The European Unemployment Dilemma

The paper’s starting observation is that European OECD unemployment, roughly at parity with the rest of the OECD through the 1970s, has “persistently exceeded the OECD average by around two percentage points” since 1983, accompanied by a sharp rise in long-term unemployment – more than half of European unemployment was long-term (one year or more) by 1989, up from under a third in 1979 (Section 1; Figure 1; Table 1). The authors note existing demand-side explanations (insider-outsider conflicts per Blanchard and Summers 1986 and Lindbeck and Snower 1988; hiring/firing costs per Bentolila and Bertola 1990; capital shortage per Malinvaud 1994) and contrast their own labor-supply-side focus: “we focus on the effects of the welfare state on the supply of labor,” motivated by the observation that high taxation and generous benefits distort labor-supply decisions.

The authors note that most earlier cross-country studies “failed to find any cross-country correlation between unemployment benefits and aggregate unemployment,” and that benefit levels were even negatively correlated with unemployment in the 1960s and early 1970s – yet one OECD study found unemployment responds to benefit entitlements with “a considerable lag of 5 to 10 years, and in some cases 10 to 20 years” (Section 1). Their thesis directly explains this puzzle: they argue the lags “are purely coincidental,” since in their model it is not benefit levels themselves but a changed, more turbulent economic environment interacting with pre-existing generous benefit structures that produces high unemployment, so “the generosity of welfare programs has been increasing steadily… without any discrete jump at the time when the unemployment rate rose” (Section 8).

Turbulence is parameterized as the variance of the distribution of skill losses a worker suffers at layoff; a worker’s human capital is modeled as residing “partly in the worker, partly in the job,” so that losing a job can mean losing a job-specific or industry-specific component of skill (Sections 1-2), a modeling choice the authors connect to Neal’s (1995) finding that displaced workers’ subsequent earnings depend on how closely their new industry matches their old one. They motivate the turbulence story with Gottschalk and Moffitt’s (1994) finding of increased earnings-change dispersion between the 1970s and 1980s, and with Jacobson, Lalonde, and Sullivan’s (1993) finding that long-tenured displaced Pennsylvania workers in the 1980s suffered an eventual earnings loss of “about 25% less than earned at the pre-displacement job” even after reemployment – citing oil shocks, the shift from manufacturing to services, new information technology, deregulation, and globalization as real-world sources of such turbulence.

Under the baseline (no added turbulence) calibration, the welfare-state (WS) and laissez-faire (LF) economies have nearly identical GNP per capita (1.542 vs. 1.555), average productivity (1.657 vs. 1.659), and average skill levels (1.876 vs. 1.880); the WS unemployment rate, at 6.95%, is “less than seven tenths of a percentage point higher” than the LF rate of 6.28%, and the discounted welfare cost of the benefit system is only about “four weeks of per capita GNP” (Section 4, Table 4). The authors conclude that “the efficiency costs associated with the welfare system are relatively small” in tranquil times, even though average unemployment duration is somewhat longer in the WS economy (13.3 vs. 11.8 weeks) and long-term unemployment is modestly more prevalent (12.6% vs. 9.8% of the unemployed with spells of six months or more).

A WS worker’s reservation wage rises with past earnings (since benefits are indexed to them), so a worker who once held a high-skill, high-wage job but has since lost most of his skill through unemployment becomes “extremely picky”: the paper’s example gives such a worker a reservation wage of 0.93 (near the top of the wage distribution) despite his current skill being at the lowest level, and an optimal search intensity of “a mere 0.06” (Section 4, Figures 4-5). Tracking a cohort of high-skill workers laid off with mid-range past earnings, the model’s hazard rate of finding a job starts similar in the WS and LF economies but “diverges considerably” after two years, eventually settling at 4.5% per two-week period in the WS economy versus 15.8% in the LF economy (and as low as 1.7% for workers with the highest past earnings) – a gap the authors attribute to disillusioned long-term unemployed workers raising reservation wages and cutting search effort as their benefit-protected living standard comes to look better than available job offers.

The authors model an unanticipated, one-period twentyfold spike in the layoff rate (from 0.009 to 0.18) in which newly laid-off workers lose all accumulated skill; both economies’ unemployment rates “jump up initially by roughly 16 percentage points,” but thereafter diverge sharply: the LF economy’s unemployed “bite the bullet” and quickly accept lower-paying jobs, while the WS economy suffers a prolonged episode in which, at its peak, 66.6% of the unemployed have been out of work six months or more (versus 33.3% in the LF economy) and 49.5% have been out a year or more (versus 4.2% in the LF economy) (Section 5, Figures 8-9). GNP falls by about 17% in both economies, but average productivity of the employed falls less in the WS economy (4.7% versus 7.2% in the LF economy at one year) only because, the authors stress, skill-depreciated WS workers are “slowly phased in” rather than quickly reemployed – so the WS economy’s relatively better measured productivity is “concealing the severity of the exogenous shock to workers’ skills,” not evidence of a milder impact. The government budget deficit in the WS economy, financed by lump-sum taxes, peaks at “12.5% as a share of steady state GNP.”

Raising the turbulence parameter across four calibrations, the WS unemployment rate rises from 6.95% to 7.13%, 8.84%, and finally 14.87% at the highest turbulence level, while the LF unemployment rate moves in the opposite direction, from 6.28% down to 5.81%, 5.77%, and 5.73% (Section 6, Table 5) – the paper’s central illustration of why a welfare state that performed well under tranquil postwar conditions can become dysfunctional once the economic environment turns more turbulent, without any change in the benefit rules themselves. Average WS unemployment duration rises from 13.3 to 31.8 weeks across this range, and the share of WS unemployed with spells of six months (one year) or more rises from 12.6% (1.3%) to 63.1% (55.6%), while the corresponding LF figures stay essentially flat. Welfare costs also diverge: the discounted-consumption welfare loss from moving to the highest-turbulence calibration is 8.0% in the LF economy but “close to twice as large,” 15.8%, in the WS economy. The authors further report that attempting to solve for a WS steady state at even higher turbulence produces a computational “vicious circle” of exploding unemployment and unemployment-compensation spending with no feasible balanced-budget steady state – which they read as mirroring “a potential instability of a generous welfare state.”

Simulating artificial earnings panels under “tranquil” and “turbulent” parameterizations (matching the turbulence levels used for the 1970-78 versus 1979-87 subperiods), the authors find that increasing their turbulence parameter “spreads the distributions of both components of the Gottschalk-Moffitt decomposition in the direction observed,” though the magnitude of the simulated increase in earnings variability falls short of the actual U.S. data, and the simulated permanent-earnings distribution is narrower than the data’s, which the authors attribute to their model’s workers being ex ante identical, unlike the heterogeneous population in Gottschalk and Moffitt’s PSID sample (Section 7).

The authors frame a generous welfare state as a kind of “time bomb”: it “can function well” and conceal its incentive costs so long as “good jobs” remain available to re-employ laid-off workers at conditions similar to their old ones, but once a large or sustained shock removes those good job matches, generous benefits “hinder… the process of restructuring the economy” by letting workers with depreciated skills choose to live on benefits rather than accept the retraining and lower initial pay that re-employment would require (Section 8). They note this is consistent with the OECD’s (1992a) observation that “former manufacturing workers tend to be over-represented among the long-term unemployed, reflecting the impact of structural adjustment in industry,” and – while endorsing the broad OECD (1994a,b) conclusion that an inability to adapt to rapid structural change is the principal cause of persistent unemployment – the authors state they “believe that a greater emphasis should have been put on reforming benefit systems” than the OECD’s policy report gave it.