Zero-hours Contracts in a Frictional Labour Market

Layer 1: Overview

Dolado, Lalé, and Turon build a structural equilibrium model of the U.K. low-wage labour market to evaluate zero-hours contracts (ZHCs), employment agreements under which firms are not required to guarantee any minimum working hours and workers may decline any hours offered. The paper’s central question is whether ZHCs raise or lower welfare in general equilibrium, and through which channels. The model features two-sided heterogeneity in a random-search-and-matching environment: firms differ in the volatility of their labour demand, workers differ in their relative preferences for flexible versus regular employment, and wages are fixed at or near the statutory minimum wage. Three mechanisms operate simultaneously. First, a job-creation effect: firms facing highly volatile demand that cannot profitably hire under regular terms enter the market only because ZHCs exist. Second, a substitution effect: some firms that could hire under regular contracts instead post ZHC vacancies, crowding out regular employment. Third, a labour-force-participation effect: workers with a strong preference for flexible schedules join the labour force specifically because ZHCs exist and would withdraw if ZHCs were banned.

The model is calibrated to U.K. Labour Force Survey data for the low-pay segment (roughly 16 percent of total employment), covering September 2018 through March 2020, with a sample of 9,342 individuals aged 16 to 69. A mixture-of-exponentials approach due to Karlis and Xekalaki (1999) applied to job-tenure and unemployment-duration distributions reveals statistically exactly two worker types in both ZHC employment and unemployment, and only one in regular employment, consistent with the presence of R-best workers (who prefer regular employment but accept ZHCs as a stepping stone) and Z-only workers (who would exit the labour force without ZHCs) but not R-only or Z-best workers. Calibrated parameters include a biweekly job-finding rate of λ(θ) = 0.051, a job-destruction probability of δ = 0.005, an on-the-job search efficiency of x = 0.352, and a share of R-best workers of ζ_{R-best} = 0.969. The matching function elasticity ψ is estimated to be 0.65 from U.K. occupation-level hiring and vacancy data (range 0.60–0.70 across specifications). ZHC employment accounts for 6.5 percent of the low-wage employment stock but 19.4 percent of vacancies, because higher turnover in ZHC jobs causes them to be re-advertised more frequently.

A ban on ZHCs — simulated as an extreme tightening of flexible-work regulation — raises the unemployment rate by 2.0 to 2.7 percentage points depending on the assumed volatility of ZHC firms’ demand. When ZHC workers have a low enough disutility of labour that they remain in the workforce after a ban (accepting regular jobs instead), the employment rate falls by the same 2.0 to 2.7 p.p., and sectoral GDP falls by only 0.02 to 0.14 percent, because higher average hours per employed worker partially offset the employment decline. When ZHC workers’ disutility is high enough that they withdraw from the labour force, the employment-rate fall is larger — 4.8 to 5.4 p.p. — and sectoral GDP falls by 2.9 to 3.2 percent. Decomposing via the model’s analytical formula (Proposition 4a), lower job creation alone would reduce regular employment by almost 30 percent in isolation (λ(tilde-θ)/λ(θ) = 0.71), but this is partially offset by reduced vacancy competition (+24 percent, ceteris paribus) and improved search efficiency for regular jobs (+15 percent, ceteris paribus) after the ban.

Welfare effects are measured in consumption-equivalent variation units. In general equilibrium, R-best workers (those who prefer regular jobs but sometimes hold ZHCs as a stepping stone) suffer welfare losses of −0.5 to −0.6 percent of consumption from a ZHC ban, driven primarily by longer expected unemployment spells. Yet in a partial equilibrium experiment that converts their ZHC jobs to regular jobs while holding all other equilibrium objects fixed, these same workers gain approximately +0.2 percent: the substitution effect is genuinely welfare-improving for them in isolation, but the job-creation channel dominates in general equilibrium and more than reverses that gain. Z-only workers — those who would exit the labour force if ZHCs were banned — suffer general-equilibrium welfare losses of −1.7 to −2.0 percent (low-disutility scenario) or approximately −1.8 to −2.1 percent (high-disutility scenario). These losses exceed the losses to R-best workers because Z-only workers are also forced into a type of employment they strictly prefer to avoid. The paper concludes that a ZHC ban is welfare-reducing for all workers in general equilibrium, and proposes that policy instead target ZHC use toward matches where workers voluntarily choose flexibility (Recommendation P1) and toward small firms that cannot diversify demand volatility across many positions (Recommendation P2).

