Long-term unemployment, and the effect of duration itself

Exit rates from unemployment decline with the duration already elapsed. That fact is observed in every country with longitudinal data on trajectories. By international convention, unemployment is described as long-term beyond one year.

Two explanations produce the same curve. Selection: the most employable leave first, and those remaining differ on unobserved characteristics. Duration dependence: duration itself reduces chances, through skill loss, discouragement or the signal sent to employers.

Economists formalised this identification problem as early as the 1980s, notably James Heckman and George Borjas. With observational data alone, the two mechanisms are very hard to separate, which is what led researchers to look for experimental designs.

An experiment published in 2013 by Kory Kroft, Fabian Lange and Matthew Notowidigdo consists in sending fictitious applications identical except for the reported duration of inactivity.

The response rate declines with that duration, which isolates a signalling effect, independent of any real difference between candidates. The measured effect fades beyond a certain duration threshold. The design measures callbacks for interview, not hires, and it says nothing about the other two mechanisms, skill loss and discouragement.

The existence of duration dependence justifies early rather than late intervention, an argument taken up in the design of most European support schemes.

Evaluations of those schemes find clearer effects for early intensive support than for long training started late, a result consistent with the identified mechanism. The argument has a symmetrical limit: intervening early commits resources to people some of whom would have found work without help.

Duration is counted from registration or from the last period of employment depending on the source, and a brief return to work sometimes resets the counter. Administrative sources, based on registration with the public employment service, and surveys based on the International Labour Office definition do not follow the same rules.

Those conventions directly affect the numbers counted as long-term unemployed, and make cross-country comparisons more fragile than overall rates.

The share of long-term unemployed rises mechanically in the years following a recession, even when inflows into unemployment have stopped rising.

That inertia is an arithmetical property of the stock and not a signal of further deterioration, a distinction rarely made in cyclical commentary. It does, however, connect with the notion of hysteresis, proposed by Olivier Blanchard and Lawrence Summers in 1986, according to which prolonged unemployment can durably raise an economy's level of unemployment.