Job Protection — It is Good to be an Insider

The value, VHAH,nPHO,j−,nPHN,j−,j {V_H}\left( {{A_H},n_{PHO,j}^ - ,n_{PHN,j}^ - ,j} \right) for a high-productivity firm that has survived j periods has the recursive form: (2.3) VHAH,nPHO,j−,nPHN,j−,j=AH(nPHO,j+nPHN,j+nTH)α−nPHN,jwPHN−nTHwT−nPHO,jwPHO− nPHN,j−nPHO,j−−nPHN,j−cqθP−nTH+cqθT+ βδπHVHAH,nPHO,j+ζnPHN,j,1−ζnPHN,j,j+1+ βδ1−πHVLAL,nPHO,j+ζnPHN,j,1−ζnPHN,j,j+1 ∀j \matrix{ {{V_H}\left( {{A_H},n_{PHO,j}^ - ,n_{PHN,j}^ - ,j} \right) = } \hfill & {{A_H}{{({n_{PHO,j}} + {n_{PHN,j}} + {n_{TH}})}^\alpha } - {n_{PHN,j}}{w_{PHN}} - {n_{TH}}{w_T} - {n_{PHO,j}}{w_{PHO}}} \hfill \cr {} \hfill & { - \;\left( {{n_{PHN,j}} - n_{PHO,j}^ - - n_{PHN,j}^ - } \right){c \over {q\left( {{\theta _P}} \right)}} - n_{TH}^ + {c \over {q\left( {{\theta _T}} \right)}}} \hfill \cr {} \hfill & { + \;\beta \delta {\pi _H}{V_H}\left( {{A_H},{n_{PHO,j}} + \zeta {n_{PHN,j}},\left( {1 - \zeta } \right){n_{PHN,j}},j + 1} \right)} \hfill \cr {} \hfill & { +\, \beta \delta \left( {1 - {\pi _H}} \right){V_L}\left( {{A_L},{n_{PHO,j}} + \zeta {n_{PHN,j}},\left( {1 - \zeta } \right){n_{PHN,j}},j + 1} \right)\;\;\;\;\;\;\;\;\;\;\forall j} \hfill \cr }

Where n_TH and n_PHN,j decision variables of the firm. n_TH denotes temporary workers hired in the current period. n_PHN,j will be the number of workers that the firm will have at the end of the period, i.e. any new hires and nPHN,j− n_{PHN,j}^ - . nPHN,j− n_{PHN,j}^ - is the number of workers remaining in less protected permanent contracts that the firm starts with from the previous period. nPHO,j− n_{PHO,j}^ - indicates the permanent workers that obtained full protection in any of the j previous period which since in the high productivity state no worker is fired is equal to n_PHO,j, the number of fully protected workers working this period at this firm. Clearly, a firm that is looking to hire, will be able to save hiring costs if it already has a stock of workers nPHO,j−+nPHNj−, n_{PHO,j}^ - + n_{PHN,j}^ - , . The relative number of less and fully protected permanent workers depends on how many periods the firm has survived and will be presented in section 2.4 in detail. ζ is the transition rate of permanent workers to the more protected state and in the final line of (2.3) we already substituted the number of permanent workers with either level of protection as they will be present at the start of the next period constituting nPHO,j+1− n_{PHO,j + 1}^ - and nPHN,j+1− n_{PHN,j + 1}^ - (in case of high productivity) or nPLO,j+1− n_{PLO,j + 1}^ - and nPLN,j+1− n_{PLN,j + 1}^ - (in case of low productivity). Wages are denoted w_PHN, w_PHO and w_T where the subscripts P and T denote permanent and temporary workers and the subscripts N and O denote the less and more protected permanent workers. The H subscript denotes the productivity of the firm in the current period (suppressed for temporary workers as they only exist in high productivity stats. The future values are discounted at β and δ (discounting and destruction probability) and the relative number of different permanent workers in the future are determined by how many permanent workers transition at probability ζ into the more protected status. Let us note again that this means that a firm starts any period with a certain number of permanent workers with either f₁ or f₂ protection, keeps them in both states of productivity (and in equilibrium does not hire any additional permanent workers, unless it is the first period after creation), and then at the end of the period, the less protected permanent workers transition with probability ζ to the higher protection and then the firm enters with this new mix of permanent workers into the new period. Temporary and permanent contracts are offered to indifference, and firms are bound by the equal treatment clause for new permanent hires16

This ensures firms cannot negotiate an upfront payment prior to concluding a permanent contract.

