Open data · CC BY 4.0 what each one claims, and what replicated no ads · no tracking
Bias Atlascognitive biases

Hyperbolic discounting

also known as — Present bias · Quasi-hyperbolic discounting · Beta-delta discounting · Dynamic inconsistency · Delay discounting
People discount the near future more steeply than the far future, so a reward available now can outweigh a larger one later that they would have chosen when both were distant.
Does it replicate? Mixed Two meta-analyses disagree about money, giving beta of 0.97 with 1 not rejected and 0.938 with 1 excluded, and both find more present bias for non-monetary rewards.
VerdictMixed
First published1975
FieldDecision and judgement
Sources18
Last checked2026-08-17
What it claims

What does Hyperbolic discounting mean?

Exponential discounting applies the same rate to every interval, so a choice made today between two future dates survives the wait. Hyperbolic discounting applies a rate that falls as the horizon lengthens, so the ranking flips once the sooner reward comes into reach. Someone may prefer $110 in 31 days to $100 in 30 days and still take $100 now over $110 tomorrow.

Economic work commonly summarises that pattern with a simplified version, the quasi-hyperbolic or beta-delta form, in which one parameter discounts everything except the present and ordinary exponential discounting handles the rest. Beta below 1 is present bias and beta of 1 is dynamic consistency. That parameter, rather than the shape of any curve, is what the experiments test.

Verdict

Repeats disagree, or the effect holds in some conditions and not others. The other 248.

The field
43 of the 85 entries in decision and judgement share this verdict.
Does it replicate?

Has Hyperbolic discounting been retested?

Mixed Two meta-analyses disagree about money, giving beta of 0.97 with 1 not rejected and 0.938 with 1 excluded, and both find more present bias for non-monetary rewards.

Andreoni and Sprenger gave 97 subjects a convex time budget, splitting an allocation between a sooner and a later date across 45 decisions, and equalised the transaction costs and the delivery risk of the two payments. They estimated beta at 1.004, standard error 0.002, alongside an annual discount rate of 0.300. The rejection of beta = 1 was marginal, F(1,96) = 2.82, p < 0.10, and the favoured alternative was future bias.

Augenblick, Niederle and Sprenger then ran the same allocation over real work, seven weeks of Greek transcription and Tetris. The 80 subjects with usable effort data gave beta = 0.888, standard error 0.033, and the 75 of them with complete monetary data gave 0.974, standard error 0.009, with the two rejected as equal, chi-squared(1) = 6.37, p = .01.

Their between-subjects repeat recruited 200 further Berkeley students, of whom 194 finished, and the estimates rest on 99 money and 95 effort cases. It gave 0.997 for money and 0.892 for effort. There the effort estimate only just cleared the line, chi-squared(1) = 3.73, p = .05, and money and effort were no longer separable at the 5% level, chi-squared(1) = 3.50, p = .06.

Offered a commitment device at a price of zero, 47 of 80 subjects took it, raising from 0.1 to 0.9 the chance that their earlier allocation counted. Present bias over effort predicted that choice, though the raw correlation was 0.225, p = .04, and explained about 5% of the variance. Augenblick and Rabin recruited 100 participants for a seven-week study of task loads, of whom 79 finished and 72 allowed individual estimation.

Participants wanted 10 to 12 percent fewer tasks in the present than at any future date, and the aggregate beta ran from 0.812 to 0.835 across five specifications. Imai, Rutter and Camerer collected 220 estimates from 31 studies in 28 articles using the convex time budget, then averaged 217 of them after dropping three influential observations. The random-effects average is 0.9663, standard error 0.0147, p = .0297 against beta = 1, and the multi-level average is 0.9518.

The 193 monetary estimates give 0.9723 and 0.9758, neither distinguishable from 1. The 24 effort estimates, drawn from nine studies, give 0.8815 and 0.8802, p = .0001 and p = .0004. Cheung, Tymula and Wang then published a larger meta-analysis, of 86 studies, and reached a different answer for money: 0.938 after correcting for selective reporting, 95% interval 0.905 to 0.972, with non-monetary rewards at 0.750, interval 0.643 to 0.857.

Only their abstract could be obtained. Money is not settled either. Balakrishnan, Haushofer and Jakiela paid 494 participants in Kenya through M-Pesa and found beta between 0.902 and 0.924, but only where the same-day payment landed immediately after the session. Where it landed at the close of business the estimates ran 0.982 to 0.992 and 1 could not be rejected.

