Discussion Paper

No. 2016-24 | June 15, 2016
Job Placement Agencies in an Agent-based Model of the Local Labor Market with the Long-term Unemployed and on-the-job Flows


In this paper, an agent-based search model of the labor market with heterogeneous agents and an on-the-job search is developed, i.e. the long-term unemployed and other job seekers compete for vacancies which differ in skills demands and in the sector of the economy. Job placement agencies help both types of unemployed persons find the proper vacant job by improving their search effectiveness and by sharing leveraged job advertisements. The agents’ interactions take place in an artificial world drawn from labor market search theory. Six global model parameters were calibrated with the Latin hypercube sampling technique for one of the largest urban areas in Poland. To investigate the impact of parameters on model output, two global sensitivity analysis methods were used, i.e. Morris screening and Sobol indices. The results show that both programs considerably influence unemployment and long-term unemployment ratios as well as the level of wages, duration of unemployment, skills demand and worker turnover. Moreover, strong cross-effects were detected: programs aimed at one group of job seekers affect other job seekers and the whole economy. This impact is sometimes positive and sometimes it is negative.

Data Set

JEL Classification:

C63, C69, J48, J63, J64


  • Downloads: 779


Cite As

Marcin Wozniak (2016). Job Placement Agencies in an Agent-based Model of the Local Labor Market with the Long-term Unemployed and on-the-job Flows. Economics Discussion Papers, No 2016-24, Kiel Institute for the World Economy. http://www.economics-ejournal.org/economics/discussionpapers/2016-24

Comments and Questions

Stefano Staffolani - Referee Report 1
July 12, 2016 - 12:56

See attached file

Marcin Wozniak - Reply to Referee Report 1
July 18, 2016 - 10:46

Below I attach the pdf file with responses to your remarks. Hope my answers have sufficiently addressed your concerns and your feedback is sincerely appreciated.

Anonymous - Referee Report 2
July 13, 2016 - 12:45

The topic is interesting, the project is well defined but the paper is badly written and imprecise. Take Section 5 as an example. Most of the authors quoted are not in the refences or if they are have different dates. Furthermore, the title should be shorter and the introduction reformulated.

Marcin Wozniak - Reply to Referee Report 2
July 18, 2016 - 12:33

Thank you very much for reading my paper and making remarks.
I would like to refer briefly to the comments at a few following points:
1) I admit that there are some discrepancies between the table of references and the main text. The references will be corrected immediately.
2) It ...[more]

... is a good idea to make the title shorter. I consider the change in the revised version of the manuscript.
3) According to your suggestion, I plan some reformulations in the introduction. A more place in the section should be devoted to the justification of choosing the agent-based methodology and strengths of this approach.
4) I also acknowledge the remarks on the language of the paper. Be sure I am aware of the importance of the proper writing style. In fact, I am not an English speaking person, but the manuscript was proofread by a certified professional service, which in my opinion should be enough.

Josef Orczyk - Comment
July 15, 2016 - 11:26

See attached file

Marcin Wozniak - Reply to the comments
July 19, 2016 - 12:57

Many thanks for appreciating essential features of the paper.
I recognize that most of the pointed disputes should be resolved:
1) A number of job-seekers and number of firms were set to capture the dependency between the actors in the simulated labor market. The empirical proportion is lower (5 job-seekers ...[more]

... per 1 firm) than the relation in the model. However, in the model, a maximum number of vacancies the firm can open is three. In real, the firms can create as many vacancies as they want. As we do not dispose of the detailed vacancies statistics, I assume that the proportion should be bigger in the model.
2) 5 skills levels correspond with 5 education stages in Poland, which was distinguished on a basis of International Standard Classification of Education. We can then write Level 1 as a Primary School; Level 2 as Middle School; Level 3 as Vocational School; Level 4 as High School and Level 5 as Higher Education.
3) The size of the world (number of patches) was adjusted according to two criteria 1) reasonable time execution of a simulation; 2) accuracy of the results and possibility of free movements of the agents on the grid.
The bigger world needs usually fewer repetition and should provide better results, however, the simulation time increases significantly with the number of patches. Thus I needed to find the golden mean. In that case, it was the grid of 20 patches. With this setup, one period of the model run lasts approximately 1 second. A single execution of the simulation (156 periods) takes ~156 seconds. Consider Morris screening of the model. To obtain comparable results we should use ~1000 repetitions of the simulation. Total execution time would be ~156000 seconds = 43 hours.
4) Of course, all the editorial issues will be fixed.
5) Finally, I must consider some reformulations in the literature review point, all the more so the third Referee also suggests modifications in the section.

International Standard Classification of Education 2011, UNESCO.

Local Data Bank of Poland (https://bdl.stat.gov.pl/BDL/start).

Oremland, Matthew and Laubenbacher, Reinhard (2014) 'Optimization of Agent-Based Models: Scaling Methods and Heuristic Algorithms' Journal of Artificial Societies and Social Simulation 17 (2) 6.