Where the decision variables of the firm are n_{PLN j}, and n_PLO,j which are analogous to above the numbers of workers of either f₁ or f₂ protection at the end of this period. Essentially, the firm is deciding how many of these workers to keep. Notation follows the same convention as above. A few differences to (2.3) merit comment. The indicator functions in the fourth and fifth term mean that firms only consider the firing taxes at the margin if they actually downsize and the numbers at the end of the period of permanent workers of either protection level is below the number the firm started with in this period, nPLN,j− n_{PLN,j}^ - and nPLO,j− n_{PLO,j}^ - . This means the firm does not have to consider firing taxes as far as size decisions are concerned as, during hiring, they were not hiring beyond the point where a permanent worker would get paid more than they produce at low productivity. The firing costs do however feature in the wage negotiation. There will be no temporary workers since hiring them incurs higher costs than retaining a permanent worker in a period of low productivity.

These value functions are derived assuming permanent workers are never fired unless the firm ceases to exist. We also do not refer to the size of the firm, since there are only two sizes that map one to one to the productivity. The value function for unemployment, Y (after matching happened and, thus, represents the fallback) satisfies: (2.5) Y=βpθPWPHN+β1−pθPY=βpθTWTH+β1−p(θT)Y {\rm{Y}} = \beta p\left( {{\theta _P}} \right){W_{PHN}} + \beta \left( {1 - p\left( {{\theta _P}} \right)} \right.\;{\rm{Y}} = \beta p\left( {{\theta _T}} \right){W_{TH}} + \beta \left( {1 - p({\theta _T})\;{\rm{Y}}} \right. where W_PHN and W_TH are the values of being employed as a newly hired permanent worker (i.e. with low protection f₁) and being temporarily employed, respectively. Slightly transformed to: pθPWPHN+(1−pθPY=pθTWTH+(1−pθTY p\left( {{\theta _P}} \right){W_{PHN}} + (1 - p\left( {{\theta _P}} \right){\rm Y} = p\left( {{\theta _T}} \right){W_{TH}} + (1 - p\left( {{\theta _T}} \right){\rm Y} the equation shows that applying for a temporary or a permanent job needs to ex ante provide the same payoff. It implicitly defines the labor market tightnesses for permanent and temporary positions as θP=VPμU {\theta _P} = {{{V_P}} \over {\mu U}} and θT=VT1−μU {\theta _T} = {{{V_T}} \over {\left( {1 - \mu } \right)U}} where V_P and V_T are aggregate numbers of permanent and temporary vacancies (to be derived below), and U is the number of aggregate unemployed. Again, we suppress time subscripts since we solve for the Steady State. Also, note that in equilibrium low productivity firms do not hire workers at all so workers do not enter these from unemployment but rather transition into low productivity with an employer they matched with before. μ is the proportion of unemployed electing to apply for a permanent position. For this proportion to be stable, the above condition needs to hold, equalizing ex ante returns from applying for any type of position. Value functions for protected/permanent workers are: (2.6) WPHN=wPHN+1−ζβδπHWPHN+ζβδπHWPHO+ζβδ1−πHWPLO+ 1−ζβδ1−πHWPLN+β1−δY \matrix{ {{W_{PHN}} = } \hfill & {{w_{PHN}} + \left( {1 - \zeta } \right)\beta \delta {\pi _H}{W_{PHN}} + \zeta \beta \delta {\pi _H}{W_{PHO}} + \zeta \beta \delta \left( {1 - {\pi _H}} \right){W_{PLO}}} \hfill \cr {} \hfill & { + \;\left( {1 - \zeta } \right)\beta \delta \left( {1 - {\pi _H}} \right){W_{PLN}} + \beta \left( {1 - \delta } \right)\;{\rm{Y}}} \hfill \cr } (2.7) WPNO=wPHO+βδπHWPHO+βδ1−πHWPLO+β1−δY {W_{PNO}} = {w_{PHO}} + \beta \delta {\pi _H}{W_{PHO}} + \beta \delta \left( {1 - {\pi _H}} \right){W_{PLO}} + \beta \left( {1 - \delta } \right)\;{\rm{Y}} (2.8) WPLN=wPLN+1−ζβδπHWPHN+ζβδπHWPHO+ζβδ1−πHWPLO+ 1−ζβδ1−πHWPLN+β1−δY \matrix{ {{W_{PLN}} = } \hfill & {{w_{PLN}} + \left( {1 - \zeta } \right)\beta \delta {\pi _H}{W_{PHN}} + \zeta \beta \delta {\pi _H}{W_{PHO}} + \zeta \beta \delta \left( {1 - {\pi _H}} \right){W_{PLO}}} \hfill \cr {} \hfill & { + \;\left( {1 - \zeta } \right)\beta \delta \left( {1 - {\pi _H}} \right){W_{PLN}} + \beta \left( {1 - \delta } \right)\;{\rm{Y}}} \hfill \cr } (2.9) WPLO=wPLO+βδπHWPHO+βδ1−πHWPLO+β1−δY {W_{PLO}} = {w_{PLO}} + \beta \delta {\pi _H}{W_{PHO}} + \beta \delta \left( {1 - {\pi _H}} \right){W_{PLO}} + \beta \left( {1 - \delta } \right)\;{\rm{Y}} where the first value function is for newly hired permanent workers, and the second is for permanent workers after their probation period. The third value function is for new permanent workers in less productive firms (they do not exist in equilibrium), the fourth for permanent workers after their probation period in a low productivity firm. Lower case w's denote wages where the subscripts are analogous to the value subscripts. Workers that go into unemployment remain unemployed for one period and only then may find another job. We see the probability for the less protected workers to transition to more protection, ζ, present in the value function for the less protected workers. The probabilities, π_H and π_L of the firm entering or remaining in higher or lower productivity are also reflected in all the value functions as this has a direct effect on the wage of the workers. Finally, in case of destruction of the firm, common in all above value functions, with probability 1 − δ, even permanent workers return to unemployment. For temporary workers, the value function is: (2.10) WTH=wTH+βY {W_{TH}} = {w_{TH}} + \beta {\rm{Y}}

