2. Institutional background

The Chilean pension model was implemented in 1981. It has a three-pillar design: a social pillar funded by public funds (Pillar 0), a mandatory individual contributory pillar associated with personal savings accounts managed by private firms (Pillar 2), and a voluntary savings plan introduced after a major reform in 2002 (Pillar 3). Contributors are dependent workers who must save 10% of their paychecks (before taxes) in an individual retirement account (Pillar 2). These contributions are automatically deducted and saved into individual savings accounts within the retirement system. The 10% contribution rate applies up to a monthly earnings cap, above which no additional contributions are made. As of 2024, the cap is set at monthly earnings of US$3,394, equivalent to an annual wage of nearly US$41,000 (Dirección del Trabajo, 2024). As of 2020, independent workers must also contribute to their savings account at a contribution rate that started at 0.75% in 2020 and will increase to 7% in 2028 (Superintendencia de Pensiones, Reference de Pensiones2022e).

Pensions within the social pillar may take two forms. The first is a basic monthly social pension of US$215.25 for individuals who are not enrolled in the formal pension system and have no other source of retirement income. Alternatively, there is a means-tested pension supplement for individuals who are enrolled in the system but whose pension falls below a designated threshold.Footnote ⁴ In November 2021, over 1.8 million individuals received funds from Pillar 0 (representing over 90% of retirees enrolled with the system). Sixty-one percent of beneficiaries were female. However, mandatory contributions from Pillar 2 remain the primary source of retirement income among retirees (Berstein et al., Reference Berstein, Fuentes and Villatoro2013).

Private firms known as pension fund administrators have only one objective: to manage contributors’ investments and savings. For their services, individuals pay management fees defined as a portion of their salary. In April 2022, the average market fee was 1.15% (Superintendencia de Pensiones, Reference de Pensiones2022c). Individuals can allocate their savings across one or two funds selected from five different options. These funds are labeled A, B, C, D, and E and differ in their financial risk.Footnote ⁵

Employers have a reduced role in the current system. Employers can only contribute to workers’ retirement savings using a specific collective voluntary account (within Pillar 3) where the employer opens an account with an authorized institution, and both the employer and the employee agree to contribute (Superintendencia de Pensiones, Reference de Pensiones2022a). However, this instrument has yet to be used in larger shares. By November 2021, only 870 collective individual accounts were registered for the system’s over 11 million enrollees (Superintendencia de Pensiones, Reference de Pensiones2022b), of which almost 6 million were active contributors (Superintendencia de Pensiones, Reference de Pensiones2021b).

The latest reforms in Chile (2008 and 2021) have only considered extensions of Pillar 0. Table 1 compares the social components of Pillar 0 across the top-10 pension systems. In this comparison, Chile ranks relatively well regarding the generosity of its social pensions relative to per capita GDP and minimum wage. However, the contribution rates for Pillars 1 and 2 are substantially lower in Chile than in the top-10 systems, suggesting a future reform should consider raising contributory rates. Table 1 reports the contribution rates in the top pension systems and the Chilean model. The comparison across systems is not straightforward, as countries have different institutional backgrounds and system designs. There is no unique optimal model; instead, there are different models for different settings. For example, in the Netherlands, the state contributes with funding from general funds (Pillar 1) (Parada-Contzen and Provoste, Reference Parada-Contzen and Provoste2021). This contribution is despite the ample contribution rate for employers and workers. It is the only country in the Table that has this design. In the same way, some countries have generous social pensions (Pillar 0) despite having relatively low contribution rates (e.g., New Zealand, Australia, and Denmark) by providing social pensions funded through general taxes (Parada-Contzen and Provoste, Reference Parada-Contzen and Provoste2021). In this setting, Chile ranks very well in terms of social pension as a fraction of per capita GDP or minimum wages, but it has the lowest contribution rate of all countries.

Table 1.

Comparison between top-10 pension systems and Chile of Pillar 0 and contribution rates for Pillars 1 and 2

Note: (a) Own elaboration based on Parada-Contzen and Provoste (Reference Parada-Contzen and Provoste2021) and references therein. (b) Top-10 pension system based on Mercer (2020). (c) Wage data come from Eurofound (2021), Ministry of Manpower (2021), and WageIndicator (2021). (d) There is no legal universal minimum wage for Denmark, Finland, Norway, Singapore, and Sweden as wages depend on collective contracts by sectors. Thus, minimum wage was computed using the average agreed salary for the three jobs with the lowest wages. For Canada, the minimum wage in Ontario was used. (e) For comparison purposes, the hourly wage was transformed, assuming 45 hours per week and 4 weeks per month for all countries. (f) W = Worker. E = Employer. A hyphen (-) indicates that there is no contribution for that agent/pillar (meaning 0 contribution).

