lasting changes. The relevant causal contrast concerns the intervention designs, rather than the causal effect of signing among self-selected adopters. This also illustrates a dynamic feedback: behavior induced by a temporary contract can change subsequent habits or perceived costs.
Taken together, the studies support heterogeneous patterns across financial choices, subscriptions, and health-related effort. The object of measurement changes with the market. A contractual payment flow is easy to observe, but the expected utility of activity, liquidity, and self-control is more difficult to recover.
International evidence and the boundaries of transfer
Giné, Karlan, and Zinman (2010) report that approximately 11 percent of smokers offered their voluntary commitment product took it up. The offer raised the probability of passing the six-month biochemical test by about three percentage points, with an effect also observed at unannounced 12-month testing. These are offer effects measured using a specific outcome; neither the take-up percentage nor the test result directly identifies the preference mechanism.
The international scope of this dossier follows the provenance of the studies rather than an assumed universal effect. The Philippine bank experiment studied an actual deposit product offered to existing or former bank clients. The separate CARES smoking cessation study by Giné, Karlan, and Zinman (2010) involved adult smokers in the Philippines. The intervention allowed voluntary deposits that could be forfeited following unsuccessful cessation under specified conditions. The underlying article is Giné, Karlan, and Zinman (2010).
The two Philippine applications are analytically different. Saving commitment restricts access to an asset, whereas the cessation intervention makes return of deposited money conditional on a health-related outcome. The latter changes incentives through forfeiture; the former primarily changes the availability of balances. Their target outcomes and risks cannot be combined into one effect size. A loss of deposited savings may have substantial welfare consequences even if a health intervention increases the probability of achieving a target.
The US gym records involve private subscription contracts, while the workplace exercise experiment embeds a facility and intervention within employment. The bank and workplace institutions differ in enforcement costs, income regularity, alternative financial services, and social relations. Transfer between countries requires knowing which mechanism is intended to persist across such differences.
A household's access to formal accounts, trust in enforcement, transaction costs, and exposure to shocks determine whether a contract can operate effectively. Institutional comparisons should specify at least the legal feasibility of withdrawal restrictions, consumer protection, deposit risk, repayment alternatives, and practical access to emergency funds. These are moderators to measure, not grounds to assign a representative discount parameter to an entire region.
The verified comparative corpus includes selected US and Philippine settings, agricultural savings in Malawi, and laboratory and field choice experiments. It supplies a mechanism-based international analysis but does not estimate global prevalence, characterize all continents, or constitute a harmonized cross-country meta-analysis. Generalization of the qualitative mechanism is a theoretical question; prediction of quantitative effects in other populations requires further evidence.
Commitment savings and agricultural seasonality
The randomized Philippine savings experiment of Ashraf, Karlan, and Yin (2006) identified a difference between the offer of a restricted account and ordinary access to deposit services. Restrictions were tied to a target balance or date, and the commitment account paid the ordinary interest rate. Approximately 28.4 percent of the 710 clients offered the product opened one. Twelve months later the offer effect on balances at the study bank was about 411 pesos, about 81–82 percent relative to baseline balances in the authors’ summaries. The approximately 411-peso estimate (Tables VI and IX) is significant at the 10 percent level. The estimate is an intention-to-treat effect of access to a particular product. It does not measure the causal effect of adoption on the adopters, the household's overall net worth, or a structural present-bias parameter.
Brune, Giné, Goldberg, and Yang (2016) studied a related but institutionally different problem in agricultural savings in Malawi. Farmers receive funds seasonally but must pay for subsequent inputs. Savings facilities can bridge those dates, and a commitment restriction can preserve funds earmarked for the next planting season. The study includes an ordinary-account treatment and a treatment with access to a commitment account. In Table 5, the estimated effects on savings balances across accounts on October 22, 2009 were 2,475 Malawi kwacha for commitment access and 1,301 for ordinary access; the reported equality test gives p = .019. This is a comparison of assigned offers. It is not a comparison of individuals selected into contracts, and the distinction between the effects of the two treatment arms varies across outcomes.
Seasonal receipts create a stronger connection between the commitment horizon and a measurable production opportunity than do arbitrary calendar restrictions. Nevertheless, higher recorded balances might come from diverted cash, reduced transfers to other households, or lower consumption. The depositor's options also depend on when funds become available, the ability to meet emergencies, and the viability of the planned investment. The ordinary-account treatment is scientifically valuable because it allows some features of savings access to be distinguished from the additional restriction.