For these workers, they essentially get their wages and immediately return to unemployment in the next period, reflecting the precariousness of temporary employment.

Whether total employment increases or decreases after a rise in f₁ depends on market tightness (amplified if the elasticity of matching w.r.t. vacancies is larger) and, therefore, on the cost of recruitment, as is the case under the CTP policy. If the latter drops sufficiently, we see an increase in employment. This is because firms that compete too aggressively impose negative congestion externalities upon each other in the matching. f1 acts as an indirect hiring tax and mitigates this dynamic. Since ∂nPHN⋆+nTH∂f1=11−αnPHN⋆+nTHa {{\partial \left( {n_{PHN}^ \star + {n_{TH}}} \right)} \over {\partial {f_1}}} = {1 \over {1 - \alpha }}\left( {n_{PHN}^ \star + {n_{TH}}} \right)a where a is a fraction whose denominator is positive and whose numerator is given below in (3.14) and can be either positive or negative, such that the reduction in hiring costs outweighs the increase in firing taxes. So, we can derive the following relationship: (3.14) 0=ηcβ∂θP∂f1+βpθP+βf1∂pθP∂θP∂θP∂f1+1−δβ+βδζ1−η−1−δβπHcqθP2∂qθP∂θP∂θP∂f1 0 = \eta \left( {c\beta {{\partial {\theta _P}} \over {\partial {f_1}}} + \beta p\left( {{\theta _P}} \right) + \beta {f_1}{{\partial p\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} + \left( {1 - \delta \beta } \right) + \beta \delta {\zeta \over {1 - \eta }}} \right) - \left( {1 - \delta \beta {\pi _H}} \right){c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} defining f1⋆ f_1^ \star beyond which an increase in f₁ decreases employment.