Layer 2: Deep Dive

What is the model’s core structure and what frictions drive the results?

The model is a discrete-time steady-state random-search-and-matching model with two-sided heterogeneity. Workers are heterogeneous in their flow payoffs from regular employment (ω^i_R), flexible ZHC employment (ω^i_Z), and non-employment (ω^i_N), with these payoffs shaped by CRRA utility over consumption and a type-specific disutility of hours worked (α^i). Firms are heterogeneous in the volatility of their demand shock (σ_j), which determines the expected profit flow under each contract type. Flow profits depend on how actual hours h deviate from a stochastic target h-tilde via a quadratic loss specification. Market tightness θ is determined endogenously by free entry. The key friction is random search: workers cannot direct their search to their preferred contract type, so R-best workers sometimes end up in ZHCs and must search on-the-job to move to regular employment.

How are worker types identified empirically, and why only two types?

The paper adapts a mixture-of-exponential distributions procedure from Karlis and Xekalaki (1999), applied separately to the duration distribution of ZHC employment, regular employment, and unemployment in LFS data. A bootstrapped sequential hypothesis test determines the number of latent classes M* that best fits the survival function. For ZHC employment, two exponential components are needed (p-value for M=1 vs. M≥2 is 0.01; for M=2 vs. M≥3 it is 0.74). For regular employment, one component suffices (p-value for M=1 vs. M≥2 is 0.99). For unemployment, again two components (p-values 0.01 and 0.93 respectively). Cross-referencing which types are present in which states using the model’s theoretical exit-rate table rules out R-only and Z-best workers, leaving only R-best and Z-only workers as consistent with all three distributions simultaneously.

What is the identification strategy and what are the main threats to it?

Identification rests on three steps. First, the mixture-of-exponentials procedure identifies the number of worker types from shape of duration distributions; this step relies on recalled job tenure and unemployment duration, which the authors acknowledge may suffer from recall bias and heaping (rounding to salient durations). Second, the turnover parameters are calibrated by minimizing distance between model-implied and empirical transition matrices across U, Z, and R states from the longitudinal LFS; the main limitation noted is that the two moments (transitions and durations) are not jointly consistent because they come from different measurement processes. Third, flow profits and payoffs are calibrated to external moments (minimum wage, replacement rate, business creation costs) and the preference for ZHC hours; the hours volatility parameter σ_Z has no direct empirical counterpart and is varied across scenarios. The model abstracts from wage bargaining, treating wages as fixed at the minimum wage, which reduces scope for confounding but is an approximation even in the low-wage sector.

How are the three channels — job creation, substitution, and labour-force participation — distinguished in the quantitative analysis?

The job-creation channel is captured by Z-only firms (firms with σ_Z = 6 such that regular employment is not profitable): removing ZHCs forces these firms out of the market entirely, reducing labour market tightness θ and hence the aggregate job-finding rate λ(θ). The substitution channel is captured by Z-best firms (σ_Z = 3): these firms could profitably hire under regular contracts but choose ZHCs, and after a ban they convert vacancies to regular posts, with incomplete crowd-out due to general equilibrium adjustment. The labour-force-participation channel is captured by Z-only workers: those with disutility α^i above the threshold (WTP > £7.9 per week to avoid regular work) withdraw from the labour force when ZHCs are banned, while those below the threshold remain and take regular jobs. The paper runs scenarios that vary both the firm side (low vs. high volatility) and the worker side (low vs. high disutility) to disentangle the magnitude of each channel.

What is the decomposition of the effect on regular employment (Proposition 4a)?

Under the calibrated parameters (no Z-best workers), regular employment in the baseline relative to the no-ZHC counterfactual equals the product of three multiplicative terms. The job-creation term is λ(θ)/λ(tilde-θ) = 1/0.71 ≈ 1.41, meaning that ZHCs raise the job-finding rate by about 41 percent relative to the no-ZHC counterfactual. The vacancy-competition term vR/v ≈ 0.81 (80.6 percent of vacancies are for regular jobs, while the remaining 19.4 percent for ZHC jobs dilute the pool). The search-efficiency term captures the fact that some R-best workers are in ZHC employment and search on-the-job at reduced intensity x < 1. The ceteris paribus decomposition at the ban scenario indicates: job creation alone would cut regular employment by 29 percent; competition reduction adds 24 percent; and search-efficiency gains add 15 percent — so the post-ban equilibrium has higher regular employment despite worse job creation overall.