Ruggeri and colleagues put five intertemporal choice anomalies to 13,629 people in 61 countries on a preregistered plan. The choices were hypothetical, and their present-bias item was a single question between 12 and 24 months rather than a test against an immediate reward. Rates for the four main anomalies ran from 13.8% for absolute magnitude to 40.1% for gain-loss asymmetry, 54% of participants showed at least one, and every anomaly appeared in every country.

Read off Imai, T., Rutter, T. A., & Camerer, C. F. (2021). Meta-analysis of present-bias estimation using convex time budgets. The Economic Journal, 131(636), 1788-1814.

Examples

What are some examples of Hyperbolic discounting?

The class booked in Januaryillustration

In January a course starting in March looks obviously worth the two evenings a week. On the March morning, with the alarm going and rain on the window, the same two evenings lose to a quiet night in. Nothing about either option changed except which one is now immediate.

The raise already spentillustration

A saver decides that next year's increase goes straight into the pension, and means it. The increase lands and the money goes on a kitchen. The plan made at a distance and the choice made on the day are the same choice, ranked in opposite orders.

Monday morning and Sunday eveningillustration

On Sunday the week ahead is one of early starts and no drinking. By six on Monday the extra hour in bed is worth more than the run, and by Thursday the drink is worth more than the morning after. Each reversal arrives as the nearer option comes into reach.

Cases marked as illustrations describe nobody in particular. They are there to make the pattern recognisable, not to report that a particular thing happened.

The experiments

What experiments is Hyperbolic discounting based on?

The design

The classic demonstration is an implied discount rate that falls as the horizon lengthens. Thaler asked respondents what sum in one month, one year or ten years would leave them indifferent to $15 now. Frederick, Loewenstein and O'Donoghue give the median answers as $20, $50 and $100, implying annual rates of 345%, 120% and 19%. Those are median answers from one sample under one matching procedure, and Thaler's paper could not be obtained during this check, so the figures are attributed to that review rather than read off the original.

The same review reports that fitted hyperbolic functions beat exponential ones on such data, citing Kirby, Kirby and Marakovic, Myerson and Green, and Rachlin, Raineri and Cross. The testable core is the preference reversal, and the review cites four human studies and two pigeon studies for it. A footnote records that all of them are synchronic: the subject holds both preferences at one sitting.

The diachronic version, in which subjects brought back 30 days later have switched, follows from the synchronic one only if discounting is stationary. Laibson used the beta-delta form to explain why consumption tracks income, why consumers have asset-specific marginal propensities to consume, and why financial innovation might lower welfare by supplying too much liquidity.

4 primary research7 replication and reanalysis2 commentary and critique5 background and reference

Origin

Who first described Hyperbolic discounting, and when?

Year1975
WhoGeorge Ainslie
WherePsychological Bulletin

"Specious reward: A behavioral theory of impulsiveness and impulse control" is a review of economic, psychoanalytic and behavioural accounts of impulsiveness. Only excerpts could be obtained, from Ainslie's own site at picoeconomics.org, and they cover his survey of Jevons, Mill, Freud and Strotz rather than his own model. Frederick, Loewenstein and O'Donoghue record the function Ainslie proposed as D(t) = 1/t.

The year before, Ainslie had reported in the Journal of the Experimental Analysis of Behavior on pigeons given a small immediate food reinforcement for pecking a key and a larger delayed one for not pecking it. Most subjects pecked the key on more than 95% of trials. Three of 10 learned to peck a differently coloured key earlier on, which removed the tempting option.

Those pigeon figures are read from the abstract, which is what PubMed Central serves for that article. Laibson's 1997 Quarterly Journal of Economics paper, 'Golden Eggs and Hyperbolic Discounting', carried the idea into economics through a consumer whose only commitment device is an illiquid asset.

The claim, and the replication this entry reads its figures from.

Where it runs out

When does Hyperbolic discounting not apply?

Searches of Crossref, OpenAlex and Semantic Scholar in August 2026 located no incentivised multi-site replication of the beta-delta estimates on the Many Labs or Registered Replication Report pattern. The one multi-site preregistered study found is Ruggeri and colleagues, and its choices were hypothetical. Four lines of work put the falling discount rate down to the elicitation rather than to time preference.

Read, Frederick, Orsel and Rahman found that discount rates drop with horizon when a delay is described as an interval and stay flat when the same delay is given as a calendar date, an interaction of F(3,165) = 9.4, p < .0001. Halevy separated stationarity, time consistency and time invariance, and reports that about half of his time-inconsistent subjects had stationary preferences, which present bias alone cannot produce.

