Why a Longer Life Can Leave a Retirement Plan Unchanged
The subject of this series is a randomized information experiment about longevity beliefs and retirement planning among 1,000 working-age adults in Trentino–South Tyrol, in northeastern Italy. It asks a deceptively simple question: if people receive an official estimate of how long people like them tend to live, do they revise their own expectations and their retirement plans? The answer matters because a realistic planning horizon can affect saving, work, pension choices, and the risk of running short of money. It also supports a more precise judgment than “information works” or “people ignore facts”: an accurate statistic may still fail to become part of a person’s model of their own future.
That gap between public information and personal relevance is the central theme. The study examines not only whether respondents know the average longevity of similar people, but also whether that knowledge is connected to what they think will happen to themselves. To understand the result, we need to keep three things distinct from the beginning: a population statistic, a belief about one’s own lifespan, and a plan for financing retirement.
A precise experiment behind a broad question
The evidence comes from an IZA discussion paper, based on a survey conducted in February 2024. Respondents were between 25 and 60 years old and lived in one of the Trentino–South Tyrol counties. Interviews were conducted by computer-assisted telephone interviewing. The sample was designed around characteristics such as age, gender, county, employment status, and primary language, so the researchers could compare people who received different information while holding the basic structure of the survey constant.
The intervention was deliberately modest but carefully targeted. A randomly selected half of respondents were told the official actuarial life expectancy for a person with their age, gender, and county of residence. The other half were told that the national statistical office calculates life-expectancy estimates using those characteristics, but they were not given a numerical estimate. This control condition matters: the comparison was not between people who had heard of longevity and people who had not. It was between people who received a specific population figure and people who received only the information that such figures exist.
The researchers then asked about several outcomes. Respondents reported how long they expected to live, when they expected to retire, and how satisfied they were with their retirement planning. The survey also asked about complementary pension participation and related choices, including whether respondents planned to join a scheme or eventually take private pension funds as an annuity. These measures let the researchers examine a chain: information might first change a belief about lifespan, that belief might alter a retirement expectation, and the expectation might influence a broader sense of preparedness or a pension choice.
The study’s headline tension is that the chain did not behave as ordinary intuition suggests. Providing the tailored actuarial figure did not produce a statistically detectable change in respondents’ own longevity expectations or expected retirement age. It also did not change most of the measured complementary-pension outcomes. The result is not evidence that every form of longevity communication is futile. It is evidence that this brief, population-level message did not cross the relevant boundary in this sample and under this design.
That boundary is easy to miss if the subject is described too broadly. This is not simply a survey of whether people are financially literate, nor is it a general verdict on retirement calculators. It is a study of the relationship between beliefs about similar others, beliefs about the self, and choices that depend on a future income horizon. The distinction gives the null result its meaning.
Why an expected lifespan belongs in a retirement plan
Retirement planning is fundamentally a problem of allocating resources across time. While working, a person earns income and may build assets. After retirement, they rely on some combination of public benefits, private pensions, savings, and investment returns. The plan has to last for the years that remain. Expected lifespan therefore does more than add an interesting demographic fact: it helps define the length of the period that must be financed.
The paper’s simple life-cycle framework makes this logic explicit. It treats longevity beliefs as one of at least three knowledge inputs. Longevity literacy concerns the expected age at death. Pension literacy concerns how benefits are calculated, how the public system is financed, and what income may be available after work. Financial literacy concerns concepts such as compound interest, inflation, and diversification. A person can know one of these things while misunderstanding the others. A retirement decision requires them to work together.
The direction of the theoretical pressure is intuitive but not mechanical. If someone expects to live longer, they may need resources for more years. Holding other conditions constant, that can create a reason to save more, consume less before retirement, postpone retirement, or seek a more durable income stream. But the size and even the practical form of the response depend on assets already accumulated, wages, pension benefits, investment returns, health, and the value placed on leisure. Longevity is an input into a decision, not a command that determines it.