How does the paper handle the partial versus general equilibrium distinction for welfare?

For R-best workers, the PE experiment replaces their ZHC jobs with regular jobs while keeping all other equilibrium objects (tightness θ, vacancy composition, etc.) fixed. This isolates the substitution effect and yields a welfare gain of approximately +0.15 to +0.18 percent for R-best workers. In general equilibrium, the full ban requires θ to fall (less job creation), which extends unemployment spells, and the net welfare effect is −0.50 to −0.62 percent. The difference between GE and PE therefore quantifies the job-creation externality that ZHCs provide — approximately 0.65 to 0.80 percentage points of consumption equivalent variation for R-best workers. For Z-only workers, the PE experiment replaces ZHC jobs with non-employment (their next-best option in the baseline), yielding PE welfare changes of −2.94 to −3.28 percent, which overstates the GE loss (−1.65 to −2.0 percent) because GE adjustment allows some Z-only workers to take regular jobs, partially compensating for the loss of ZHC access.

What heterogeneity is documented in the data for U.K. ZHC workers?

ZHC employment is concentrated at both ends of the age distribution: workers aged 16–29 are over-represented, as are workers aged 55–69, relative to regular employment. Mean age is 40.8 years for ZHC workers vs. 46.3 for regular workers. Gender composition is similar: 56.5 percent female in ZHCs vs. 60.4 percent female in regular employment, a difference that is not statistically significant. Educational attainment distributions are similar: 21.9 percent of ZHC workers hold a degree vs. 18.0 percent of regular workers. By industry, ZHC employment is heavily concentrated in Accommodation and food services (19.9 percent), Health and social work (20.5 percent), and Arts, entertainment and recreation (6.7 percent). Average hours worked are 18.4 per week for continuously employed ZHC workers vs. 28.1 for regular contract workers; the standard deviation of hours is 7.8 vs. 7.2. 16.6 percent of ZHC workers report wanting more hours vs. 10.1 percent in regular contracts, and 18.2 percent of ZHC workers are looking for another/additional job vs. 5.0 percent of regular workers, suggesting a minority are in involuntary underemployment while a majority are not actively seeking to change.

What are the key calibrated parameter values and how do they compare to the broader literature?

The biweekly job-finding rate λ(θ) = 0.051; the biweekly job-destruction probability δ = 0.005; on-the-job search efficiency x = 0.352 (authors note this is on the high end but consistent with estimates accounting for flexible work); share of R-best workers ζ_{R-best} = 0.969; share of type-R vacancy-posting firms γ_R = 0.950. The matching function elasticity ψ = 0.65 (estimated from U.K. data, range 0.60–0.70, higher than the commonly used 0.50 but consistent with bias-corrected estimates from Borowczyk-Martins et al. 2013). The job-filling rate is 0.21 per biweek, consistent with Kuhn et al. (2021) U.K. estimates of 0.35–0.38 per month. The vacancy posting cost κ = £36.3 per week and startup cost K = £4,376, the latter close to the £4,500 implied by U.K. business creation data. Non-employment income b = £148.8 per week (replacement ratio 80 percent). The minimum wage is set to £7.50 per hour (2017 U.K. National Living Wage); labour productivity p = £8.25, implying a 10 percent productivity premium over the minimum wage.

What robustness checks are run, and do the main results change?

The authors run three main robustness analyses. First, they vary the hours parameters: an alternative calibration uses σ_Z = 4.5 for both firm types but differentiates by mean hours (µ_Z = 20 for Z-best, µ_Z = 16 for Z-only); employment and unemployment effects are modestly smaller than the baseline but welfare effects are nearly identical. Second, they hold µ_Z = 18 and vary σ_Z to 1.0 (low) and 8.0 (high); results move in the expected direction and remain broadly consistent. Third, they vary the targeted job-filling rate: at λ(θ)/θ = 0.16 (25 percent lower than baseline), the unemployment response to a ZHC ban is only 0.33–0.51 p.p. and GDP effects are positive in the low-disutility case; at λ(θ)/θ = 0.26 (25 percent higher), unemployment rises by 4.1–5.5 p.p. and sectoral GDP falls by up to 6 percent. The authors conclude that the baseline calibration of 0.21 is the most plausible. The qualitative conclusions — that GE welfare effects are negative for all workers — are robust across specifications.