Cross-country comparison therefore concerns identifiable institutional margins. The Philippines evidence concerns clients of a particular bank; the Malawi experiment concerns agricultural producers with seasonal payment patterns. Both establish that particular savings offers can change measured behavior. Their treatment magnitudes have different denominations, baseline balances, and economic environments. Converting currencies does not make them estimates of a common parameter.
The underlying analytical lesson is that a commitment treatment must be described as an offer with a defined choice set, enforcement rule, and population. When a package combines account access, payment facilitation, information, and withdrawal rules, the estimated effect identifies that package. A theory of present bias can explain why restrictions are valued, but competing explanations such as protection from social claims and earmarking must remain in the empirical comparison.
Future wage increases and adjustable commitment
Thaler and Benartzi's (2004) Save More Tomorrow program attached a future increase in employees' retirement contributions to subsequent pay increases. In the initial implementation, 78 percent of employees offered the program enrolled, 80 percent of those enrollees remained through the fourth pay raise, and their average contribution rate increased from 3.5 to 13.6 percent over 40 months. These are implementation statistics for participating employees, not randomized treatment effects or estimates of total household saving.
A future-dated increase avoids an immediate reduction in take-home pay at enrollment. As earnings rise, the contribution schedule changes automatically unless the employee elects to leave the program. The mechanism combines a prior authorization, timing, persistence, and an option to exit. It differs from an irrevocably locked account. In a simple two-date example, an employee earning 100 and contributing 4 takes home 96 before taxes. If earnings increase to 110 and the new contribution is 7, take-home pay increases to 103 despite the higher saving rate. This mechanical property can reduce the immediate perceived cost of raising contributions.
The relevant counterfactual is what employees would have saved absent the arrangement, including outside accounts and debt. A higher payroll contribution can coincide with new borrowing or reductions in other saving. An employer-based program may also be inaccessible to informal workers or those with irregular earnings. The extent of voluntary enrollment, the cost and clarity of exit, and the ability to change contributions in emergencies are properties of the institution itself.
The design illustrates an important distinction between binding contractual penalties and persistence produced by a prior instruction. The latter can offer meaningful assistance even when no penalty is imposed. A comparison of commitment designs should measure not just rigidity but enactment, reversibility, and the degree to which a consumer needs to take repeated affirmative actions to maintain a chosen plan.
Consumption domains and recent measurement evidence
Money is a transferable resource, whereas a meal or a unit of effort is itself an object of consumption or disutility. Experimental payment dates are therefore not automatically consumption dates. Even a carefully implemented laboratory choice over money might capture arbitrage opportunities, liquidity needs, or expected ability to borrow.
Danzer and Zeidler (2024), in the provisional IZA discussion paper Present Bias in Choices over Food and Money, compare food and monetary choices in a college canteen setting. The study involves 73 participants and 3,666 food-item decisions. Its authors report no significant change in aggregate meal healthiness between advance and immediate decisions, although the balance of choices across categories changes. They find no significant correlation between their monetary and food measures of dynamic inconsistency. Their analysis proposes that commitment uptake can coexist with internal self-control. As a discussion paper, the work remains provisional; its limited sample, domain-specific measures and design should constrain generalization.
The unit of observation matters. Repeated dish choices provide information about within-person behavior, but thousands of choices do not constitute thousands of independent consumers. If item-level errors correlate within participants, inferential precision depends on the dependence structure. Furthermore, aggregate nutritional quality is a constructed outcome whose interpretation depends on its scoring rule. A reversal concerning dessert can coexist with stable meal-level quality when another category adjusts. This cautions against treating one revealed outcome as a complete measure of intertemporal preferences.
Consumption decisions are multidimensional. Consider two dated meals, each consisting of a main dish and a dessert. If an immediate self prefers a higher-calorie main dish but compensates with a lower-calorie dessert, an aggregate calorie total may be constant while component rankings change. A stable aggregate outcome alone does not distinguish commitment, internal substitution, or indifference over bundles. The proper model includes the feasible bundle and the attributes of its components.
Together with the monetary and real-effort studies considered earlier, the food experiment motivates a research design that estimates task-specific choice rules and then tests cross-domain restrictions. A single scalar parameter is a substantive hypothesis. Its usefulness depends on predictive stability across contexts, not on the convenience of reporting a common behavioral coefficient.
Welfare, autonomy, and the commitment trade-off
Welfare evaluation becomes substantive when rankings differ across dates. The date-zero evaluator values the protection of a long-term plan, the date-one evaluator values access to immediate consumption, and an experienced-utility criterion may aggregate realized benefits and costs according to a stated rule. The selection of a criterion is a normative decision. There is no algebraic theorem that all improvements for the earlier evaluator improve all other evaluations.