Chakraborty, Calford, Fenig and Halevy re-analysed the Andreoni and Sprenger data and found 43 violations of the weak axiom of revealed preference among the 36 subjects who chose only corners. Thirty-four ran towards less impatience under the convex time budget than under a choice list and 9 the other way. Seventeen of those 36 always chose the later reward and so could not register a violation at all.

Enke, Graeber and Oprea paid subjects to value a sum 'shrunk 12 times' by 4% each time, a task with no delay in it, and the hyperbolic pattern appeared anyway. A beta-delta model fitted to those atemporal choices returns 0.85, standard error 0.01, where the design forces the true value to 1. Behaviour on the two tasks correlates at 0.34.

Their reading is that present bias identified from the shape of the discount function is inflated, while front-end delay designs, which push both payments back by a common lag, are not confounded. Heterogeneity in the Imai meta-analysis is close to total for money: I-squared is 98.08 across all estimates and 98.13 among monetary studies, so those pooled figures describe a spread rather than a value.

Among the 24 effort estimates it is 45.96. Imai and colleagues find funnel-plot asymmetry in the effort studies, p = .0016, running towards over-reporting of present bias, and their corrections move the effort average anywhere from 0.90 to 0.99 depending on the method used. The underlying measurement has never converged. Frederick, Loewenstein and O'Donoghue plotted every study they could locate that reported a discount rate or let one be computed, and give implied annual rates from -6% to infinity.

They add that the range of estimates is not shrinking over time, and that once studies with horizons under a year are dropped the correlation between horizon and discount factor is -0.0026. Cohen, Ericson, Laibson and White argue that money-earlier-or-later choices cannot isolate time preference, because a subject's financial flows on a date are only weakly linked to consumption on that date.

They report that measures of time preference explain less than 5% of the cross-sectional variation in economic behaviours involving intertemporal trade-offs, while noting that such results are often statistically reliable.

Commonly misread as

How is Hyperbolic discounting commonly misread?

Three claims travel under one name and the evidence differs for each. That people discount the future steeply is not in dispute. That the discount function is hyperbolic rather than exponential is a model-fitting claim, and both Read and colleagues and Enke and colleagues produce the hyperbolic pattern with no change in time preference behind it. That preferences reverse as a date approaches is a third claim, and the studies Frederick, Loewenstein and O'Donoghue cite for it are synchronic.

The state above is about the beta parameter, which asks only whether the present carries extra weight. Nearly every estimate above comes from the quasi-hyperbolic beta-delta form, which is not a hyperbola. It marks down the future in one step and then discounts exponentially. The second error is reading a null result over money as a null result for present bias.

Andreoni and Sprenger make the point themselves: the models of temptation and self-control are about consumption, and a subject with access to modest liquidity should not be expected to show present bias over dated cash, which can be arbitraged. Imai and colleagues find that studies run where more people hold credit cards or can draw on liquid savings report less present bias.

The third error is treating a null in one payment protocol as the last word. The Kenyan results turn on whether 'today' means today or within the hour. The fourth is enrolling the Global Preference Survey as a measurement of present bias. Falk and colleagues surveyed 80,000 people in 76 countries and report patience, which is the level of discounting rather than the way it changes with horizon.

A fifth is the calibrated beta of around 0.7 that circulates from macroeconomic work; Andreoni and Sprenger cite Laibson, Repetto and Tobacman for that figure and set their own 1.004 against it.

