Time Preference

Time Preference, Pure Impatience and Discount Rates for Health

Time preference, in its pure sense, is the part of discounting that rests on timing alone. Someone who would rather have a given health gain this year than the same gain in ten years, with nothing else different, shows positive time preference. It matters because the rates applied to future costs and QALYs are often defended by appeal to it, yet pure impatience supplies only 0.5 of the 3.5 percentage points in the UK Treasury rate. The Treasury derives its 3.5% and 1.5% rates from the Ramsey formula, and the NICE manual (PMG36) sets its own reference-case rate. The present-value formula is on the discount rate page and the case for a lower health rate on the differential discounting page.

Pure time preference and the other reasons for discounting

Frederick, Loewenstein and O'Donoghue separate time discounting, which covers any reason for caring less about a future consequence, including uncertainty or changing tastes, from time preference in the narrower sense of a preference for immediate utility over delayed utility. The TEMPUS report by Cairns and van der Pol gives three reasons individuals are not indifferent to timing, namely that they cannot be sure they will be alive, they expect to be better off and so value further increments less, and what Pigou called a defective telescopic faculty.

Attema, Brouwer and Claxton list the reasons used in economic evaluation as the return on the next best investment, which represents the opportunity cost of a programme; pure time preference, or impatience, which they describe as widely observed; catastrophic risk; and consumption growth, since extra consumption adds less utility to someone who already has more. Only the second is time preference in the strict sense. The others would remain if impatience were set to zero, so a debate about whether society should be impatient about future health does not by itself settle the size of a discount rate.

How the Green Book builds its rates from the Ramsey formula

HM Treasury's supplementary guidance on discounting, published in February 2026, derives the Green Book's social time preference rate (STPR) from two components. Time preference is the rate at which consumption and public spending are discounted assuming no change in per capita consumption, and the wealth effect reflects expected growth in consumption, which lowers the utility of future consumption.

$$STPR = \rho + \mu g, \qquad \rho = \delta + L$$

where $\rho$ is time preference, $\delta$ is pure time preference (society's impatience), $L$ is catastrophic risk, an allowance for unpredictable risks not normally included in appraisal, $\mu$ is the elasticity of marginal utility of consumption and $g$ is the expected annual growth rate of real per capita consumption.

For years 1 to 30 the guidance assumes $\delta$ of 0.5%, $L$ of 1.0%, $\mu$ of 1.0 and $g$ of 2.0%, which gives the following.

$$STPR = (0.5\% + 1.0\%) + 1.0 \times 2.0\% = 1.5\% + 2.0\% = 3.5\%$$

where the bracketed term is time preference and the product is the wealth effect.

Health and life effects are discounted at 1.5% because the wealth effect is excluded, since the welfare or utility from additional years of life does not decline as real incomes rise. The health rate is therefore the Treasury's time-preference term $\rho$ alone, which is broader than pure time preference because it adds catastrophic risk to impatience. The guidance cites Freeman, Groom and Spackman for a plausible range of 0.0% to 1.0% for pure time preference, so 0.5% is a choice within that range.

Beyond 30 years the rates decline because the future values of the parameters are uncertain. The STPR falls to 3.0% for years 31 to 75 and 2.5% for years 76 to 125, and the health rate to 1.286% and 1.071% (guidance Table 3.A). For horizons beyond 50 years the guidance asks for extra sensitivity analysis with pure time preference set to zero, 3.00% for years 1 to 30, citing the Stern Review conclusion that pure social time preference is not ethically defensible where significant and irreversible wealth transfers from the future to the present are involved. The difference between the two results estimates the wealth transfer attributable to pure social time preference.

Individual and social time preference

Individual time preference describes how one person trades off their own outcomes, and the TEMPUS report notes that it influences behaviour such as smoking, exercise and following dietary restrictions. Social time preference describes how a public decision maker should weigh outcomes for different people and generations, which is partly a normative choice. Attema and colleagues describe two routes to a social rate, either estimating the Ramsey components, which assumes all of them, impatience included, are normatively relevant, or measuring individual time preferences, which may be problematic for normative and practical reasons. They record the argument for a zero social rate so as not to discriminate against future generations, and judge that it does not adhere to compelling economic arguments for a positive rate.