The remainder of this section reviews existing evidence on labor demand elasticities, which are critical for predicting the employment impacts of pension reforms that increase labor costs. One of the canonical works in estimating labor demand elasticities is reported in Hamermesh (Reference Hamermesh1993), which reports an average elasticity of −0.45 with typical ranges between −0.75 and −0.15. There is large heterogeneity in estimates across countries and industries. However, a consensus is that a 10% increase in labor costs leads to a 3% decrease in employees, implying a point estimate of −0.30 (Hamermesh, Reference Hamermesh, Heckman and Pagés2004, Reference Hamermesh2021). The scarce evidence for Chile suggests an employment drop of 0.3% for a 1% wage increase. Early evidence from Chile (1981–1986) estimates a labor demand elasticity from −0.48 to −0.32 (Hamermesh, Reference Hamermesh, Heckman and Pagés2004), while a study covering the 2000s finds an elasticity of −0.19 for the manufacturing sector (Lichter et al., Reference Lichter, Peichl and Siegloch2015).

In the literature on labor demand elasticities, variation across industries, worker type, regions, or other characteristics is typically introduced by estimating separate regression models on subsets of data (Slaughter, Reference Slaughter2001; Maiti and Indra, Reference Maiti and Indra2016). In a comprehensive meta-analysis, Lichter et al. (Reference Lichter, Peichl and Siegloch2015) explore sources of heterogeneity in wage elasticities of labor demand with information from 151 studies containing 1,334 estimates from 37 countries between 1980 and 2012. They find that estimated wage elasticities in the literature show that labor demand is inelastic.

The average elasticity across all 1,334 estimates is −0.55, while estimates for Latin America show a smaller elasticity, averaging −0.37. For the United States, Maiti and Indra (Reference Maiti and Indra2016) report labor demand elasticities somewhat below those reported for the entire sample in Lichter et al. (Reference Lichter, Peichl and Siegloch2015). Across sectors, the labor demand elasticity in construction is larger than that for manufacturing, services, and other sectors (Lichter et al., Reference Lichter, Peichl and Siegloch2015). While estimating labor demand elasticities has attracted significant attention in the literature, there is only so much evidence for Chile specifically, except for two studies in the manufacturing sector reviewed by Lichter et al. (Reference Lichter, Peichl and Siegloch2015).

From Lichter et al. (Reference Lichter, Peichl and Siegloch2015), estimation procedures that account for potential endogeneity between wages and labor demand by instrumenting for wages show higher elasticities (in absolute value) than when a single equation model is estimated. In Latin America, labor demand elasticities shift by 7 to −0.44 pp when using instrumental variables (IV) methods. When considering heterogeneity across workers, evidence shows that labor demand for blue-collar and low-skilled workers is more elastic than for white-collar and high-skilled workers. This is consistent with the hypothesis that more specialized workers are harder to substitute for. A summary of the estimates from Lichter et al. (Reference Lichter, Peichl and Siegloch2015) is presented in Table 2.

Table 2.

Summary of prior evidence on labor demand elasticities

Note: (a) Own elaboration based on Lichter et al. (Reference Lichter, Peichl and Siegloch2015) complemented with Hamermesh (Reference Hamermesh, Heckman and Pagés2004) for estimates in Chile (b) N = number of estimates.

In a meta-analysis for Germany based on 705 estimates, Popp (Reference Popp2023) find results consistent with the meta-analysis of Lichter et al. (Reference Lichter, Peichl and Siegloch2015). The average labor elasticity is −0.43, with important heterogeneity across industries, regions, and workforce characteristics. Elasticities increase in about 18% (7 pp) when using IV approaches (Popp, Reference Popp2023). Regarding workforce skill, Popp (Reference Popp2023) find average elasticities of −0.44 for high-skilled workers, −0.33 to −0.21 for medium-skilled workers, and −0.72 for low-skilled workers.