An initial observation is that there is a close connection between the proportion of new permanent workers and the wages that are paid to those workers. If wages increase and total employment goes down, the model makes clear predictions. If, however, the hiring costs decrease sufficiently, there may be cases that are consistent with lower wages as well as lower proportions of permanent workers. In what follows, we outline three cases. To establish the cases, we consider the following two equations that are the derivative of the employment decisions for nPHN⋆ n_{PHN}^ \star and nPHN⋆+nTH n_{PHN}^ \star + {n_{TH}} , respectively, that is, the number of permanent workers and total employment in a high-productivity firm. Both are rearranged to yield the elasticity of employment relative to firing taxes. (3.15) f1∂nPHN⋆∂f1nPHN⋆=−f11−αηcβ∂θP∂f1+βpθp+βf1∂PθP∂θP∂θP∂f1+1−δβ+βδς1−n+δβπHcqθP2∂qθP∂θP∂θP∂f1ηβPθPcqθP+f1+1−βδf1−βδςf2−f11−η−βδπcHqθP {{{f_1}{{\partial n_{PHN}^ \star } \over {\partial {f_1}}}} \over {n_{PHN}^ \star }} = \left( {{{ - {f_1}} \over {1 - \alpha }}} \right){{\left[ {\eta \left( {c\beta {{\partial {\theta _P}} \over {\partial {f_1}}} + \beta p\left( {{\theta _p}} \right) + \beta {f_1}{{{\partial _P}\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} + \left( {1 - \delta \beta } \right) + \beta \delta {\varsigma \over {1 - n}}} \right) + \delta \beta {\pi _H}{c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}}} \right]} \over {\eta \left( {{\beta _P}\left( {{\theta _P}} \right)\left( {{c \over {q\left( {{\theta _P}} \right)}} + {f_1}} \right) + \left( {1 - \beta \delta } \right){f_1} - \beta \delta \varsigma {{{f_2} - {f_1}} \over {1 - \eta }}} \right) - \beta \delta \pi {c \over {{H_q}\left( {{\theta _P}} \right)}}}} (3.16) f1∂nPHN⋆+nTH∂f1nPHN⋆+nTH=−f11−αηcβ∂θP∂f1+βpθp+βf1∂PθP∂θP∂θP∂f1+1−δβ+βδς1−n−1−δβπHcqθP2∂qθP∂θP∂θP∂f1ηβPθPcqθP+f1+1−βδf1−βδςf2−f11−η−1−βδπHcqθP {{{f_1}{{\partial \left( {n_{PHN}^ \star + {n_{TH}}} \right)} \over {\partial {f_1}}}} \over {n_{PHN}^ \star + {n_{TH}}}} = \left( {{{ - {f_1}} \over {1 - \alpha }}} \right){{\left[ {\eta \left( {c\beta {{\partial {\theta _P}} \over {\partial {f_1}}} + \beta p\left( {{\theta _p}} \right) + \beta {f_1}{{{\partial _P}\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} + \left( {1 - \delta \beta } \right) + \beta \delta {\varsigma \over {1 - n}}} \right) - \left( {1 - \delta \beta {\pi _H}} \right){c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}}} \right]} \over {\eta \left( {{\beta _P}\left( {{\theta _P}} \right)\left( {{c \over {q\left( {{\theta _P}} \right)}} + {f_1}} \right) + \left( {1 - \beta \delta } \right){f_1} - \beta \delta \varsigma {{{f_2} - {f_1}} \over {1 - \eta }}} \right) - \left( {1 - \beta \delta {\pi _H}} \right){c \over {q\left( {{\theta _P}} \right)}}}}

These two equations are similar and only differ in as far as the decision on total employment considers the cost of hiring a worker in the first place, whereas the decision on permanent workers considers the situation as to how many of them would be retained. Now, we can state three results covering three cases:

This occurs if both numerators of (3.15) and (3.16) are negative, which is the case when the increase in f1 increases the wages of recently hired permanent workers more than the resulting reduction in vacancy posting reduced the aggregate costs of labour search by alleviating the congestion externality. The denominator of (3.15) is more negative than that of (3.16) due to the positive term cqθP2∂qθP∂θP∂θP∂f1 {c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} in (3.16), which increases it. Additionally, the denominator of (3.16) is larger, further reducing its numerical value and establishing that (3.15) is more negative than (3.16), which reduces the ratio of permanent workers relative to total employment.

This result is straightforward as it entails that the numerator of (3.15) is negative while that of (3.16) is positive on account of the term cqθP2∂qθP∂θP∂θP∂f1 {c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} , which captures that it has become cheaper to hire permanent workers (as well as temporary workers) but not cheap enough to reduce the wages of permanent workers in the low-productivity firms. The effect is to reduce the number of permanent workers and increase the overall hires, which naturally only occurs with an increase in temporary workers. This dynamic establishes that the ratio of permanent to total employment falls.

This is the most involved result, and there are two sub-cases involved. The ambiguity arises because if the numerator of (3.15) and (3.16) is positive, that of (3.15) is naturally larger because of the additional positive term cqθP2∂qθP∂θP∂θP∂f1 {c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} . However, it is not clear if (3.15) or (3.16) is larger since the denominator of (3.16) is larger, thus diminishing the effect. We find that which effect dominates depends on the following inequality: (3.17) −η(cβ∂θP∂f1+βp(θp)+βf1∂P(θP)∂θP∂θP∂f1+(1−δβ)+βδς1−n)+δβπHcq(θP)2∂q(θP)∂θP∂θP∂f1η(βP(θP)(cq(θP)+f1)+(1−βδ)f1−βδςf2−f11−η)−βδπcHq(θP)<−cq(θP)2∂q(θP)∂θP∂θP∂f1cq(θP) \matrix{-{{{\eta \left( {c\beta {{\partial {\theta _P}} \over {\partial {f_1}}} + \beta p\left( {{\theta _p}} \right) + \beta {f_1}{{{\partial _P}\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}} + \left( {1 - \delta \beta } \right) + \beta \delta {\varsigma \over {1 - n}}} \right) + \delta \beta {\pi _H}{c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{{\partial _q}\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}}} \over {\eta \left( {{\beta _P}\left( {{\theta _P}} \right)\left( {{c \over {q\left( {{\theta _P}} \right)}} + {f_1}} \right) + \left( {1 - \beta \delta } \right){f_1} - \beta \delta \varsigma {{{f_2} - {f_1}} \over {1 - \eta }}} \right) - \beta \delta \pi {c \over {{H_q}\left( {{\theta _P}} \right)}}}}} \hfill \cr {\;\; < - {{{c \over {q{{\left( {{\theta _P}} \right)}^2}}}{{\partial q\left( {{\theta _P}} \right)} \over {\partial {\theta _P}}}{{\partial {\theta _P}} \over {\partial {f_1}}}} \over {{c \over {q\left( {{\theta _P}} \right)}}}}} \hfill \cr }

