Optimal timing of unemployment benefits: evidence from Sweden

A public program of unemployment benefits aims to protect people against job loss, but it should ideally be designed so that it doesn’t encourage them to stay out of work too much longer than they otherwise would. This research explores how policy can achieve the ideal balance between maximizing the insurance value of benefits while minimizing the incentive cost. Analyzing data from Sweden on unemployment, consumption, income, and wealth, the findings indicate that, contrary to recent reforms that push towards making the generosity of benefits decline over the unemployment spell, it is more socially desirable to reduce benefits for the short-term unemployed in order to raise them for the long-term unemployed.

The key objective of social insurance programs is to provide insurance against adverse events that affect people’s earnings potential, such as job loss, or their spending, such as poor health. The value of unemployment insurance, for example, lies in the fact that households would otherwise be unable to weather the shock of losing their job: their resources and consumption would go down. Unemployment benefits provide extra resources at a time when households value them the most.

But while social insurance programs provide protection against adverse events, they must also ensure that incentives to avoid or reduce the likelihood of these events remain in place. For example, providing higher benefits when workers are unemployed distorts their incentives to reduce the time they spend in unemployment. This makes the cost of providing insurance higher than it otherwise would be.

The design of a social insurance program is therefore a subtle balancing act between maximizing the insurance value while minimizing the incentive cost.

The design of a social insurance program is a subtle balancing act

Adverse events and the protection provided by social insurance are inherently dynamic. If workers lose their job today, they might end up having a shorter or longer spell of unemployment. An unemployment insurance policy needs to determine how much social protection to provide for workers experiencing shorter or longer spells out of work.

The key question here is how the insurance value and incentive costs depend on the time spent in unemployment. In general, the design of social insurance policies tends to be a dynamic problem, and it should aim to specify a schedule of benefits and taxes that balances incentives and insurance over time.

There are various reasons why the insurance value and incentive costs of benefits may change over the course of a spell of unemployment. The challenge is how to set up the profile.

In practice, there seems to be little consensus. Unemployment insurance policies vary substantially across countries, not just in their overall generosity, but also in how benefits are timed over a spell of unemployment. In the United States, for example, benefits are paid only during the first six months of unemployment. In other countries, such as Belgium and Sweden, the unemployed could in principle receive the same level of benefits forever.

Recent policy reforms have reduced benefits for the long-term unemployed relative to the short-term unemployed. There seems to be an overall push towards making the generosity of benefits decline over the unemployment spell.

Dynamic selection and duration dependence

The main theoretical argument underpinning a steeper decline of insurance profiles over a spell of unemployment is the idea of forward-looking incentives: people who lose their job are expected to exert more effort to leave unemployment early in a spell when they know that their benefits will decrease or expire if they remain unemployed.

But do we know whether this argument is strong enough to make declining benefit profiles socially desirable? Solving the dynamic problem of balancing incentives and insurance can prove daunting, especially when taking account of two important features of unemployment: ‘dynamic selection’ and ‘duration dependence’.

Notes: The figure displays estimates of the elasticity of the remaining duration of unemployment, conditional on remaining unemployed until t, with respect to changes in the benefit level (together with 95% robust confidence intervals). The estimates can be interpreted as percentage increases of the unemployment duration if unemployment benefits go up by 1%. Unemployment duration is defined as the number of weeks between registration at the public employment service (PES) and exiting the PES or finding any employment (part-time or full-time employment, entering a PES program with subsidized work or training, etc.). The sample is restricted to unemployed individuals with no earnings who report being searching for full-time employment.

Dynamic selection means that the pool of unemployed people varies significantly at different unemployment durations. What’s more, there may be critical differences in the characteristics of those individuals that affect their value of receiving benefits or their unemployment response to insurance coverage.

For example, older people are more likely to become long-term unemployed. But they may also have more assets and savings to help them cope with unemployment compared with younger people. At the same time, they might be less attractive to employers, and therefore have less control over their employment prospects.

Duration dependence means that the circumstances for a given individual may change depending on the time spent in unemployment: a worker may start a spell with some liquid resources but will deplete these as the spell continues. Workers’ skills may also depreciate the longer they stay unemployed; or employers may discriminate against those who have been unemployed longer, viewing it as a negative signal (Kroft et al, 2013).

Solving for the optimal profile of unemployment insurance is notoriously challenging in the presence of dynamic selection and duration dependence. The difficulty lies in the fact that it is hard empirically to identify the precise forces and mechanisms at play over a spell of unemployment. But they have drastically different implications for the optimal profile of unemployment benefits.

A new methodology to identify the optimal profile

Our study circumvents the issues encountered in previous work by developing a framework in which to characterize the optimal time profile of unemployment benefits and to evaluate the consequences of changes in the profile of existing unemployment insurance policies. In doing so, we aim to bridge three different research strands:

  • First, there is an influential body of theoretical work on optimal dynamic policies. These are typically derived in stylized models that are often difficult to connect with the data (for example, Shavell and Weiss, 1979; Hopenhayn and Nicolini, 1997; Shimer and Werning, 2008).
  • A second empirical body of work analyzes the structural dynamics of unemployment, but without drawing the consequences for dynamic policies (for example, Van den Berg, 1990; Eckstein and Van den Berg, 2007).
  • Finally, a recent but growing body of research has started evaluating social insurance design using the so-called sufficient statistics approach. This work has been mostly silent about the dynamic features of social insurance programs (for example, Chetty, 2008).