3. Methods and data

3.1. Empirical model

To find the impact on employment of implementing an employer’s contributory pillar, we estimate the unconditional labor demand elasticity by using the following reduced-form log-linear specification.Footnote ⁶ From the specification, we can directly interpret the coefficients as elasticities.

(1)\begin{equation}\log {y_{it}} = \beta \log {w_{it}} + \gamma {X_{it}} + {\phi _i} + {\theta _t} + {\varepsilon _{it}}\end{equation}

where ${y_{it}}$ represents the number of workers employed at firm $i$ in period $t$, ${w_{it}}$ represents the pre-tax average salary per worker at the firm level, ${X_{it}}$represents other characteristics of the firm, ${\phi _i}$ represents firm fixed effects, ${\theta _t}$ represents time fixed effects, and ${\varepsilon _{it}}$ represents an idiosyncratic shock. The time fixed effects control for shocks to the number of workers in a particular year common to all firms. Firm characteristics and fixed effects are included to control for firm-level determinants of worker pay, such as amenities, market power, productivity, and its sharing (Card et al., Reference Card, Cardoso, Heining and Kline2018; Sorkin, Reference Sorkin2018; Berger et al., Reference Berger, Herkenhoff and Mongey2022; Lamadon et al., Reference Lamadon, Mogstad and Setzler2022). Generally, firm fixed effects capture time-invariant variation across firms that affects the number of workers per firm.

The log-log specification is chosen to consider changes in percentage instead of levels; thus, the firm’s size does not distort the analysis. Our parameter of interest, $\beta $, represents the average elasticity of employment with respect to wages. The specification follows the canonical labor demand model in the literature (Castillo-Freeman and Freeman, Reference Castillo-Freeman, Freeman, Borjas and Freeman1992; Hamermesh, Reference Hamermesh1993; Angrist et al., Reference Angrist, Graddy and Imbens2000; Hull et al., Reference Hull, Kolesar and Walters2022). Specifically, the key coefficient $\beta $ represents the percent change in employment for a 1% change in salary per worker (i.e., the labor demand elasticity).

To explore heterogeneity in elasticities across industries and workforce characteristics, we estimate Equation 1 separately by subsets of industry and worker type, defined by industry as follows: (1) agriculture, (2) mining, (3) manufacturing, (4) utilities, (5) construction, (6) commerce, (7) transportation, (8) tourism, and (9) financial services; and by workforce category as follows: (1) managers, professionals and technicians, (2) clerks and office workers, (3) skilled production workers, and (4) unskilled production workers. This subsample analysis allows us to better predict the heterogeneous impacts of the proposed pension reform on employment, depending on specific characteristics of the Chilean labor market.

We estimate a reduced-form specification of labor demand at the firm level, so our results reflect partial equilibrium employment effects. Our estimates thus reflect employment losses for currently employed workers, holding labor force participation and hours worked fixed. This assumption is reasonable in the Chilean context, where labor laws limit firms’ ability to adjust hours downward. There are also limitations on firms’ reducing wages or replacing workers with lower-paid hires in the same roles. However, our estimates may overstate employment losses if the pension reform causes voluntary labor force exits.

Because wages might be endogenous to employment at firms, we also estimate an IV model where we instrument for wages using exogenous variation in labor market conditions at the region-industry level (see Equation 2). Specifically, we construct four instruments: (1) average wage per worker in other regions, (2) average wage per worker in other industries, (3) regional minimum wages, and (4) regional unemployment rates. These instruments isolate variation in firm-level wages driven by market-level labor supply and demand shocks, which are plausibly uncorrelated with firm-specific labor demand shocks. This approach follows the widely used Bartik (Reference Bartik1991) IV strategy, leveraging the fact that labor productivity and wages vary across region-industry cells based on their underlying labor market characteristics (Enrico, Reference Enrico, Card and Ashenfelter2011; Maiti and Indra, Reference Maiti and Indra2016).