Dominik Buttler - comments
July 15, 2016 - 18:38

in the attachment

Marcin Wozniak - Some remarks on labor market search models
July 21, 2016 - 14:42

I appreciate that you read my study so carefully and made new insight into the discussion. In the next version of the paper, I will surely try to establish a deeper link with the ALMP evaluation studies. I would like to refer only to one of the points you have ...[more]

... made which I strongly disagree. You wrote 'In my opinion DMP models are less useful as a policy evaluation tool due to the numerous simplifying assumptions(...)'. Let me refer to the sentence in three arguments:
1) Existing research in labor economics proves that the search and the matching models are a natural and powerful framework for labor market policy evaluation (e.g. Card et al. 2010; Michaillat 2012). They can also shed light on how the labor markets function when policy externalities arise (policy side-effects) (Crepon et al. 2012).
2) Simplifying assumptions are the basis of the most economic theories, which, nonetheless, still works fine. ;-)
3) It must be also stressed that the model in the paper is not an aggregate DMP but agent-based implementation. Remind some advantages of ABM comparing to classic approach i.e. 'real' heterogeneity, spatial aspect of simulation, memory-based decision-making of agents.

Card, D., Kluve, J. and Weber, A. (2010), Active Labour Market Policy Evaluations: A Meta-Analysis. The Economic Journal, 120, pp. 452–477.

Crépon B., Duflo E., Gurgand M., Rathelot R., Zamora P. (2012), Do labor market policies have displacement effects? Evidence from a clustered randomized experiment, National Bureau of Economic Research, Working Paper 18597, Cambridge, MA.

Michaillat P. (2012), Jobs in a recession, CentrePiece - The Magazine for Economic Performance 365, Centre for Economic Performance, LSE.

Anonymous - Referee Report 3
July 19, 2016 - 10:23

see attached file

Marcin Wozniak - Reply to Referee Report 3
July 25, 2016 - 13:27

Please, see attached pdf file. Hope my responses are comprehensible. If you have any further comments or questions, I would be glad to address them.

Anonymous - Comments
July 21, 2016 - 14:05

See in attachment.

Marcin Wozniak - Reply to the comments
July 22, 2016 - 11:42

I want to thank you for spending your time reading and commenting my paper. I am aware of the fact that the problem is not the main domain of your interests all the more I appreciate your contribution.

Roger Bivand - Invited Reader Comment
July 28, 2016 - 08:28

This discussion paper covers a number of topical areas with regard to the analysis of active labour market programmes (ALMP) and long-term unemployment (LTU) in a large urban labour market. One major topical area is the development of an agent-based model of the individual-level decisions made by those in work, ...[more]

... those unemployed, and among these long-term unemployed. Another is the calibration of the model largely using NUTS2 regional data, while a third is the examination of the sensitivity of model output to changes in calibration and treatment settings.

The discussion of why to use an agent-based approach is sound and convincing, although it seems that the discussion of closeness in local submarkets suggested on p. 6 (neighbourhood patches) does not seem to be followed up. It is easy to see that this could add too much detail to the model, but maybe the use (or not) of neighbourhood patches deserves clarification. Further points might be to discuss age and/or gender decomposition of the labour market, as for example maternity leave is a major issue in this (and other) labour markets.

With regard to calibration, it is encouraging to see the care with which available data are used. The comment regarding the probably fictitious nature of the minimum wage is well taken. How much do we know about multiple employment, or labour-only one-person firms? For readers who are unaware of the relative stability of this labour market through the financial crisis, it might be worth discussing the external shock link to the crisis years (which I believe was much less than might have been expected, in all three considered sectors, possibly partly because of the buffer given by the PLN/EUR rate, and by the use of EU structural fund transfers; note also EURO 2012, which involved a good deal of infrastructure investment).

The model runs, results and sensitivity tests are described quite briefly, given the complexity of the output. In microsimulation, one might use stereotypical "biographies" to describe agents' "experiences", but whether this would help here is unknown (say start with a skilled 25-year-old changing jobs on-the-job in services with a specific skill set, and add other "characters"?). Maybe such "stories" could complement the rather rapid discussion of the results stemming from the sensitivity analyses.
The availability of code files is valuable, although I have not attempted to re-run any analyses.

I did notice that the references are a little disorganised towards the foot of p. 32 (and I couldn't find Wallach et al. 2006 referenced on p.
21). Might it be sensible to include citations in the references to the R packages used here and listed on p. 2?

Marcin Wozniak - Reply to the comment
August 01, 2016 - 09:06

Thank you very much for adding your voice to the discussion. You have pointed a few important features that may extend the agent-based model presented in the paper. I must admit that the biographies of agents have inspired me. I think they may be an interesting and unique element of ...[more]

... the paper. I have one doubt which is the size of the paper. As I am revising the text, the number of pages is increasing continuously. Owing to the fact, I must consider the pros and cons of adding another component to the manuscript.