If this inequality holds, we have a decrease in the ratio of permanent to total employment and the reverse if it does not hold. Inspection of the left-hand side and right-hand side of (3.17) reveals that the left-hand side represents a measure of the reaction of wage costs for permanent workers to firing taxes (net of any additional effects and a higher marginal product due to downsizing) while the right-hand side represents the elasticity of the hiring costs. Naturally, if the elasticity of the hiring costs with respect to firing taxes is higher, this implies that the cost of hiring workers is falling faster than the cost of retaining them. This means that the marginal hiring decision is affected more favorably than the retention decision. Once we realize that this wider hiring margin translates to more temporary workers while the retaining margin translates to more permanent workers, it is clear that a greater hiring elasticity with respect to firing taxes translates to a decrease in permanent workers relative to total employment.

One result is that when separation taxes for recent hires f₁ increase, this does not alter hazard rates for recent permanent matches or established permanent matches. This is due to the buffering effect of temporary workers in the same firms as the following paragraph outlines:

Recently hired permanent workers are not dismissed in the model at this stage as there is no instance in the model where their firm would become less productive when they are still recently hired19

Even if the firm were still in the current period when higher firing taxes were instituted, firms would not fire the permanent workers but would be forced to pay them a higher wage. Naturally, firms will also not hire more permanent workers in such a situation and will actually curtail their hiring.

In the model, separation taxes do not influence job security for permanent workers because temporary workers absorb the blow. Releasing temporary workers is cheaper. This is a result of firm's motivation to protect core workers from adverse productivity shocks (noted by, e.g. Houseman et al., 2003) to avoid separation taxes and unevenly distributing risk. Further, since temporary workers are only hired for a fixed period and would not assume another contract upon expiration, we expect no measured change in hazard rates of temporary workers.

Figure 4 shows average inflows of workers in treatment and control areas normalized by their values in December 2005, when the policy was announced. The figure shows 2,172 incoming workers in the CTP and 1,265 incoming workers in the control areas, on average. This reflects that the average control area is approximately 30% smaller than the average treatment area but, otherwise, quite similar. In fact, we see little difference in the employment trajectory before the policy is implemented (April 2006). This confirms the groups’ share trends, indicating the propensity score was successful in identifying control zones sufficiently similar to the treatment zones. Additionally, there also is no clear trend in general employment flows in the treatment zones after the policy announcement, which is disappointing, but the model is ambiguous concerning the general employment effect. There is a seasonal pattern; substantial hiring occurs in January and in July. We see a smaller spike each year, however, this may be due to the impending Great Recession. Looking at the stock of workers (Figure 5), employment declined relative to control areas after the announcement, although not statistically significantly. This (weakly) indicates reduced economic activity due to more expensive workers, which is one possibility outlined by result 1 in the model when the increased firing taxes are too large and go beyond correcting for high incentives that encourage firms to hire workers. The increased help for workers seeking jobs is not efficient. However, disentangling the positive effects of the improved job-seeking assistance and the negative effects of the firing taxes would necessitate information on the job finding rates of the unemployed, which, unfortunately, we do not have since the data only cover the employed.

Figure 4

Figure 5

The model predicts fewer permanent workers (results 1 and 2). This section provides evidence that supports this. For flows of permanent workers, the groups share the same trend; that is, fewer permanent workers are hired. There is a spike of 9% in permanent employment in the control areas in August 2006, after the implementation of the policy, that is not shared by CTP areas, although they share the same trend and hire in similar numbers. Such an event seems missing in 2007, but in September 2008, we observe a relative rise in permanent hires in the control group. Note that the significance levels of the difference between the treatment and control groups for flows29

Available upon request.