The key identifying assumption for the shift-share instrument is that firm-level wage changes apply to the entire firm across locations and are driven by shocks at the region-industry level that are uncorrelated with local labor market conditions. This means that the instruments require that demand for labor in region A is independent of the demand for the same labor in region B in such a way that shocks in one region do not affect all markets. For example, a currency shock would violate this assumption because it would affect all regions. A similar analysis can be extended for industries or sectors. The assumption behind the labor market structure variables is that an individual firm does not influence the aggregated labor market characteristics (i.e., an individual firm’s employment decisions do not significantly influence aggregate labor market characteristics, ensuring exogeneity of the instruments). This version of the instrumenting approach was introduced by Hausman (Reference Hausman, Bresnahan and Gordon1996) and applied by Nevo (Reference Nevo2001), where the prices of products in other markets served as an instrument. We also follow the recent evidence by Hazell et al. (Reference Hazell, Patterson, Sarsons and Taska2022) for the construction of these instruments. In this version of the model, we use instrumented wages as follows:

(2)\begin{equation}\log {y_{it}} = \beta \log {\hat w_{it}} + \gamma {X_{it}} + {\phi _i} + {\theta _t} + {\varepsilon _{it}}\end{equation}

where ${\text{log}}{\hat w_{it}}$ represents the predicted wage (in logs) for firm $i$ at time $t$, as obtained from the first-stage regression. To capture employment inertia at the firm level, we also estimate an Arellano and Bond (Reference Arellano and Bond1991) dynamic panel specification with lagged variables as instruments.

(3)\begin{equation}\log {y_{it}} = \alpha \log {y_{i,t - 1}} + \beta \log \left( {{w_{it}}} \right) + \gamma {X_{it}} + {\theta _t} + {\varepsilon _{it}}\,\end{equation}

Our empirical model allows us to estimate the marginal effect of increasing labor costs on employment levels. In our baseline simulation, we assume employers bear the full incidence of a 1% increase in labor costs, holding net wages fixed. This implies labor supply is perfectly elastic, as workers’ net wages are held constant. If the labor supply is upward sloping, the employment losses would be smaller, as workers would bear some tax burden through lower net wages.

The employment effects we compute are job losses for employed workers due to increased labor costs. Results thus reflect extensive margin responses rather than hours change, which we cannot measure. This focus aligns with Chilean labor laws, which limit downward wage and hours adjustments and prohibit replacing workers with lower wage hires in the same roles. Because we estimate a demand-side model, we do not capture potential shifts in labor supply.

3.2. Data

We construct a panel of firms using five waves of the ELE for 2007, 2009, 2013, 2015, and 2017. The ELE belongs to the Chilean Ministry for the Economy, Development, and Tourism and is managed by the National Institute of Statistics (INE).Footnote ⁷ The ELE covers legally registered firms in all regions of Chile and is representative at the industry and firm size (based on sales) level. This allows us to characterize the universe of Chilean firms (Santi and Santoleri, Reference Santi and Santoleri2017). The ELE features a rotating sample design to prevent sample size reductions and maintain representativeness.

Our estimation sample comprises 26,556 firms pooled over the five waves. The panel is unbalanced, with 3,257 firms appearing in all waves, which requires redefinition of estimation samples for alternative specifications that required lagged values in one or more periods. The ELE oversamples large firms and undersamples microfirms relative to the population (Santi and Santoleri, Reference Santi and Santoleri2017).Footnote ⁸ We observe the same pattern in our research sample, so our main results should be interpreted as most relevant for formal employment in larger firms. Microfirms may react differently to the pension reform due to greater labor cost sensitivity and informality.

Variable definitions and summary statistics for our estimation sample are presented in Table 3. The key variables are firm-level employment and average monthly wage per worker. Summary statistics of these two variables by industry are reported in Table A1 of the Appendix.

Table 3.

Summary statistics for our estimation sample

Note: Sample size = 26,556.

5. Policy implications

The main policy implication of our research is that a uniform contributory pillar would increase unemployment, especially for blue-collar workers and industries that adjust more easily to increases in labor costs. Our findings suggest this policy would be regressive, disproportionately harming low-wage and vulnerable workers. These results provide significant evidence for considering alternative reform proposals and arrangements.

One alternative is a policy where employer contributions apply only to high-wage workers, while including a redistributive component to reduce inequality among retirees. High-wage workers accumulate more lifetime savings, meaning their pension income is higher. Implementing an employer’s contributory pillar to a segment of workers prevents layoffs of unskilled workers who accumulate less pension wealth in individual-funded accounts because of lower wages and higher risk of unemployment and discontinuous labor trajectories.

