Delayed gratification, honestly bounded

The same choice, moved twelve months out

Delayed gratification is usually described as a character trait. The more useful description is a shape: the weight a reward carries falls fastest over the first stretch of waiting and much more slowly after that. A plan comes apart not because resolve failed but because the ranking changed.

Take the smallest version of it. You are offered ₹10,000 today or ₹10,600 in a month. Now take the identical offer with twelve months added to both dates: ₹10,000 in a year, or ₹10,600 in thirteen months. The increment is ₹600 either way, the wait is one month either way, and the second pair differs from the first only in how far away both of them are. Round figures, chosen to make the arithmetic visible.

The pattern reported in the choice experiments is that these two pairs are not always ranked the same way, and the reversals that turn up run predominantly in one direction — patience in the far pair, impatience in the near one. George Ainslie set out the framework in Specious Reward (Psychological Bulletin, 1975), and Kris Kirby and Richard Herrnstein reported reversals of exactly this two-dates form in Preference Reversals Due to Myopic Discounting of Delayed Reward (Psychological Science, 1995). Richard Thaler had already found, from hypothetical questions put to a small sample, in Some Empirical Evidence on Dynamic Inconsistency (Economics Letters, 1981) that the discount rate implied by people's answers falls as the delay lengthens, which is the same fact seen from the other side.

Sit with what a reversal means, because it is stronger than it looks. Nothing about the two amounts changed between the pairs. Nothing about the gap between them changed. If the ranking flips anyway, then the ranking was never a property of the two options — it was partly a property of where the person doing the ranking happened to be standing in time.

Which is the sentence this whole article is built on, and it is worth stating flatly: a preference between two dated amounts is not stable under moving both dates. Every consequence below follows from that and from nothing else.

Why a constant rate would not produce this

Because a constant rate multiplies both options in the far pair by the same factor as it moves them twelve months out, leaving their ratio untouched — so the ranking cannot change. A constant rate makes reversals mechanically impossible, which is what makes an observed reversal worth something: it separates the part of this that is arithmetic from the part that is measurement.

Paul Samuelson wrote down that standard model in A Note on Measurement of Utility (Review of Economic Studies, 1937): a future amount is discounted at a constant rate for each period of delay. Samuelson himself was careful to say the model was a convenience rather than a description of how people behave, a caution the following decades largely ignored.

So a reversal, if it happens, is evidence about the shape of the curve rather than evidence about the person. It requires the discount to be steeper over the first stretch of delay than over later stretches of the same length — and once it is, moving both options out by a year shrinks the near-term penalty applied to the later one and the ranking can turn over. David Laibson's Golden Eggs and Hyperbolic Discounting (Quarterly Journal of Economics, 1997) formalises this as an extra discount applied once to everything that is not now, with an ordinary constant rate running underneath it. Two components, and it is the first of them that makes a reversal possible at all.

Now the honest bound, and it is the reason this article quotes no discount rate at all. Shane Frederick, George Loewenstein and Ted O'Donoghue reviewed the field in Time Discounting and Time Preference (Journal of Economic Literature, 2002) and found the estimated rates across studies varied so widely that they could not sensibly be treated as measurements of one underlying quantity. The elicitation method, the size of the stake and the framing all moved the answer. Read the direction as well supported and any specific rate as an artefact of how the question was asked.

That gives the three tiers of claim their boundaries. That a steeper near-term segment produces reversals is mechanical, and follows from the arithmetic above whatever the parameters are. That the curve actually has this shape is empirical, supported in direction and unsettled in magnitude. And that anything follows for how a portfolio should be arranged is judgement, treated in the last two sections and labelled as such.

The study everyone quotes, and what later work found

No article on this subject survives without addressing the marshmallow test, and most of them repeat a version of it that the evidence no longer supports.

The procedure is real and the original work is careful. Walter Mischel, Ebbe Ebbesen and Antonette Zeiss described it in Cognitive and Attentional Mechanisms in Delay of Gratification (Journal of Personality and Social Psychology, 1972): a small child, one treat available now, two treats available if they wait. Yuichi Shoda, Walter Mischel and Philip Peake followed a group of those children into adolescence (Developmental Psychology, 1990) and reported associations between how long a child had waited and later measures of competence.