The control group has a higher proportion of permanent workers than CTP areas of between two and 10 percentage points, starting in January 2006, after the policy was reported in Le Monde (12 December 2005). This means that firms immediately adjusted their stock of permanent workers as a response to the policy. The beginning of the year is typically a time of higher permanent work contract flows, as can be seen in the seasonal pattern in Figure 4, which may have been a factor. It is further noticeable that the confidence intervals cease, including for the other group, at the time of the policy announcement. Figure 8 shows the statistical significance of the difference between the ratios of permanent workers of the treatment and control areas. The difference is not significant before the policy at any conventional significance level and, precisely when the policy is announced, the ratio in treatment areas drops and becomes statistically significantly different at the 10% level (and higher) for most months. This is evidence in favor of the model (but not exclusive to it). Results 1 and 2 predict fewer permanent workers as a response to increased initial firing taxes along with less employment. An important question is: What was the margin of adjustment? From the graphs, in particular figure 5, we see that employment always dips between November and December of any given year. This applies to temporary jobs and permanent jobs. The ratio of permanent jobs increases rather strongly at the beginning of each year, only to peter out throughout the year and regain ground once again at the beginning of the next year. The end of the year is a time when many jobs, permanent and temporary, are terminated. However, relatively more temporary jobs are dissolved automatically increasing the proportion of permanent jobs. Incidentally, the policy was announced at exactly that time (December 2005), so the firms could adjust immediately.

Figure 6

Figure 7

Figure 8

The reader may wonder, why the proportion of the stock of permanent workers in the treatment zones is falling so much relatively to the control zones in November/December of each year as we can see in figure 7. This is due to the overall reduction in hiring that we can observe in figure 5 as well as a shift of employers to use temporary workers more extensively and employ them for slightly longer (but still temporary) time horizons in response to the changed regime. This may be especially pronounced in the final months of the fiscal and budgetary year because decision makers may not have fully internalized their lower “budget” of permanent vacancies yet since the CTP was novel, but the exact dynamics are beyond the scope of this study.

We are tempted to construct graphs for all subsets of firms, but this may be misleading since there are general equilibrium effects between firms with less than and more than 1,000 workers, and in such graphs it is not possible to control for confounding factors. The aggregate of the treatment and control zones is informative since the propensity score balances potential confounding factors. This is why we move on to propensity score adjusted regressions.

For wages the regression specifications are: (5.1) log hourlywage=a1+b1X1+c1X2+kY log\;\left( {hourlywage} \right) = {a_1} + {b_1}{X_1} + {c_1}{X_2} + kY (5.2) log hourlywage=a2+b2X1+c2X2+d2X3+e2X4+kY log\;\left( {hourlywage} \right) = {a_2} + {b_2}{X_1} + {c_2}{X_2} + {d_2}{X_3} + {e_2}{X_4} + kY (5.3) log hourlywage=a3+b3X1+c3X2+d3X3+e3X4+f3X5+kY log\;\left( {hourlywage} \right) = {a_3} + {b_3}{X_1} + {c_3}{X_2} + {d_3}{X_3} + {e_3}{X_4} + {f_3}{X_5} + kY (5.4) log hourlywage=a4+b4X1+c4X2+d4X3+e4X4+f4X5+g4X6+h4X7+kY log\;\left( {hourlywage} \right) = {a_4} + {b_4}{X_1} + {c_4}{X_2} + {d_4}{X_3} + {e_4}{X_4} + {f_4}{X_5} + {g_4}{X_6} + {h_4}{X_7} + kY

The a's are intercepts. X₁ is the indicator of the match existing in a CTP zone after implementation; that is, after 15 April 2006, X₂ interacts being in one of the seven affected CTP zones after implementation with being in a permanent match. X₃ − X₆ respectively interact being in the CTP zones (after implementation) with an indicator if it is a match in a firm with fewer than 1,000 workers (the upper threshold for the CTP scheme), a permanent position in a firm with fewer than 1,000 workers, if the worker previously held a permanent job (i.e. an ‘insider’) and, importantly, an interaction of being in a CTP zone after implementation if the worker previously held a permanent job and the match itself is permanent and would be subject to higher firing taxes in case of dissolution. X₇ adds various interactions (see footnote) of being eligible, previously employed, and so on. The interaction reported in the table below is the interaction of being in an affected zone (after implementation) with an indicator if the worker was eligible for the policy before the current job; that is, the worker held a permanent job in a CTP zone before, potentially benefitted from the policy, and holds a permanent job. This variable is important if we want to determine if the policy had any effect beyond increasing dismissal taxes. Finally, kY is a vector of additional controls such as a proxy for productivity, year fixed effects, firm size, sectors of activity, age, and sex

