How do search frictions, firm hiring rules, and heterogeneous worker skills generate unemployment duration distributions, wage dispersion, and Beveridge curve dynamics that aggregate models compress away?
Background
The Diamond-Mortensen-Pissarides (DMP) search-and-matching framework treats the labor market as an encounter between a representative vacancy and a representative unemployed worker, with matching governed by an aggregate matching function. This compression is powerful for understanding average unemployment and vacancy dynamics, but it cannot speak to the distribution of unemployment durations, the heterogeneity of wage offers across otherwise similar workers, or the network effects through which a plant closure propagates through local labor markets. Labor market ABMs exist to put these distributional and network dynamics back into the analysis. Neugart (2008) built one of the first policy-oriented labor ABMs, studying the effects of active labor market programs on matching efficiency. Fagiolo, Dosi, and Rovelli (2010) embedded a labor market with heterogeneous firms and workers into a macro ABM, showing that aggregate matching functions can emerge from individual search behavior but that the emergent function is not stable across policy regimes.
The core mechanism is decentralized matching with heterogeneous agents. Workers differ in skills, reservation wages, search intensity, and location. Firms differ in productivity, vacancy posting behavior, and wage-setting rules. A worker does not encounter 'the labor market'; they encounter specific vacancies with specific firms, and the match quality depends on the bilateral skill-productivity fit. This means that aggregate unemployment can coexist with unfilled vacancies not because of a calibrated matching-function elasticity but because the skill distribution and vacancy distribution are misaligned. The Beveridge curve becomes an emergent object, not an imposed relationship.
Policy institutions use labor ABMs for questions where heterogeneity is the mechanism. The European Commission's DG Employment has funded ABM research on active labor market policies (training, job subsidies) to evaluate their impact on different worker cohorts. The ILO has explored ABM approaches to study informal employment dynamics in developing economies. The Bank of England's Dawid et al. (2014) Eurace framework includes a labor module used to study the interaction between labor market flexibility and financial stability. Academic work spans matching efficiency under directed vs. undirected search (Hamill-Gilbert 2016), skill mismatch and hysteresis (Seppecher-Salle 2015), and the interaction between labor market institutions and macroeconomic volatility (Dosi et al. 2017, 2018).
How the Parts Fit Together
The economy is assembled from three agent populations. Workers (N = 1,000 to 20,000) carry state variables: skill type (discrete or continuous), current wage (if employed), reservation wage, unemployment duration (if unemployed), search intensity, and a saving/consumption rule that links labor income to aggregate demand. Firms (F = 100 to 2,000) carry: productivity level, current workforce, vacancy count, a hiring rule (skill requirement, offered wage), and a firing/layoff rule triggered by demand shortfalls. The third agent is a policy block: unemployment insurance parameters, minimum wage, hiring subsidies, and training programs.
Matching happens through a decentralized protocol. Each period, unemployed workers sample a subset of posted vacancies (the number sampled depends on search intensity). For each sampled vacancy, the worker checks whether the skill requirement is met and whether the offered wage exceeds the reservation wage. If both conditions hold, the worker applies. Firms collect applications, rank them by skill fit (and possibly wage cost), and hire the best match up to their vacancy count. Unmatched vacancies and unmatched workers carry over to the next period. This bilateral protocol is what generates the emergent Beveridge curve: the aggregate vacancy-unemployment relationship is a consequence of the micro matching, not an input.
The demand side closes the loop. Employed workers earn wages and consume; consumption drives firm revenue; firm revenue drives labor demand (vacancy posting and retention decisions). A negative demand shock reduces firm revenue, triggers layoffs, increases unemployment, reduces consumption further, and feeds back. This multiplier mechanism is present in Keynesian macro models, but the ABM adds the distributional dimension: the workers who lose their jobs are not random draws from the population. They are concentrated in specific firms, skill brackets, and locations, and the pattern of who loses and who keeps a job matters for the recovery dynamics.
Applications
The European Commission's DG Employment has used labor ABM frameworks to evaluate active labor market policies (ALMPs) across heterogeneous worker populations. A training subsidy that is effective for medium-skilled workers may be wasteful for high-skilled workers who would find jobs anyway and ineffective for low-skilled workers who face structural barriers beyond training. The ABM's ability to decompose aggregate employment effects by worker type makes it useful for policy targeting. Neugart (2008) showed that the aggregate matching function shifts endogenously when ALMPs change the composition of the unemployed pool, a result that cannot be obtained from a model with a fixed matching function.