The model gives this intuition a measurable scale. Under its assumptions, a one-year increase in expected longevity would raise the optimal retirement age by an average of about 0.43 years, or roughly five months. That is a model-based comparative static: it asks how the optimal choice would respond if the belief changed, conditional on the rest of the model. It is not the observed effect of the information treatment. In the experiment, the crucial upstream belief did not move, so the downstream theoretical response had no reason to appear.
Errors in the planning horizon can matter in both directions. Someone who expects to die earlier than they actually do may exhaust assets sooner and face a sharp fall in consumption. Someone who expects to live longer than they actually do may save more than necessary and give up consumption that could have improved their life while they were alive. The paper’s illustrative calculations suggest that these costs can be asymmetric when a state pension provides a floor after private assets run out. The figures are simulations under chosen parameters, not measurements of respondents’ welfare, but they clarify why an apparently abstract belief can have material consequences.
That is why the information puzzle is important. The value of a more realistic longevity belief does not depend on believing that a single number can solve retirement planning. It depends on recognizing that the number has to enter a larger system of expectations. If it stays outside that system, its theoretical importance will not automatically translate into a changed retirement age or a better-funded future.
The crucial distinction: similar others are not the self
The survey was designed to observe this distinction directly. Respondents first estimated the average age at death for people with the same relevant demographic and geographic characteristics. They later estimated their own expected age at death. The first answer measured a form of population longevity knowledge. The second captured a personal expectation that could incorporate private information, including how respondents viewed their own health or circumstances. The two questions concerned related populations, but they were not asking the same thing.
That difference changes how an official statistic can be interpreted. An actuarial estimate describes an average for a reference group. It does not say where a particular individual sits within that distribution. A respondent may therefore treat the estimate as useful for describing other people while believing that their own health or other private circumstances place them somewhere else. The personal belief can remain optimistic even when the population estimate is revised upward.
The paper’s control-group results show how large that separation can be. Respondents expected to live to about 85 years and four months themselves, around three years longer than the lifespan they assigned to similar people. Twenty-one percent believed that similar others would die before age 80, but only 1% expected to die themselves before reaching that age. This is not merely a small rounding difference. It is a systematic optimism gap in how mortality is assigned to the self and to comparable others.
The relationship between the two beliefs was correspondingly weak. Their overall correlation was 0.09. Among respondents who put similar others below age 80, it was slightly negative, at −0.03; among those who put similar others above 80, it was positive but still modest, at 0.22. A statistic can be related to a person’s self-belief in principle, yet the observed relationship can be too weak for a population update to travel far.
The researchers discuss several possible interpretations, including a tendency to form beliefs about oneself through introspection and beliefs about others through observable characteristics. They also consider self-image protection and information avoidance. These are plausible mechanisms, not diagnoses established by the experiment. The evidence does not tell us that every respondent consciously rejected the statistic, or that distrust was the decisive explanation. It does show that the self–other link needed for automatic updating was largely absent.
The paper’s conceptual model says that information must be new and credible, but also relevant to the inference a person is making. The actuarial figure was official and tailored by age, gender, and county. Yet tailoring a population estimate is not the same as providing an individualized medical forecast, and credibility is not the same as personal relevance. A respondent can accept that the figure is correct for similar people without using it to revise the expected age at their own death.
This is the question that should guide the rest of the analysis: not simply whether people saw a number, but whether the number entered the belief that governs a decision. The IZA experiment makes that question unusually clear. It begins with a statistic about a group, follows the possibility that the statistic will become a belief about the self, and then asks whether the self-belief reaches retirement planning. The first step can be accurate while the later steps remain disconnected.
The next task is therefore to examine the retirement equation itself: what must be combined with a longevity belief, and why would a change in that belief matter for work, saving, and consumption in the first place? Only then can the experiment’s unchanged plans be read as a result about a broken link rather than as a paradox about information in general.