One suggested bridge is two-stage discounting, which discounts each person's stream of health at an individual rate and then discounts differences in timing between people at a social rate, which can be lower. Survey evidence on own versus others' health is summarised below.

Measuring time preferences for health

Most evidence comes from stated choices between health outcomes at different dates. The TEMPUS project sent postal questionnaires to a random sample of 5,120 adults in England, Scotland and Wales. In its open-ended method a respondent imagined 20 days of ill health in 2 years' time and stated the maximum days of ill health in a later year at which a one-off minor treatment delaying the illness would still be worthwhile; its second method was a discrete choice experiment.

Median implied rates were 6.1% for own health and 6.2% for others' health with the open-ended method, a difference that was not statistically significant, and 5.0%, 4.6% and 3.8% for own health at delays of 5, 8 and 13 years in the discrete choice experiment, although the authors warn that the sample was unrepresentative. Van der Pol and Cairns found that most respondents had positive rates but a substantial number had negative or zero rates, and TEMPUS links negative time preference to dread. Across studies, Attema and colleagues report average rates for health between 0 and 45%. They also note that time trade-off answers already reflect individual discounting, so discounting QALYs built on them risks double discounting.

Present bias and hyperbolic discounting

The standard discounted utility model assumes a constant rate, and its key axiom is stationarity, under which the preference between two outcomes depends only on the interval between them. Frederick and colleagues call hyperbolic discounting, a rate of time preference that declines with delay, the best documented anomaly. Someone may prefer USD 110 in 31 days to USD 100 in 30 days, yet prefer USD 100 now to USD 110 tomorrow. A simple form that captures much of this is the quasi-hyperbolic discount function, defined as follows.

$$D(k) = 1 \;\text{for}\; k = 0, \qquad D(k) = \beta \lambda^{k} \;\text{for}\; k > 0$$

where $D(k)$ is the weight on an outcome $k$ periods away, $\beta$ is a one-off discount below one on anything not immediate, and $\lambda$ is the per-period discount factor thereafter (written $\delta$ by Frederick and colleagues, a different quantity from the Green Book's $\delta$). The one-off discount makes preferences inconsistent over time, which Hunter and colleagues call present-biasedness, meaning an apparent dynamic inconsistency between the preferences of the short-term and long-term self.

TEMPUS found stationarity violated and hyperbolic models fitting better than the discounted utility model. In an axiomatic test with 203 university students, van der Pol and Cairns found initial and further delays both discounted at a decreasing rate, which suggests the quasi-hyperbolic form may not suit analyses of health behaviour. In a field experiment with real money, Hunter and colleagues found that office workers who were present-biased and had higher discount rates did significantly less physical activity, and note that immediate financial incentives can override present bias for one-off or infrequent behaviour such as vaccination or screening, but may not sustain repetitive behaviour such as physical activity, which links present bias to self-control and intervention design. For economic evaluation, Attema and colleagues note that hyperbolic discounting explains individual behaviour better but leads to inconsistency over time, and that almost all guidelines prescribe constant discounting. Functional forms are on the hyperbolic discounting page.

How time preference connects to the NICE discount rates

The NICE manual (PMG36, last updated 13 October 2026) states in section 4.5.1 that, for the reference case, "costs and health effects should be discounted at the same rate of 3.5% per year". Section 4.5.3 lets the committee consider 1.5% a year for both costs and health effects only when all three of its criteria are met, and section 4.5 states these rates without breaking them into time preference, risk and growth.

Numerically the NICE reference-case rate equals the Green Book STPR for years 1 to 30, which in the Treasury derivation includes a 2.0-point wealth effect, whereas the Green Book discounts health effects at the 1.5% time-preference term only. The 1.5% allowed by PMG36 applies to costs as well, so it is a lower rate for both streams, not the Treasury's health-only rate.