For a transparent calculation, consider an anticipated commitment gain G measured in date-zero utility in states where the consumer would otherwise deviate. Let the restriction impose loss H in unforeseen emergencies and upfront utility cost F. If the two states are disjoint and exhaustive with emergency probability q, the expected difference under the stated evaluator is (1-q)G - qH - F. Commitment is beneficial exactly when this quantity is positive. With G=20, H=80, F=0, the break-even emergency probability is 0.2. A riskier population can have lower expected net benefit from the same device even if its self-control conflict is equally severe. The calculation is an illustrative counterfactual, not an estimated welfare function.
The cost term includes foregone flexibility, contract fees, forfeiture, enforcement errors, and downstream borrowing. The benefit term includes realized long-term consumption or health improvements and, where the model warrants it, relief from temptation or effort spent resisting. Each element must be expressed in commensurable units and assigned to the relevant evaluator. A larger nominal balance is not itself a complete measure of welfare.
Voluntary choice provides information about expected value under the consumer's beliefs, but take-up does not establish realized benefit. A naive consumer may refuse a useful device due to inaccurate predictions. A sophisticated consumer may decline a restrictive contract because the option value of liquidity is high. Conversely, inaccurate beliefs about achievement can generate excessive uptake of a costly penalty contract. Policy should account for these possibilities without assuming all deviations reflect faulty judgment.
Autonomy includes the ability to undertake voluntary, intelligible future constraints. It also includes the practical ability to respond to emergencies, contest errors, and understand the consequences of a contract. A welfare assessment should therefore state the decision criterion, compare feasible alternatives, measure adverse outcomes, and describe the distribution of benefits and costs.
Market organization and contract design
A competitive supplier of commitment products can earn revenue by serving consumers with demand for credible restrictions. Another supplier can profit from predictable delay in cancellation, missed deadlines, or the purchase of an unnecessarily expensive plan. Both can advertise convenience and future benefits. Their incentives differ with fee schedules, switching costs, and contractual enforcement.
Consider a subscription priced F per period and an outside per-use charge p. Expected visits n make the simple price-only break-even condition F = pn. When perceived n differs systematically from realized n, the contract menu can direct users toward a costlier arrangement under their own initial comparison. To diagnose systematic overprediction, one needs ex ante forecasts and the full service and price menu, not merely realized usage. For the economic implications of such discrepancies, DellaVigna and Malmendier's (2006) health club contract evidence is especially relevant.
A beneficial commitment product is easier to evaluate when the provider discloses the trigger, amount, duration, exceptions, renewal mechanism, and treatment of forfeitures. Objective verification can reduce disagreement but introduces privacy and measurement costs. An exercise contract based on facility entry can mismeasure meaningful physical activity. A savings penalty can be enforced precisely yet punish a necessary withdrawal. Better observability is not equivalent to better welfare.
Institutional design can offer a menu of liquid savings, notice-period withdrawals, bounded penalties, and stronger restrictions. Consumers may sort according to their expectations, liquidity needs, and beliefs about their future decisions. Sorting complicates observational inference but provides useful data when paired with randomized menu variation and measured outcomes. A comparison should track both deposits and total household balance sheets.
Public interventions may target information, defaults, disclosure, access to low-cost commitment, and protection from exploitative terms. The precise legal setting varies by jurisdiction; this monograph makes no claim about current statutory obligations. The economically relevant question is whether a change in the contractual environment improves implementation under an explicitly defended welfare criterion after accounting for flexibility and error costs.
Household heterogeneity, measurement, and robustness
Households differ in income timing, access to borrowing, baseline assets, transaction costs, household membership, and the reliability of expectations. A consumer-level preference estimate may not explain a household-level balance when multiple people control expenditure. A partner may constrain access to resources or pool risk, and an account designated as commitment may function primarily as protection against other household members' claims. Studies should distinguish intrapersonal conflict from interpersonal bargaining.
A credible panel should record measurement dates, proposed cash or effort dates, elicited beliefs, actual actions, and changes in the feasible set. For income and wealth, both stocks and flows require consistent reference periods and currency units. Administrative linkage may reduce recall error but introduces consent, matching, and selection issues. Surveys of intended spending are useful for prediction but are not direct substitutes for binding choices.
Robustness checks begin with the timing variable. Was the immediate option actually available immediately? Were delayed rewards paid in the same channel, with comparable collection costs and confidence? Were participants offered equivalent risk? Were estimates sensitive to utility curvature or background consumption? The convex budget methodology and real effort evidence offer complementary ways to expose the consequences of these choices.