How does this paper relate to and differ from closely related prior work?

The closest model-based study is Scarfe (2019) on casual work in Australia. Scarfe’s model features homogeneous agents ex ante, with contract choice driven by luck (stochastic match productivity), while Dolado et al. emphasise ex ante heterogeneity in preferences/profitability as the primary source of variation. The empirical study of Datta et al. (2019) documents U.K. ZHC characteristics using LFS, online survey, and matched employer-employee data from the social care sector; Dolado et al. use the LFS but impose structural discipline to recover preference parameters and conduct GE welfare analysis. The paper differs from the dual labour market literature (Cahuc et al. 2016, 2020; Créchet 2022) in that temporary jobs in that literature have a fixed expiration date, whereas ZHCs are jobs with potentially long tenure but endogenously lower expected duration due to on-the-job search quit-outs, not contractual termination. Mas and Pallais (2017) and Angelici and Profeta (2020) use field experiments to estimate workers’ valuation of flexibility; Dolado et al. instead recover this from duration distributions, allowing for general equilibrium job-creation and participation effects that field experiments cannot capture.

What are the sorting patterns in the equilibrium, and what sustains ZHC jobs?

In the baseline equilibrium, 66.8 percent of filled ZHC jobs are held by R-best workers (workers who prefer regular employment but accept ZHCs as a stepping stone). Only 4.8 percent of employed R-best workers are in ZHCs at any point in time, because most vacancies are for regular jobs (80.6 percent of vacancies). This sorting has a crucial implication: ZHC vacancies would not be viable without the presence of R-best workers, because Z-only workers alone are too few to sustain the ZHC sector in equilibrium. A firm posting a ZHC vacancy accepts a higher worker-turnover risk (R-best workers quit on-the-job once they find a regular vacancy) in exchange for the profit advantage of hours flexibility; the trade-off is viable only because the random search pool contains enough R-best workers willing to take ZHC jobs temporarily.

What are the policy implications and their scope conditions?

The paper identifies four recommendations. P1: restrict ZHCs to matches where the worker voluntarily chooses the flexible contract when offered a choice; this would protect R-best workers who currently end up in ZHCs due to search frictions from the substitution effect without eliminating the job-creation channel. P2: prioritise access to ZHCs for small firms (as a proxy for inability to diversify demand shocks), limiting substitution by large firms while preserving genuine job creation by high-volatility operators. P3: recognise that the allocation of hours-flexibility between firms and workers is often an implicit and incomplete contract rather than an explicit one. P4: regulate the sharing of hours flexibility — specifically, who controls the timing and quantity of work — to reduce the income uncertainty that generates the main political objections to ZHCs. The scope conditions for all recommendations are: the low-wage sector of the U.K. labour market; the results do not directly apply to higher-wage workers with more bargaining power, or to markets where exclusivity clauses remain common.

What key empirical facts about ZHC flows does the paper document?

From the transition matrix estimated from LFS data: 11 percent of exits from unemployment are to ZHC employment. The rate of transition to unemployment is almost 50 percent larger in ZHC employment than in regular employment (6.2 percent vs. 4.4 percent semi-annually). Job-to-job transitions from ZHC to regular employment are 6.5 percent semi-annually; the reverse (regular to ZHC) is only 0.5 percent. Nearly half of ZHC workers report job tenures longer than two years. 9.2 percent of ZHC workers were recruited in the last three months vs. 3.4 percent of regular workers; 30.3 percent of ZHC workers have been with their employer less than one year vs. 14.3 percent in regular contracts. The non-employment rate for this low-pay segment is 11.2 percent; ZHCs account for 4.6 percent of the overall sample (5.2 percent of employees), about 1.5 times the aggregate U.K. incidence rate.

What does the model say about time spent out of regular employment following a ZHC ban?