Then came the part that rarely travels with the anecdote. Tyler Watts, Greg Duncan and Haonan Quan revisited it in Revisiting the Marshmallow Test (Psychological Science, 2018) using a larger and considerably more diverse sample. The associations they found were much smaller, and were substantially attenuated once family background and early cognitive ability were controlled for. The finding did not vanish; it stopped carrying the weight the popular retelling puts on it.

Read carefully, that is not a debunking so much as a relocation. If a large part of what looked like a stable trait is accounted for by the circumstances a child grew up in, then waiting time was reporting on those circumstances as much as on the child. Celeste Kidd, Holly Palmeri and Richard Aslin pushed at the same joint from another angle in Rational Snacking (Cognition, 2013), where children's willingness to wait differed according to whether an earlier promise made to them by the experimenter had actually been kept. That was a small sample and one manipulation, so it is worth reading for direction and not for size.

Carrying that into adult finances is a reading rather than a finding — none of the work above tested an adult — but it names the specific mistake this cluster exists to avoid. A person who has watched a plan get overturned by a medical bill, a delayed salary or a family obligation is not failing a character test when they discount a future promise. They are pricing in an environment that has not honoured such promises before, which is a reasonable thing for an estimate to do. The arrangements in the rest of this article work better when the near-term claims on the money are covered first, and that ordering is a consequence of the mechanism rather than an afterthought.

Expecting the reversal is a different problem from having it

There is a second distinction that decides which arrangement is worth anything, and it has nothing to do with how patient anyone is.

Ted O'Donoghue and Matthew Rabin drew it in Doing It Now or Later (American Economic Review, 1999). Two planners can have the identical discount curve and behave completely differently, because one of them expects the reversal and the other does not. The planner who does not expect it writes a plan that assumes the later self will simply carry it out. The planner who does expect it treats the later self as a party to be constrained rather than instructed, and spends effort on the constraint rather than on the resolution.

Richard Thaler and Hersh Shefrin had put the same split more vividly in An Economic Theory of Self-Control (Journal of Political Economy, 1981), modelling a person as a planner who sets the policy and a doer who faces the moment. The planner has the horizon and no power; the doer has the power and a very short horizon. Every arrangement below is the planner buying leverage over a doer it will never meet.

This is where the practical difference sits. “I will invest whatever is left at the end of the month” is a plan written by someone who expects the later self to comply, and it hands the doer both the timing and the amount. A dated instruction that moves the money before it is available to spend hands the doer neither. Both people can want the same outcome equally; only one of them has arranged for the want to be irrelevant. That principle has its own article in paying yourself first, which is this mechanism applied to a salary date.

And it names the failure mode worth remembering. Any plan whose execution requires a fresh decision each month has a renegotiation window each month, and the window opens at precisely the moment when one option is near and the other is distant. The plan is not being broken. It is being re-run, correctly, under the conditions that make the near option win.

Deciding while both options are far away

If the steep segment applies to whatever is happening now, then the only lever with any purchase is when the decision gets made — not how firmly. A decision taken about two distant dates is taken on the flat part of the curve, where the extra discount has nothing to bite on.

The clearest applied demonstration is Richard Thaler and Shlomo Benartzi's Save More Tomorrow (Journal of Political Economy, 2004), in which employees at American firms committed in advance to directing part of future pay rises into retirement saving. Take-up was higher than for an equivalent increase starting immediately. The arithmetic of the two offers is the same; what differs is that the commitment was made about money not yet in anyone's hand, so the near side of the curve had nothing to weigh. The authors credit more than timing alone — inertia holds people in the plan once they are enrolled — so the timing is one limb of that result rather than the whole of it.

Field evidence for restricting your own access points the same way. Nava Ashraf, Dean Karlan and Wesley Yin studied a savings product that blocked withdrawals until a date or amount the saver chose themselves, in Tying Odysseus to the Mast (Quarterly Journal of Economics, 2006), and found balances rose against the comparison group. Note the sample: that study was run in the Philippines, and we are not citing an Indian field experiment, because we do not have one to cite. Laibson's 1997 paper makes the same point theoretically — an illiquid asset is doing a job that a liquid one cannot, and the illiquidity is the feature.

In Indian portfolios the same structure appears in ordinary instruments, and it is worth seeing them as commitment mechanisms rather than as products. An auto-debit instalment dated near the salary credit removes both the timing and the amount from the monthly decision. An equity-linked savings scheme carries a statutory lock-in of three years. Provident fund contributions are deducted before the salary arrives and restrict access until defined events, as does the National Pension System. None of these makes anyone patient. Each one moves a decision to a date on which nothing was tempting.