The Eurace framework (Dawid et al. 2014) uses its labor module to study the interaction between labor market flexibility and financial stability. In a flexible labor market (low firing costs, short UI duration), firms adjust employment quickly to demand shocks, but the resulting income volatility amplifies consumption cycles and feeds back into financial stress through household balance sheets. In a rigid labor market (high firing costs, long UI duration), employment adjusts slowly, but the fiscal cost of UI and the hysteresis from long unemployment spells create a different kind of fragility. The ABM captures this tradeoff because it tracks individual worker histories, not just the aggregate unemployment rate.
The model breaks down when the labor market question is primarily about wages in the upper tail of the distribution (executive compensation, superstar effects) or about long-run human capital accumulation over decades. These dynamics require richer models of skill investment, career progression, and firm-worker co-investment that go beyond the matching framework. For pure wage inequality analysis at the top of the distribution, a competitive assignment model (Gabaix-Landier 2008) or a tournament model is more appropriate.
Literature and Extensions
Key Papers
- Neugart (2008) -- labor market policy ABM with active labor market programs and endogenous matching efficiency
- Fagiolo, Dosi, and Rovelli (2010) -- emergent matching functions from heterogeneous agent search
- Dawid et al. (2014) -- Eurace labor module with skills, search, and macro feedback
- Dosi et al. (2017) -- K+S labor extension with endogenous skills and hysteresis
- Seppecher and Salle (2015) -- skill mismatch, wage rigidity, and hysteresis in ABM labor markets
- Hamill and Gilbert (2016) -- directed vs. undirected search in ABM matching
Named Variants
- Multi-sector labor ABM (sector-specific skills and inter-sector mobility)
- Spatial labor ABM (geographic matching with commuting costs)
- On-the-job search extension (job ladder and poaching)
- Endogenous skill accumulation (learning by doing, training investment)
- Gig economy extension (flexible hours, platform matching)
Open Questions
- Does the emergent matching function remain stable across policy regimes, or does Lucas critique apply to ABM labor markets?
- What is the minimal heterogeneity needed to reproduce the stylized facts of unemployment duration distributions?
- Can ABM-based active labor market policy evaluations outperform randomized controlled trials for cost-effectiveness ranking?
Components
An individual worker with state vector (skill, wage/reservation wage, employment status, unemployment duration, search intensity). Decisions: search, accept/reject offers, set reservation wage.
A producing firm with state vector (productivity, workforce, vacancies, revenue, hiring rule). Decisions: post vacancies, screen applicants, hire, fire/lay off.
The aggregate relationship between unemployment U and vacancies V that arises from bilateral matching. Not imposed but measured from simulation output. May shift endogenously with skill mismatch.
Periods since worker i's last employment spell ended. The duration distribution is a key emergent output: right-skewed with a long tail that lengthens during recessions.
Minimum wage worker i will accept. Typically declines with unemployment duration (adaptive) or is a fixed fraction of the last earned wage. The reservation wage rule drives the acceptance/rejection decision.
Policy instrument: replacement rate and maximum duration. Higher replacement raises reservation wages and search selectivity; maximum duration creates a cliff in search behavior.
Assumptions
Workers sample a finite number of vacancies per period (typically 2 to 10) rather than observing the entire vacancy distribution.
If violated: If workers observe all vacancies, matching frictions vanish and the model collapses to a frictionless competitive labor market.
Both workers and firms are heterogeneous in characteristics that affect match quality and surplus.
If violated: Homogeneous agents on either side eliminate wage dispersion among identical workers and remove the skill-mismatch channel.
Workers adjust reservation wages downward as unemployment duration increases, following a declining schedule.
If violated: Fixed reservation wages produce unrealistically long unemployment durations for some workers and eliminate the duration-dependence pattern observed in data.
Firms post vacancies as a function of expected demand (revenue or orders), not as a search for optimal labor input in a production function.
If violated: Optimization-based hiring reintroduces the representative-firm logic that the ABM is designed to move beyond.
In the baseline specification, employed workers do not search for better jobs. Job-to-job transitions are excluded.
If violated: Adding on-the-job search enriches wage dynamics (workers climb the job ladder) but roughly doubles computational cost and requires additional calibration targets.
The baseline model has no trade, no migration, and a single production sector.
If violated: Sectoral and spatial extensions exist but are not part of the baseline specification.
Unlock Full Macro Model Library with Starter.
This feature is exclusively available to Starter, Research, and Pro. Upgrade when you need this workflow, review pricing, or send a question before changing plans.
What you keep on Free
- Create and edit one custom board
- Use up to 3 widgets on each Free board
- Browse indicators and calendar