Despite higher aggregate unemployment rates after the ban, R-best workers spend less total time out of regular employment: the duration of non-regular-employment spells decreases by 7 weeks. This is because ZHCs, by acting as a stepping stone, expose workers to more frequent labour market transitions — they cycle through unemployment, ZHC employment, and regular employment rather than simply unemployment and regular employment. The ban removes the ZHC stepping stone, so workers face longer individual unemployment spells but avoid the ZHC-employment phase, and on net spend more time in regular employment. However, this does not translate into a welfare gain because (a) ZHC employment, even if imperfect, provides utility above the unemployment level, and (b) the longer unemployment spells that do occur under a ban are more costly than the shorter ZHC spells they replace.

Key Concepts

Zero-hours contract (ZHC): In the paper’s sense, an employment arrangement under which the employer is not obligated to provide any minimum guaranteed hours of paid work, and the worker is not required to accept any hours offered. Workers on ZHCs in the U.K. hold ‘worker’ status (between employee and self-employed), entitling them to holiday pay, minimum wage protections, and Universal Credit, but not redundancy pay. The key feature for the model is that actual hours worked equal the firm’s demand realisation, eliminating the quadratic deviation costs that arise under fixed-hours regular contracts.

R-best workers: In the paper’s worker taxonomy, individuals for whom the asset value of regular employment strictly exceeds that of ZHC employment, which in turn exceeds the asset value of non-employment (W^i_R > W^i_Z > N^i). These workers accept ZHCs as a stepping stone when regular jobs are unavailable, and search on-the-job (at reduced efficiency x) for regular vacancies. They constitute 96.9 percent of the low-wage sector in the calibration and account for two-thirds of filled ZHC jobs.

Z-only workers: Workers for whom the asset value of ZHC employment exceeds both the value of regular employment and non-employment (W^i_Z > N^i > W^i_R, or W^i_Z > W^i_R > N^i), and who prefer non-employment to regular work. Without ZHCs, these workers’ participation in the labour market depends on whether their disutility parameter α^i implies ω^i_R > ω^i_N. A subset — those with high disutility (WTP > £7.9 per week to avoid regular work) — exit the labour force if ZHCs are banned, generating the participation effect.

Z-only firms: In the paper’s firm taxonomy, firms with high demand volatility (σ_Z = 6 in the calibration) for which regular employment is not profitable (V^j_R < 0 < V^j_Z). These firms can only operate and post vacancies because ZHCs allow them to set actual hours equal to realised demand. A ban on ZHCs causes Z-only firms to exit entirely, generating the pure job-creation loss.

Z-best firms: Firms with moderate demand volatility (σ_Z = 3 in the calibration) that could profitably post regular vacancies (V^j_R > 0) but prefer ZHC vacancies because the hours-flexibility profit advantage outweighs the higher quit risk from R-best workers. A ban redirects these firms to regular contracts, constituting the substitution effect on the firm side.

Stepping-stone effect: The mechanism by which R-best workers accept ZHC employment when unemployed, using it as a bridge to search on-the-job for regular employment. ZHCs therefore simultaneously reduce unemployment duration and extend the time workers spend out of regular employment. The paper documents that a ZHC ban reduces total time out of regular employment by 7 weeks for R-best workers despite raising the unemployment rate, precisely because the stepping-stone pathway — which adds a ZHC phase before reaching regular employment — is eliminated.

Consumption equivalent variation (welfare measure): The percentage permanent change in consumption that would make a worker indifferent between the baseline equilibrium (with ZHCs) and the counterfactual (ZHC ban). The paper uses this metric to express welfare effects: R-best workers suffer losses of −0.50 to −0.62 percent, and Z-only workers suffer losses of −1.65 to −2.0 percent, in general equilibrium following a ZHC ban.

Mixture-of-exponentials identification of worker types: A statistical procedure adapted from Karlis and Xekalaki (1999) that fits the empirical distribution of job tenure or unemployment duration as a mixture of M exponential distributions. Each component corresponds to a latent class of workers exiting the labour market state at a distinct rate. The optimal number of components M* is chosen via a bootstrapped sequential hypothesis test. Applied to U.K. LFS data, the procedure identifies M* = 2 for ZHC employment and unemployment, and M* = 1 for regular employment, which the model interprets as evidence for R-best and Z-only worker types.