Unreported controls in the first specification also include: age, age squared, blue or white-collar jobs, sex, a proxy for firm productivity, if the job was in manufacturing or services, if it was a permanent job, year controls, if the worker is a resident, and interactions of year and permanent indicators. I do not report the results on indicators individually, such as being in an affected zone, being a permanent worker, due to space concerns and because they do not reveal anything of interest. The second specification adds indicators of firm size, if it was a new firm, and various interaction terms. The third specification adds controls if the workers held permanent jobs before in the control group and if workers were eligible. The third specification interacts the indicators with year dummies. The fourth specification via X7 adds combinations of controls for having held a permanent job before, being eligible for the policy while holding a permanent job now, and being in the control group.

Table 2 reports the results for regressions with the log of hourly wage as a dependent variable. Flows are workers newly hired in that year while stocks denote incumbents.

The first column shows that for newly hired workers, there is no significant effect in the first (most naive) specification. However, in the second column for the stock of workers, we see that the workers who found a permanent job were able to negotiate a 7.4% higher wage (albeit only significant at 10%). This is in line with the model's prediction that permanent workers negotiate a higher salary as separation taxes increase. The second specification in the third column shows that newly hired workers in firms with less than 1,000 workers obtain a lower wage (only significant at 10%), which is surprising, but reassuringly the finding does not hold once we add further controls. There is no significant estimate for newly hired or retained workers.

The third specification explores the situation among newly hired workers. In the fifth column, it seems (significant at 10%) that permanent workers hired in firms with fewer than 1,000 workers obtain a 6.4% lower wage, which is mitigated by workers who held a permanent job before having a 2.8% higher wage. It appears that new hires are willing to take a lower wage initially to gain a permanent position but leverage the CTP. Considering the average from the first specification does not expose a clear trend; the effects seem to cancel out. However, the sixth column reveals that workers who held a permanent position previously and are now working in a CTP region will increase their wages by 5.9% in the long run (i.e. in the stock of workers), significant at 1%. This is in line with workers extracting more surplus.

Moving to the fourth specification in columns 7 and 8 that includes controls for workers eligible for the policy, we see that it is not workers eligible for the CTP who are negotiating higher wages. It is those workers who have better adherence to the labour force as they had permanent jobs in the last 12 months. Thus, these workers were eligible for the policy. The wage increases by 4% for newly hired workers and 7.5% for the stock of workers32

Firms with fewer than 1,000 workers were not significantly affected, which shows a general equilibrium effect.

In summary, there is a wage increase for permanent workers, supporting the theory as these are workers whose initial firing taxes were changed, which, with a lower ratio of permanent workers is consistent with results 1 and 2. The treatment effect for temporary workers is positive but insignificant. One caveat is that better workers may have been hired in permanent positions, inflating estimates. This cannot be tested as we cannot follow workers over the years.

The next piece of evidence is if the hazard rates do not change with dismissal taxes. Other models (e.g. Alvarez and Veracierto, 2012) predict that hazard rates decrease for permanent workers. However, in our model, permanent workers already enjoy the most stable employment relationship possible. The following proportional hazard models are fitted: (5.6) λi=λoexpb1Xi1+c1Xi2+kY {\lambda _i} = {\lambda _o}exp\left( {{b_1}{X_{i1}} + {c_1}{X_{i2}} + kY} \right) (5.7) λi=λ0expb2Xi1+c2Xi2+d2Xi3+e2Xi4+kY {\lambda _i} = {\lambda _0}exp\left( {{b_2}{X_{i1}} + {c_2}{X_{i2}} + {d_2}{X_{i3}} + {e_2}{X_{i4}} + kY} \right) (5.8) λi=λ0expb3Xi1+c3Xi2+d3Xi3+e3Xi4+f3Xi5+kY {\lambda _i} = {\lambda _0}exp\left( {{b_3}{X_{i1}} + {c_3}{X_{i2}} + {d_3}{X_{i3}} + {e_3}{X_{i4}} + {f_3}{X_{i5}} + kY} \right) (5.9) λi=λ0expb4Xi1+c4Xi2+d4Xi3+e4Xi4+f4Xi5+g4Xi6+h4Xi7+kY {\lambda _i} = {\lambda _0}exp\left( {{b_4}{X_{i1}} + {c_4}{X_{i2}} + {d_4}{X_{i3}} + {e_4}{X_{i4}} + {f_4}{X_{i5}} + {g_4}{X_{i6}} + {h_4}{X_{i7}} + kY} \right) λ_i is the individual hazard rate, and λ₀ the baseline. The covariates alter the baseline hazard multiplicatively and are the same as previously. Crucially, the reported part of Xi7 is now an indicator of a match not being in a CTP zone, but the worker has potentially benefitted from the CTP policy and currently holds a permanent job.