The arithmetic reframe worth taking away is about counting decisions rather than counting rupees. A twenty-year monthly commitment is not one hard decision repeated 240 times; it is 240 decisions collapsed into one, taken on a day when every instalment was in the distant future. The reason it holds is not that the person became more disciplined. It is that there is no longer a monthly moment at which the near option is available to win.

Where the window opens, and what a prior decision would have settled

The window is not everywhere. It opens at identifiable moments, all of which share one feature: something is available now and the alternative is a benefit dated later. Naming them is more useful than resolving to be better at them.

Every row has the same architecture and none of them is about weakness. In each one the near option is concrete and available and the far one is an abstraction, and that asymmetry is supplied by the calendar rather than by the person facing it.

Two rows lean on mechanisms owned elsewhere in the library and are not re-argued here. The instalment-during-a-fall row is where this meets loss aversion, since skipping also avoids adding money to something currently showing red — and the mechanical effect of skipping is worked through in SIPs during market crashes. The raise row is where it meets lifestyle inflation: a raise absorbed into spending was allocated by default, and a default is a decision taken by whoever did not make one.

The general form of the fix is the one the pillar sets out at length under behavioural biases — a trigger and an action specified in advance, rather than an intention to do better. What this article adds is which trigger to reach for: the useful ones fire on a date or a threshold that is knowable now, not on how a moment feels when it arrives.

What deferral costs, and when the near claim is the right one

Three costs — access, information and the erosion while you wait — and the near claim is right whenever it is the better option rather than merely the nearer one. Delay is not free and it is not a virtue.

Start with the plainest cost. Money placed behind a lock-in is unavailable in a genuine emergency, and the mechanism that protects the plan from a festival sale protects it equally from a hospital admission. It cannot tell them apart. That is why an accessible buffer sits ahead of any committed structure in the ordering — the emergency fund is what stops a real shock from breaking the commitment. A structure broken once is a weaker constraint the next time it is set, which is a reasonable reading rather than a measured effect.

The second cost is that a commitment made early is made with less information than later. Income changes, families change, obligations appear. A structure that was correct when it was set can become wrong, and the whole point of it is that it resists being changed — which means it resists being corrected too. There is no version of this that has the benefit without the cost; the two are the same property.

Third, deferral has an arithmetic price of its own. A rupee held back is a rupee whose purchasing power is being eroded while it waits, which is the subject of inflation, and the mirror of that is the compounding treated in the time value of money. Neither says more waiting is better without limit. They say the comparison is between two dated amounts in real terms, which is exactly the comparison the steep near-term segment distorts.

And the near claim is sometimes simply correct. Clearing a borrowing is arithmetic rather than impatience when its interest rate is contractual and known while the return it would be traded against is neither. So is paying for cover against a risk that is already live, since the risk does not wait for the corpus, and so is spending on something that has a deadline the money does not. Not every present claim is the doer winning; the distinction is whether the near option was chosen because it was better or because it was near, and the two feel identical from the inside.

Which is the last honest bound. Nothing here forecasts what anyone will do, and none of the evidence above licenses a claim about how patient any individual is. The defensible statement is narrower and more useful: a ranking that reverses when both dates move is telling you about the calendar, not about the assets — and a decision taken while both options are far away is taken on the part of the curve where that distortion is smallest.

Putting a number on what the deferral bought

Everything above is an argument about when a decision is taken. The natural follow-up is to look at what a dated, uninterrupted commitment actually produced over a history nobody was choosing at the time.

FNOTrader's Mutual Funds app runs instalment and lumpsum schedules against the full AMFI NAV history — around 34 million NAV rows — and reports invested amount against value, the internal rate of return for cashflows landing on irregular dates — XIRR — the worst peak-to-trough fall along the path, and rolling returns across every available start date.

The last of those is the one that speaks to this article. A distribution across start dates separates what the commitment contributed from what the starting month contributed, which is the confound that makes any single experience of waiting a poor teacher. The maximum drawdown column is the companion figure: it describes the worst stretch a holder had to sit through, which is the thing an arrangement made in advance is actually being asked to survive.

Past performance is a record of what happened, not an indication of what will. FNOTrader is not a SEBI-registered investment adviser and does not give investment advice.

Common questions

Continue reading

More in Behavioural Finance · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article