Researchers should also report the distributions of estimates rather than only means. Pooling heterogeneous individuals may produce a smooth aggregate discount function even when individual discounting is stable. Suppose half the population has delta = 0.5 and half delta = 0.95. Their mean discount factors at horizons one and two are 0.725 and 0.57625. The ratio at horizon two relative to one is approximately 0.795, larger than 0.725. This is decreasing impatience in the averaged factor, despite exponential discounting by both individuals. Aggregation thus supplies a concrete counterexample to inferring individual dynamic inconsistency from a population discount curve.
No original large-scale microdata analysis is undertaken here. The study-level measures and design facts remain tied to the source populations. The examples are reproducible algebra; they do not carry sampling standard errors or significance claims. A meta-analysis would require comparable effect definitions, independent sample information, uncertainty measures, and a protocol for combining distinct interventions.
Analytical synthesis and research priorities
The original synthesis of this dossier separates a chain of economic propositions that are often compressed into a single statement about self-control. A reversal is a change of ranking of fixed continuations. An expectation error is a discrepancy between a prediction and its subsequent realization conditional on the state. Commitment demand is a choice over future menus or payoffs. Intervention effectiveness is a counterfactual difference in behavior or outcomes. Welfare improvement is an evaluation of the resulting outcome against a specified criterion. Each proposition requires its own evidence and several can hold without the others.
The formal analysis establishes three conditional findings. First, the exponential benchmark preserves continuation rankings under unchanged conditions, whereas beta-delta preferences admit the reversal interval beta delta b < a < delta b. Second, the price a consumer is willing to pay for intended protection differs from the penalty necessary to deter a future deviation. Third, a restriction's expected benefit can decline with the probability and severity of an emergency even when the self-control mechanism remains fixed. These findings are derived within established economic specifications and no novel named theory is claimed.
The empirical synthesis supports treating money, effort, and consumption as economically distinct. The convex budget study and the longitudinal effort study provide a direct contrast in measured patterns. Their difference does not establish that all monetary tasks lack present bias or all effort tasks exhibit it. Rather, it illustrates that the mapping from dated rewards to utility is part of identification. The commitment bank experiment then demonstrates how a particular demand question can be investigated using a real financial institution.
A future research design could combine a panel of repeated fixed-date choices, ex ante beliefs about future decisions, randomized changes in commitment terms, and comprehensive household outcomes. Directly observed forecasts and later actions allow prediction errors to be separated from preference changes. A randomized offer establishes the policy estimand. The balance-sheet and expenditure panel measures substitution and emergency costs. Prespecified heterogeneity splits by income volatility, access to credit, and baseline assets would help establish where flexibility has the highest value. Such a design must account for participant loss, statistical power, spillovers, and privacy.
Priority questions follow from the evidence: whether the same person exhibits similar reversals across money and effort under comparable state control; how expectations change after direct experience with a contract; which types of commitment reduce costly deviations without shifting distress elsewhere; and how legal and financial institutions influence enforcement and exit. These remain empirical research questions. The contribution presented here is an integrated argument about the conditions of inference, derived examples, and the institutional trade-off between protection and flexibility rather than a claim to have discovered a new behavioral law.
Conclusion
Time inconsistency is a property of dated evaluation. Commitment is an institutional response that changes later opportunities or incentives. The quasi-hyperbolic specification makes their relationship formally transparent, while the exponential benchmark clarifies why shocks, constraints, and impatience alone do not establish dynamic inconsistency. Sophistication changes the equilibrium consequences of present bias through expectations about subsequent choices.
Empirical studies report informative results in selected domains. Real-effort experiments, monetary allocations, bank offers, credit records, health-club contracts, and workplace interventions measure different parts of the intertemporal choice problem. Their results are properly understood through their particular outcomes, samples, and identification conditions. Causal estimates of an offer must remain distinct from descriptive correlations and selected adopter comparisons.
The welfare value of commitment depends on the avoided cost of a future deviation, the credibility and price of enforcement, and the value of flexibility under uncertainty. A consumer may benefit from access to commitment without benefiting from the most restrictive available contract. An intervention that increases target behavior requires further examination of costs, displacement, and autonomy before it can be characterized as welfare enhancing. These implications follow from the established models and the bounded evidence reviewed in this dossier. They provide a research-oriented foundation for the economic analysis of consumer self-regulation within Microeconomics.
