Hyperbolic Discounting

Hyperbolic discounting is a family of mathematical models for how people discount future rewards, meaning how much less a reward is worth to them the further off it is. It is defined against the classical alternative, exponential discounting, which applies the same discount rate to every period and so keeps preferences consistent over time. A hyperbolic discount curve instead falls steeply over short delays and then flattens out, so waiting from "now" to "soon" feels far costlier than waiting the same length of time from "later" to "even later." That curvature is precisely what allows preferences to reverse. The term names the shape of the discount function; the behavior that shape is built to describe, valuing the immediate too heavily and reversing an earlier plan, is called present bias.

Start with the normative benchmark. Under exponential discounting, formalized in economics decades ago, a reward received t periods from now is worth the present reward multiplied by a constant factor raised to the power t. Because the rate is constant, the model is time-consistent: if you prefer a larger reward later to a smaller reward sooner when both are distant, you will still prefer it when both draw near. Nothing about the passage of time flips the ranking.

Hyperbolic discounting, studied by George Ainslie in the 1970s, replaces the constant rate with a curve whose implied discount rate is high in the near term and lower in the far term. The practical consequence is a preference reversal: a person can genuinely prefer a larger-later reward while both options are far away, then reverse and grab the smaller-sooner reward once the sooner one becomes immediate. A fully hyperbolic function is awkward to work with, so economics mostly uses a tractable approximation, the quasi-hyperbolic or "beta-delta" model, developed by David Laibson in 1997 from earlier work by Phelps and Pollak. It keeps ordinary exponential discounting through the delta term but adds a single extra factor, beta, applied once to every future period. The present carries full weight; everything in the future is knocked down by beta on top of the usual exponential discount. When beta equals one, the model is just exponential and there is no reversal; when beta is less than one, there is an extra one-time penalty on any delay at all, which is the mathematical object that corresponds to present bias.

What the hyperbolic and quasi-hyperbolic forms predict that the exponential form cannot is the reversal itself. That is the whole point of the models: they are built to reproduce the observed fact that plans made in advance get overturned when the moment arrives. It is worth being honest about the empirical state of the field. The evidence that people are, on average, present-biased in this direction is reasonably well supported, but the size of the effect varies a great deal across studies and methods, so the models capture a real qualitative pattern more confidently than any single value of beta. And in ordinary usage the phrases "hyperbolic discounting" and "present bias" are simply swapped for each other. The clean distinction is that hyperbolic discounting is the form of the discount function and present bias is the behavior that form describes; for practical purposes that line rarely changes anything, and the applied lessons about the reversal itself belong with present bias.

In the quasi-hyperbolic model, the value today of a reward is the reward times delta raised to the power of the delay, and then, for anything not received right now, times an additional factor beta. Because beta hits every future period once, it creates a discrete drop between "now" and "any future time," while delta handles the smooth decline across future periods.

A hypothetical example, with invented numbers, showing the reversal a constant rate cannot produce. Let delta be 0.95 per year and beta be 0.7. First, an immediate choice: $100 now versus $110 in one year. The $100 now keeps full weight, $100. The $110 in a year is worth 0.7 times 0.95 times $110, which is $73.15, so this model prefers the $100 now. Next, a distant choice viewed from today: $100 in five years versus $110 in six years. Here both are in the future, so beta applies to each and does not change the comparison; the five-year reward is worth 0.7 times 0.95-to-the-fifth times $100, about $54.16, and the six-year reward is worth 0.7 times 0.95-to-the-sixth times $110, about $56.60, so the model prefers the larger-later $110. The same person prefers the sooner reward when it is immediate and the later reward when both are far off, which is the reversal. Run the identical numbers with beta equal to one (pure exponential) and the later reward wins in both cases, with no reversal at all.

Pros

• The models reproduce a real, repeatedly observed pattern, preference reversal, that the classical exponential model cannot.
• The quasi-hyperbolic "beta-delta" form is simple enough to use in economic analysis while still capturing the near-term overweighting.
• Separating the discount function from the behavior it describes keeps the math and the psychology straight.

Cons

• It is a model, not a measurement: the direction of present bias is well supported, but the estimated size varies widely across studies.
• The distinction between hyperbolic discounting and present bias is precise but, for practical decisions, rarely consequential, so insisting on it can be more pedantic than useful.
• A single parameter cannot capture everything about how a given person values the future, so the model simplifies real behavior.