The current proposal taxes risky groups to benefit workers with better financial positions in retirement. Based on our findings, we propose an alternative arrangement, with contributions levied only on high-wage workers and in industries where labor demand is more inelastic. We incorporate a redistributional component, allowing individuals with less accumulated lifetime savings to receive higher pension income from contributions made by high-wage workers. This design is consistent with recent evidence that shows that in a design where an employer contributory pillar of 6%, with all the contribution going to an individual account without redistributional arrangements, the median of the replacement rate reaches 17%; while with redistributional components (3% to individual account and 3% with redistributional arrangements) the median of the replacement rate increases up to 29% (Superintendencia de Pensiones, Reference de Pensiones2024).

The current pension system caps contributions at almost US$41,000 per year. The cap measure prevents contributions from being too high for workers at the wealthiest 1% of the distribution and allows firm managers to offer competitive wages for high-level company directors. While the cap limits total contributions under a redistributive scheme, keeping it mitigates opposition from high-wage workers and firms. In terms of labor demand elasticity, with the cap as it is today, elasticity for high-earning workers would be like that of other groups. Indeed, we find no statistical difference in estimated elasticities for wages above 90% of the wage distribution compared to other workers.

Economic sectors also vary in size, meaning that when labor costs increase, sectors with more elastic labor demand that are larger employers would experience higher job losses. For example, the mining sector has smaller labor demand elasticity than other sectors, but only accounts for 3.1% of employment, according to data from the National Statistics Institute (Banco Central de Chile, 2024). However, some sectors with larger elasticities employ larger shares of workers. For example, commerce represents 19.1% of employment, transportation represents 8.8%, and construction represents 7.9%.

A second alternative is to introduce a differentiated employer contributory pillar, with contribution rates varying across sectors and worker types, rather than a uniform contribution rate. With this design, policymakers could mitigate adverse employment effects for unskilled workers and those in sectors more sensitive to labor cost increases.

Our results are from a partial equilibrium model, so the employment effects and estimated job losses represent risks for workers currently employed due to increased labor costs for the firm. Our analysis estimates labor demand elasticities by exploiting wage variation across firms. These elasticities capture a partial-equilibrium response of labor demand to wage changes. We then use these elasticities to predict the impact of a pension reform that raises labor costs economy wide. Our results focus on the extensive margin, consistent with Chilean labor regulations. Chile’s labor legislation protects workers. If labor costs increase, firms can only adjust by laying workers off. The law limits firms’ ability to cut wages or hours, or to replace workers with lower-paid hires in the same roles.

By aggregating microeconomic responses, we get an estimate of how an economy-wide wage shock could affect aggregate unemployment. This exercise translates our partial equilibrium elasticities into a macroeconomic counterfactual. Extensive literature uses partial equilibrium elasticities to inform macroeconomic questions (Parker et al., Reference Parker, Souleles, Johnson and McClelland2013; Mian and Sufi, Reference Mian and Sufi2014; Giroud and Mueller, Reference Giroud and Mueller2017; Zwick and Mahon, Reference Zwick and Mahon2017). However, we recognize elasticities estimated from firm-level variation may differ from the economy-wide elasticity because of general equilibrium adjustments. Intuitively, when all firms experience the same shock, there can be feedback effects through prices, wages, and labor supply that either amplify or attenuate the initial response.

Several general equilibrium channels apply to our analysis. First, an economy-wide increase in labor costs could raise consumer prices as firms pass through higher costs. This product market adjustment would dampen the decline in labor demand. Second, the pension reform may alter workers’ labor supply incentives. If the reform reduces labor force participation, it would mitigate adverse employment effects. Finally, equilibrium wages may adjust. If labor supply is elastic, workers may accept lower wages, partially offsetting the reform’s direct effect on labor costs. These channels suggest the aggregate elasticity may be smaller than our microeconomic estimates.

Under what conditions are our estimates informative about the aggregate employment effects of the pension reform? Theory highlights three key considerations. First, pass-through from labor costs to product prices matters. With full pass-through, firms shift labor cost shocks to consumers, leaving labor demand unchanged. Second, the elasticity of labor supply is crucial. Highly elastic labor supply implies larger wage adjustments that offset the direct effect of the reform on labor costs. Finally, wage rigidity can play a role. If wages are slow to adjust, the microeconomic elasticities will closely approximate the macroeconomic response in the short run.