Table 3 reports the hazard rate estimates using a proportional hazard Cox model. Except for permanent workers newly hired in firms with fewer than 1,000 workers and who are less likely to be fired (only in the second specification, column 3, and only significant at the 10%-level, which makes sense as they are now more expensive to fire), no estimates in the first three specifications are significant. Importantly, there is no significant effect on the stock of workers with any contract. This makes sense in the context of the model as the firing tax prevents firms from firing permanent workers as a response to higher separation taxes. Temporary workers to whom the firing taxes do not apply will be fired at the same rate as before (particularly since temporary workers are hired for fixed terms). Since workers lose their jobs at the same occasions, we should not expect any change in the hazard rates.

The fourth specification has some significant estimates. Column 7 shows that eligible workers with a permanent job in a CTP region now have a lower hazard rate (down to 0.2 of the baseline), significant at 5%, and eligible workers who obtained a permanent job have a hazard rate increased by the factor 3 although this is only significant at 10%. This may reflect the structure of the differences between CTP and CRP. Firms may hire permanent workers under the CRP with no intention of letting them stay on after two years. Under the CTP, this is not possible. This does not overturn the conclusion of separation taxes being inconsequential for hazard rates of permanent workers, as we can see in columns 1 to 6 and column 8. Additionally, because it is only recent hires and not workers subject to higher separation taxes in general, this is a composition effect. Finally, hazard rates of the stock of workers is unaffected. In summary, hazard rates are stable as result 4 suggests

In the appendix, duration is explored. We find that workers were hired later, meaning their hazard rates did not change, but utilization decreased.

For firms not hiring new workers (permanent or temporary), a reduction in economic activity, beyond hiring fewer workers, may be reflected by fewer hours worked by those already hired. For hours worked and income, respectively, this was the set of regressions: (5.10) loghours=a4+b4X1+c4X2+d4X3+e4X4+f4X5+g4X6+h4X7+kY log\left( {hours} \right) = {a_4} + {b_4}{X_1} + {c_4}{X_2} + {d_4}{X_3} + {e_4}{X_4} + {f_4}{X_5} + {g_4}{X_6} + {h_4}{X_7} + kY (5.11) logincome=a4+b4X1+c4X2+d4X3+e4X4+f4X5+g4X6+h4X7+kY log\left( {income} \right) = {a_4} + {b_4}{X_1} + {c_4}{X_2} + {d_4}{X_3} + {e_4}{X_4} + {f_4}{X_5} + {g_4}{X_6} + {h_4}{X_7} + kY

Table 4 reports the results on hours worked and total income. The covariates are the same as previously for the fourth specification. However, now several indicators from X7 are reported: if a worker previously held a permanent job in a CTP area and holds any type of job in a CTP area currently, if a worker employed either in a CTP or control area formerly held a permanent job in a CTP area and held a current job in a CTP area, if a worker previously held a permanent job while currently holding a permanent job in a CTP area, if a worker holds a permanent job in a CTP area currently and previously held a permanent job in a CTP area and, finally, if a worker previously held a permanent job in a CTP area and is currently holding a permanent job but not necessarily in a CTP area. There seems to be a redistribution of hours worked from newly hired temporary workers towards permanent workers in new firms only, which reflects fewer overall hires, but permanent workers are used more. In column 1, workers benefitting from the policy directly work more hours in CTP but less (approximately 50%) in other regions. Eligible permanent workers in a CTP region worked less.

For income (column 3), we see a redistribution away from newly hired temporary workers towards newly hired permanent workers. Although the results on individually affected workers are significant at 10%, we see redistribution. This seems to counter the general equilibrium effect as workers that are eligible and have a permanent job seem to be worse off in CTP regions. At the same time, eligible workers seem to be better off in and outside CTP regions when they obtain permanent jobs. A worker outside a CTP region makes approximately 50% more if they are newly hired35

Such workers are hired late in the year and cannot earn as much, inflating the estimates. This is reflected in insignificant results for the stock as workers have the same time to earn.

Revenue, reported in table A1 in appendix A, was summed for each firm in the treatment and control groups and averaged. Values were normalized by the 2005 value. There is a reduction of approximately 2.8% in the treatment group in 2006 versus 2005 and even in 2007 the treatment group recovered to reach the output level of 2005. The control group's output, however, was higher by 0.9% in 2006, and in 2007 it was 7%. This is evidence of diminished economic activity in treatment areas.