Sources

Sources for Hyperbolic discounting

  1. 1Ainslie, G. (1975). Specious reward: A behavioral theory of impulsiveness and impulse control. Psychological Bulletin, 82(4), 463-496. Record and pagination verified at Crossref. Only the excerpts posted on Ainslie's own site were obtained; the full text is closed at the American Psychological Association and no copy was found through OpenAlex, Semantic Scholar or a web search.primary
  2. 2Ainslie, G. W. (1974). Impulse control in pigeons. Journal of the Experimental Analysis of Behavior, 21(3), 485-489. The pigeon figures quoted here are read off the abstract; PubMed Central carries the body only as page scans.primary
  3. 3Thaler, R. (1981). Some empirical evidence on dynamic inconsistency. Economics Letters, 8(3), 201-207. Source of the $15 matching question. The paper is closed at Elsevier and no copy was found through OpenAlex, Semantic Scholar, RePEc or a web search, so the figures quoted here come from Frederick, Loewenstein and O'Donoghue (2002). That review's own Table 1 gives the study's horizons as 3 months to 10 years, where its text gives one month, one year and ten years, and the discrepancy could not be resolved without the original.primary
  4. 4Laibson, D. (1997). Golden eggs and hyperbolic discounting. The Quarterly Journal of Economics, 112(2), 443-478. Read in a full copy of the journal typesetting hosted at Universite Paris-Dauphine.primary
  5. 5Frederick, S., Loewenstein, G., & O'Donoghue, T. (2002). Time discounting and time preference: A critical review. Journal of Economic Literature, 40(2), 351-401. Read in the authors' 30 January 2002 manuscript, whose pagination differs from the journal's.review
  6. 6Andreoni, J., & Sprenger, C. (2012). Estimating time preferences from convex budgets. American Economic Review, 102(7), 3333-3356.replication
  7. 7Augenblick, N., Niederle, M., & Sprenger, C. (2015). Working over time: Dynamic inconsistency in real effort tasks. The Quarterly Journal of Economics, 130(3), 1067-1115.replication
  8. 8Augenblick, N., & Rabin, M. (2019). An experiment on time preference and misprediction in unpleasant tasks. The Review of Economic Studies, 86(3), 941-975. Read in the accepted manuscript dated 26 March 2018 on Augenblick's Berkeley page; the journal version is closed.replication
  9. 9Imai, T., Rutter, T. A., & Camerer, C. F. (2021). Meta-analysis of present-bias estimation using convex time budgets. The Economic Journal, 131(636), 1788-1814.replication
  10. 10Cheung, S. L., Tymula, A., & Wang, X. (2026). A meta-analysis of quasi-hyperbolic discounting. Management Science. Only the abstract could be obtained, from Crossref and from the corresponding text on Cheung's own site; the article is closed at INFORMS and the December 2023 working paper on Tymula's site reports a different set of studies and different figures.replication
  11. 11Balakrishnan, U., Haushofer, J., & Jakiela, P. (2020). How soon is now? Evidence of present bias from convex time budget experiments. Experimental Economics, 23(2), 294-321. The published article is closed at Springer and Cambridge; the figures quoted here were read in NBER working paper 23558.replication
  12. 12Cohen, J., Ericson, K. M., Laibson, D., & White, J. M. (2020). Measuring time preferences. Journal of Economic Literature, 58(2), 299-347. Read as the PubMed Central JATS record of the author manuscript.review
  13. 13Falk, A., Becker, A., Dohmen, T., Enke, B., Huffman, D., & Sunde, U. (2018). Global evidence on economic preferences. The Quarterly Journal of Economics, 133(4), 1645-1692. The journal version is closed; the sample figures were read in NBER working paper 23943.context
  14. 14Read, D., Frederick, S., Orsel, B., & Rahman, J. (2005). Four score and seven years from now: The date/delay effect in temporal discounting. Management Science, 51(9), 1326-1335. The publisher's page returns 403; the statistics quoted here were read in the LSE working-paper version, LSEOR 04.65, which lists the authors in a different order.context
  15. 15Halevy, Y. (2015). Time consistency: Stationarity and time invariance. Econometrica, 83(1), 335-352. Full text not obtained: the Wiley page and the author's Toronto copy both sit behind challenge pages, the Econometric Society's own record carries no abstract, and Semantic Scholar reports the abstract as withheld. The figure quoted comes from the abstract of the Vancouver School of Economics working paper of the same title, revised 25 February 2015, at EconPapers. The supplement to the published paper was obtained and confirms an incentivised design with real payments.context
  16. 16Chakraborty, A., Calford, E. M., Fenig, G., & Halevy, Y. (2017). External and internal consistency of choices made in convex time budgets. Experimental Economics, 20(3), 687-706. Read as the Europe PMC full-text XML.context
  17. 17Enke, B., Graeber, T., & Oprea, R. (2025). Complexity and time. Journal of the European Economic Association, 23(5), 1838-1867.context
  18. 18Ruggeri, K., Panin, A., Vdovic, M., Veckalov, B., et al. (2022). The globalizability of temporal discounting. Nature Human Behaviour, 6(10), 1386-1397. Crossref lists 171 authors.replication

Every claim on this page was held against these sources on 2026-08-17. Nothing here is written from memory. Written and checked by Krishna Chagti; how entries are written and corrected. Found a mistake? Report it, with the sentence and a source.

About Hyperbolic discounting

Is Hyperbolic discounting real?

Two meta-analyses disagree about money, giving beta of 0.97 with 1 not rejected and 0.938 with 1 excluded, and both find more present bias for non-monetary rewards. The effect sizes from the original study and from the replication are printed side by side above, so the comparison is visible rather than asserted.