Worked example of implied rates from three stated choices

The figures are illustrative. A respondent answers three TEMPUS-style open-ended questions, each starting from 20 days of ill health in 2 years' time. Indifference between those 20 days and $y$ days in a later year $s$ implies an annual rate given by the TEMPUS report's equation, shown below.

$$r = \left(\frac{y}{20}\right)^{1/(s-2)} - 1$$

where $r$ is the implied annual discount rate, $y$ is the largest number of days of ill health in year $s$ that still makes the delaying treatment worthwhile, and $s - 2$ is the delay in years.

1. Short delay. In year 4 the respondent accepts up to 24 days, so the calculation runs as follows.

$$r = \left(\frac{24}{20}\right)^{1/2} - 1 = 1.0954 - 1 = 0.0954$$

where the result is an implied rate of about 9.5% a year.

2. Longer delays. The same steps give the other two answers.

A respondent with constant time preference would give the same rate at every delay; these answers imply a rate that falls with delay, the pattern TEMPUS reported as a violation of stationarity.

3. What each rate does to a QALY in year 20. The weight is $(1 + r)^{-20}$, and building up the Green Book components shows how little of a 3.5% rate is impatience.

4. Interpretation. Pure impatience removes under a tenth of the weight on a year-20 QALY (1 − 0.905 = 0.095); most of the fall to 0.503 comes from catastrophic risk and consumption growth. An individual rate like this respondent's would leave about a third, so the choice between individual and social time preference changes how prevention with distant gains is valued.

Common misreadings of time preference

Time preference is one input to a discount rate, not the rate itself, and most errors confuse a component or a measured preference with the policy rate used in cost-effectiveness analysis. Four recur.

1. Reading 3.5% as public impatience. In the Treasury derivation only 0.5 points are pure time preference.
2. Treating survey medians as the social rate. Elicited rates vary widely across studies and methods.
3. Reading the declining Green Book schedule as hyperbolic preference. The guidance attributes the decline to uncertainty about future parameter values, not to how individuals choose.
4. Taking present bias as a case for hyperbolic discounting in a reference case. Time-inconsistent weights would let the ranking of two programmes reverse a year later with nothing changed.

Sources

• Attema AE, Brouwer WBF, Claxton K. Discounting in economic evaluations. PharmacoEconomics. 2018;36(7):745-758. doi:10.1007/s40273-018-0672-z
• Cairns J, van der Pol M. Do people value their own future health differently from others' future health? Medical Decision Making. 1999;19(4):466-472. doi:10.1177/0272989X9901900414
• Cairns JA, van der Pol MM. The estimation of marginal time preference in a UK-wide sample (TEMPUS) project. Health Technology Assessment. 2000;4(1):i-iv, 1-83. doi:10.3310/hta4010
• Frederick S, Loewenstein G, O'Donoghue T. Time discounting and time preference: a critical review. Journal of Economic Literature. 2002;40(2):351-401. doi:10.1257/002205102320161311
• HM Treasury. Discounting: Green Book supplementary guidance. London: HM Treasury; February 2026. Chapters 2 and 3.
• Hunter RF, Tang J, Hutchinson G, Chilton S, Holmes D, Kee F. Association between time preference, present-bias and physical activity: implications for designing behavior change interventions. BMC Public Health. 2018;18:1388. doi:10.1186/s12889-018-6305-9
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; published 31 January 2022, last updated 13 October 2026. Sections 4.5.1 to 4.5.3.
• van der Pol MM, Cairns JA. Negative and zero time preference for health. Health Economics. 2000;9(2):171-175. doi:10.1002/(SICI)1099-1050(200003)9:2<171::AID-HEC492>3.0.CO;2-Z
• van der Pol M, Cairns J. Descriptive validity of alternative intertemporal models for health outcomes: an axiomatic test. Health Economics. 2011;20(7):770-782. doi:10.1002/hec.1628