Tide table · when to expect high water
Hyperbolic Discounting
The tendency to discount the value of a future reward by a hyperbolic (rather than constant/exponential) function of delay, so that valuation drops steeply for short delays and shallowly for long ones, producing a disproportionate preference for smaller-sooner over larger-later rewards when the sooner option is near the present.
The lunar pull · the condition that drives the surge
Why the water moves
The core empirical regularity is that the per-period discount rate declines with delay: value falls sharply over the next days/weeks and gently over distant horizons. Ainslie derived this from Herrnstein's matching law (value proportional to amount divided by delay), giving a hyperbola V = A/(1+kD). Because two hyperbolic curves of different size and delay can cross over time, the ordering of options reverses as the present approaches - hence impulsiveness and demand for commitment devices without any change in the rewards themselves. Laibson's quasi-hyperbolic 'beta-delta' form captures the key feature (an extra one-time discount on all non-immediate payoffs) in an analytically tractable way.
Whether the hyperbola is the true psychological function is disputed. Read (2001) showed that apparent hyperbolic 'declining impatience' can instead arise from SUBADDITIVE discounting (discounting more when an interval is split into subintervals); his experiments found strong subadditivity but no independent declining impatience, implying the hyperbola may be partly an artifact of how intervals are presented. Other accounts attribute the pattern to attention/salience, uncertainty about future receipt, or interval/duration perception. The dual-system neural interpretation of McClure et al. (2004) was also challenged (e.g., Kable & Glimcher's work supporting a single valuation system that discounts hyperbolically), so the mechanism behind present bias remains an open debate even though the behavioral curve shape is robust.
Prediction reliability — replication
The behavioral phenomena are well established and stable: hyperbolic functions out-fit exponential ones at the individual level (Kirby & Marakovic 1995; Kirby 1997), discount rates are trait-like with test-retest reliability roughly r=.55-.90 over weeks to a year, and steeper delay discounting correlates robustly but weakly with addictive behaviors (Amlung et al. 2017 meta-analysis, pooled r=0.17 for addiction severity and r=0.11 for quantity-frequency). What is genuinely contested is the deeper structure: Read (2001) argues subadditivity, not a true hyperbola, drives 'declining impatience'; the two-system neural account (McClure et al. 2004) is disputed by single-system hyperbolic-valuation findings; and some intertemporal anomalies (subadditivity, magnitude/sign effects) are not captured by simple hyperbolic models. So the curve shape replicates while the mechanistic claims do not converge.
Recorded tides
Ainslie (1974) - preference reversal in pigeons. Pigeons reversed preference from large-late to small-early as the smaller reward got temporally close, and 3 of 10 birds learned to peck a commitment key that removed the impulsive option - direct behavioral evidence of dynamically inconsistent (hyperbolic-like) discounting.
George Ainslie, 1974 · window: 10 pigeons
Kirby & Marakovic (1995) - hyperbolic vs exponential fit. Both functions fit well, but the hyperbolic function fit better for all delayed rewards and described the data better for the large majority of individual subjects.
Kris N. Kirby & Nino N. Marakovic, 1995 · window: 21 (Exp.1) and 18 (Exp.2) subjects
Kirby (1997) - real rewards & magnitude effect. Hyperbolic function accounted for more variance than exponential, and discounting was steeper for smaller rewards (the magnitude effect), contradicting normative exponential discounting.
Kris N. Kirby, 1997 · window: not reported here (within-subjects design)
McClure, Laibson, Loewenstein & Cohen (2004) - dual neural systems. Limbic/paralimbic regions tied to the midbrain dopamine system were preferentially activated by choices involving immediately available rewards, while lateral prefrontal and posterior parietal cortex were engaged by all intertemporal choices - offered as neural evidence for a beta-delta two-system account of present bias.
Samuel M. McClure, David I. Laibson, George Loewenstein & Jonathan D. Cohen, 2004 · window: not reported here (small fMRI sample, typical N in the tens)
King tides · logged extremes
Paying Not to Go to the Gym · 2006
DellaVigna & Malmendier analyzed 7,752 members of three U.S. health clubs. Members on flat monthly contracts (>$70) averaged about 4.3 visits/month - over $17 per visit - when a 10-visit pass would have cost $10/visit; on average flat-rate members overpaid substantially. The persistent failure to switch contracts or attend as planned is a canonical real-world signature of present-biased (hyperbolic) preferences plus inertia.
Save More Tomorrow (SMarT) 401(k) field deployment · 2004
Thaler & Benartzi designed a commitment program explicitly to counter present bias: employees pre-commit to raise 401(k) contributions at future pay raises. In the first field implementation at a U.S. manufacturing firm, participants' average savings rate rose from about 3.5% to 13.6% over roughly 40 months - a documented case of using a commitment device to offset hyperbolic discounting at scale (later adopted broadly and informing the 2006 Pension Protection Act auto-escalation provisions).
Tide warning · the safe window
Use pre-commitment that binds the impulsive future moment, and force the choice into a 'far-far' frame where both options are delayed (where preferences are consistent). Concretely: auto-escalate savings tied to future events (the SMarT design), set up automatic transfers/illiquid accounts, and decide policies in advance rather than in the moment. Reframing both options as distant, or making the future reward concrete and salient, reduces the steep near-term discount.
Thaler & Benartzi's SMarT commitment program raised participants' average savings rate from ~3.5% to 13.6% over ~40 months by binding future (not present) choices - a documented countermeasure exploiting that delay-vs-delay choices are far less present-biased.
Reading the gauge in the wild
Watch for a flip that depends only on timing, not on the rewards: you firmly plan to start saving / dieting / studying 'next month,' but when next month arrives the immediate option wins again - and the plan re-slips by another month. Other tells: choosing a smaller payout now over a clearly larger one slightly later, demanding commitment devices (locked savings, app blockers) to bind your future self, and the gap between intended and actual behavior shrinking the further away the decision is.
Neighbouring waters
- Present BiasPresent bias is the behavioral consequence most people actually mean - overweighting the immediate moment. Laibson's quasi-hyperbolic (beta-delta) model captures present bias with a single 'beta' jump rather than a full hyperbola; the two terms are often used interchangeably but are not identical.
- Dynamic Inconsistency / Preference ReversalThe crossing of two hyperbolic discount curves over time is the formal source of dynamically inconsistent preferences and the resulting demand for commitment devices.
- Exponential DiscountingThe normative discounted-utility benchmark (Samuelson) with a constant per-period rate and time-consistent preferences; hyperbolic discounting is defined as the empirical deviation from it.
- Subadditive DiscountingRead's (2001) alternative: discounting depends on how an interval is subdivided. It can mimic hyperbolic declining impatience and is the leading challenger to the hyperbola being a genuine function of total delay.
- Delay Discounting (impulsivity measure)The broader experimental paradigm in psychology/psychopharmacology; 'k' from the hyperbolic model is widely used as a trait index of impulsivity linked to addiction, obesity and ADHD.