References
Strotz, R. H. (1955–1956). Myopia and inconsistency in dynamic utility maximization. Review of Economic Studies, 23(3), 165–180. DOI: 10.2307/2295722. Myopia and inconsistency in dynamic utility maximization
Laibson, D. (1997). Golden eggs and hyperbolic discounting. Quarterly Journal of Economics, 112(2), 443–477. DOI: 10.1162/003355397555253. Golden eggs and hyperbolic discounting
O’Donoghue, T., & Rabin, M. (1999). Doing it now or later. American Economic Review, 89(1), 103–124. DOI: 10.1257/aer.89.1.103. Doing it now or later
Frederick, S., Loewenstein, G., & O’Donoghue, T. (2002). Time discounting and time preference: A critical review. Journal of Economic Literature, 40(2), 351–401. DOI: 10.1257/002205102320161311. Time discounting and time preference: A critical review
Gul, F., & Pesendorfer, W. (2001). Temptation and self-control. Econometrica, 69(6), 1403–1435. DOI: 10.1111/1468-0262.00252. Temptation and self-control
Ashraf, N., Karlan, D., & Yin, W. (2006). Tying Odysseus to the mast: Evidence from a commitment savings product in the Philippines. Quarterly Journal of Economics, 121(2), 635–672. DOI: 10.1162/qjec.2006.121.2.635. Tying Odysseus to the mast: Evidence from a commitment savings product in the Philippines
DellaVigna, S., & Malmendier, U. (2006). Paying not to go to the gym. American Economic Review, 96(3), 694–719. DOI: 10.1257/aer.96.3.694. Paying not to go to the gym
Meier, S., & Sprenger, C. (2010). Present-biased preferences and credit card borrowing. American Economic Journal: Applied Economics, 2(1), 193–210. DOI: 10.1257/app.2.1.193. Present-biased preferences and credit card borrowing
Andreoni, J., & Sprenger, C. (2012). Estimating time preferences from convex budgets. American Economic Review, 102(7), 3333–3356. DOI: 10.1257/aer.102.7.3333. Estimating time preferences from convex budgets
Andreoni, J., & Sprenger, C. (2012). Risk preferences are not time preferences. American Economic Review, 102(7), 3357–3376. DOI: 10.1257/aer.102.7.3357. Risk preferences are not time preferences
Augenblick, N., Niederle, M., & Sprenger, C. (2015). Working over time: Dynamic inconsistency in real effort tasks. Quarterly Journal of Economics, 130(3), 1067–1115. DOI: 10.1093/qje/qjv020. Working over time: Dynamic inconsistency in real effort tasks
Royer, H., Stehr, M., & Sydnor, J. (2015). Incentives, commitments, and habit formation in exercise: Evidence from a field experiment with workers at a Fortune-500 company. American Economic Journal: Applied Economics, 7(3), 51–84. DOI: 10.1257/app.20130327. Incentives, commitments, and habit formation in exercise: Evidence from a field experiment with workers at a Fortune-500 company
Giné, X., Karlan, D., & Zinman, J. (2010). Put your money where your butt is: A commitment contract for smoking cessation. American Economic Journal: Applied Economics, 2(4), 213–235. DOI: 10.1257/app.2.4.213. Put your money where your butt is: A commitment contract for smoking cessation
Bryan, G., Karlan, D., & Nelson, S. (2010). Commitment devices. Annual Review of Economics, 2, 671–698. DOI: 10.1146/annurev.economics.102308.124324. Commitment devices
Ericson, K. M., & Laibson, D. (2019). Intertemporal choice. In Handbook of Behavioral Economics: Applications and Foundations 1, Vol. 2, pp. 1–67. Elsevier. Intertemporal choice
Beshears, J., Choi, J. J., Harris, C., Laibson, D., Madrian, B. C., & Sakong, J. (2020). Which early withdrawal penalty attracts the most deposits to a commitment savings account? Journal of Public Economics, 183, 104144. DOI: 10.1016/j.jpubeco.2020.104144. Which early withdrawal penalty attracts the most deposits to a commitment savings account? Journal of Public Economics, 183, 104144
Brune, L., Giné, X., Goldberg, J., & Yang, D. (2016). Facilitating savings for agriculture: Field experimental evidence from Malawi. Economic Development and Cultural Change, 64(2), 187–220. DOI: 10.1086/684014. Facilitating savings for agriculture: Field experimental evidence from Malawi
Thaler, R. H., & Benartzi, S. (2004). Save More Tomorrow: Using behavioral economics to increase employee saving. Journal of Political Economy, 112(S1), S164–S187. DOI: 10.1086/380085. Save More Tomorrow: Using behavioral economics to increase employee saving
Danzer, A. M., & Zeidler, H. (2024). Present bias in choices over food and money. IZA Discussion Paper No. 17415. October 2024. Discussion paper
Samuelson, P. A. (1937). A note on measurement of utility. Review of Economic Studies, 4(2), 155–161. DOI: 10.2307/2967612. A note on measurement of utility