The key insight of our approach is that the insurance value and incentive cost of a dynamic policy can be expressed as a function of a limited set of identifiable and estimable statistics, without having to retrieve the precise mechanisms that drive their evolution over a spell of unemployment.

In particular, the incentive cost of benefits paid in week t of the spell is captured by the change in total unemployment duration in response to a change in benefits in week t. This is what economists call the ‘behavioral revenue effect’.

But what exactly is the behavioral revenue effect and where does it come from? When changing benefits paid out in week t, this has a direct mechanical effect on government revenues, as different benefits are paid to all individuals unemployed in week t of a spell.

But the change in benefits will also affect the behavior of the unemployed. Indeed, it may affect the behavior of the unemployed not only in week t, but also in any other week: some might change their behavior before week t in anticipation; and some after week t.

All these behavioral responses will affect the probability of staying unemployed at any point during the spell and therefore the expected benefits paid: this is the behavioral revenue effect. When this effect is large, there is a large cost to the government of providing benefits at this duration.

For the insurance value, the marginal value of giving out benefits in week t of the spell depends only on the value of an additional euro to people unemployed in week t, compared to the value of a euro to everyone else. Furthermore, the value of giving a euro to households can be approximated using their current level of consumption: everything else equal, a poor individual condemned to low consumption levels will value an additional euro more than a wealthy individual with high levels of consumption.

Evolution of incentive costs and insurance value over an unemployment spell

Equipped with this general framework, we take it to the data. We first estimate how the behavioral revenue effect of changing benefits evolves over a spell of unemployment, and then how consumption evolves over the spell. We analyze rich administrative data from Sweden, focusing on the incentive costs and insurance value of benefits provided early in a spell (before 20 weeks of unemployment) versus later in a spell (after 20 weeks).

Our results show that unemployment durations respond significantly to changes in benefit levels, whether these benefits are paid early or later in a spell. Importantly, we find that the response to changes in benefits paid earlier in a spell is 25% larger than the response to benefits paid later in the spell. In other words, the behavioral cost of giving benefits later in a spell is lower than the cost of giving benefits early in the spell.

This result may seem surprising. All else equal, the incentive cost from increasing benefits for the long-term unemployed is expected to be larger as it also discourages the short-term unemployed from leaving unemployment when they are forward-looking. This is the main argument for declining profiles of unemployment benefits. But this neglects the powerful roles of dynamic selection and duration dependence, which are large enough to offset the significant effect of forward-looking incentives.

Figure 1 illustrates the declining responsiveness to changes in unemployment benefits over a spell. For workers at the onset of a spell of unemployment, the average time spent unemployed increases by more than 1.5% when increasing benefits by 1%. For workers who have been unemployed for five months, this increase is only 0.5%.

Using a unique measure of annual consumption obtained from administrative data on the universe of unemployed workers, our results show that the insurance value of benefits increases significantly over a spell of unemployment. We show that consumption drops substantially and early in the spell: on average by 4.4% in the first 20 weeks of unemployment compared with its pre-unemployment level (see Figure 2).

Notes: This figure shows average annual consumption drops relative to average consumption in the last year prior to unemployment as a function of time t spent unemployed as of December (when annual consumption is observed in the registry data) in the year of being laid off. We can use variation in the time individuals have spent unemployed in December of their year of layoff to estimate the evolution of household consumption in 10-week intervals throughout the unemployment spell.

This drop doubles to 9.1% on average for those who are unemployed for longer than 20 weeks. This is estimated based on how annual consumption by unemployed workers depends on the number of months they have been unemployed.

We also leverage the richness of the data to document the mechanisms underlying the observed patterns of consumption, and how they affect the value of receiving benefits over a spell of unemployment. We document that most unemployed people have few liquid assets at the start of a spell, but those who do have assets use them to smooth their consumption early on.

As these assets get depleted, their consumption drops over the spell. We also document the presence of significant liquidity constraints that prevent unemployed people from using debt to smooth their consumption, especially among the long-term unemployed.

Finally, we show the limited role of adjustments in the labor supply of other members of the family to provide additional means of consumption, even for the long-term unemployed. Taken together, our evidence consistently indicates that the consumption-smoothing value of unemployment insurance is higher for the long-term unemployed.

What have we learned?

Decreasing benefits early on in a spell of unemployment in order to increase benefits later on would be socially desirable

Putting this evidence together, our results indicate that declining benefit profiles may not be optimal. In the context of Sweden, we see that benefits given early in an unemployment spell have a larger incentive cost and a much lower value to the unemployed than benefits received later in the spell. This means that decreasing benefits early on to increase benefits later on would be socially desirable.

These results are of course specific to Sweden’s current policy and institutional environment. But some of the dynamic forces that we identify are likely to operate in similar ways in other countries, suggesting that our conclusions may be applicable beyond the Swedish context.

This article summarizes ‘The Optimal Timing of Unemployment Benefits: Theory and Evidence from Sweden’ by Jonas Kolsrud, Camille Landais, Peter Nilsson and Johannes Spinnewijn, published in the American Economic Review in 2018.

Jonas Kolsrud is at the National Institute of Economic Research, Stockholm. Camille Landais and Johannes Spinnewijn are at the London School of Economics. Peter Nilsson is at the Institute for International Economic Studies, Stockholm.