Ultimately, our counterfactual predictions depend on the empirical magnitudes of these general equilibrium forces. If pass-through is limited, labor supply inelastic, and wages rigid – all plausible in the Chilean context in the short run – then microeconomic estimates can offer a reasonable guide to the aggregate labor response. If these assumptions are violated, microeconomic estimates may overstate the aggregate employment effects. Our estimates would still indirectly inform the macroeconomic question. For example, Nakamura and Steinsson (Reference Nakamura and Steinsson2018) argue microeconomic elasticities can distinguish between important classes of structural models (that are beyond the scope of the present paper). An alternative interpretation is that our estimates offer a useful bound on the potential macroeconomic impacts, with the true effect lying somewhere between the partial and general equilibrium responses.

Because this is a demand-side model, we do not estimate the policy’s impact on labor supply. Increasing contribution rates to workers may cause the exit of workers, as net wages will decrease, for example. Workers may also remain invariant, depending on their intertemporal elasticity of labor substitution, since they will receive additional benefits in the future. While estimating labor supply elasticities is beyond our scope, evidence from the literature suggests the effects of pension reforms on aggregate labor supply are small for plausible values of the intertemporal elasticity of substitution for labor (Imrohoroglu and Kitao, Reference Imrohoroglu and Kitao2009). The literature also suggests changes in labor supply responding to adjustment due to social security reforms should be prominent at the intensive margin (hours) (Cottle Hunt and Caliendo, Reference Cottle Hunt and Caliendo2022). Because flexibility regarding the number of work hours is rigid in Chile, labor supply should remain relatively constant as a new pillar enters the pension system. Labor supply responses are also heterogeneous depending on economic, institutional, and cultural differences across countries (Prescott, Reference Prescott2004). Thus, it could be interesting to investigate specific labor supply changes for Chile in future research.

6. Conclusion

This paper estimates labor demand elasticities in Chile using five waves of a representative firm survey. Elasticities are estimated by industry (nine categories) and worker classifications (four categories: managers, clerks, skilled production workers, and unskilled production workers). The estimates are then used to simulate the employment effects of implementing an employer contributory pillar in Chile’s pension system. Our simulated reform raises gross labor costs while holding net wages constant. From a labor demand perspective, implementing an employer contributory pillar can be modeled as an exogenous shock to labor costs. Our counterfactual predictions assume the labor force remains fixed, and job losses among the currently employed. While future work could relax the fixed labor force assumption, we cannot model job losses on the intensive margin, since we do not observe hours in our data.

We find labor demand elasticities are statistically negative and inelastic in almost all specifications. Our estimates are consistent with the range of elasticities in the literature. Estimated elasticities range between −0.31 and −0.91. Our preferred specification yields a decrease of 0.31% in employment for a 1% wage increase, with a labor demand elasticity 90% confidence interval of [−0.36; −0.26]. This result is consistent with previous evidence from Chile and the consensus range for labor demand elasticities worldwide. We observe substantial differences across sectors and worker types when estimating elasticities by industry and worker category. Labor demand elasticity is higher for unskilled workers and in industries where labor is easily substituted. These results are consistent with previous evidence. On average, we find an elasticity of −0.35 for white-collar workers and −0.51 for blue-collar workers.

We predict increases in unemployment rates between 0.20 and 0.59 pp from a baseline unemployment rate of 6.51% as wages increase by 1%. In our benchmark model, we predict an unemployment rate of 6.69%-6.73%. Regarding layoffs, we predict an average job loss of 17,848 positions in our most conservative model. Job losses are higher for unskilled workers. The baseline unemployment rate used in the calculations is relatively low regarding unemployment rates after 2017. There was a critical increase in unemployment rates during COVID-19. However, the recovery has been slow. By 2024, unemployment rates are between 8.3% and 8.7%, versus 6.51% in our baseline period.

Our model predicts employment effects, holding the total labor force constant in response to gross wage increases. While this assumption could be relaxed by considering estimates of employment elasticity to pension reforms, our results provide a baseline prediction of the impact of such a reform. Most of the employment effect predictions were performed considering our most conservative estimation of elasticity, meaning that we should consider the predicted effects as a baseline level. In our baseline estimations, we also control for potential endogeneity between wages and employment at the firm. While our instruments pass the standard tests for validity and the general inelastic result holds for all specifications, the estimates are sensitive to the instrumentation of wages. For example, when aggregating all sectors and workers, labor demand elasticity doubles after instrumenting wages. Future work could explore other methods and better ways to account for simultaneous processes at the firm level. Extensions should also consider both the demand and supply effects of increasing labor costs and implementing an employer contributory pillar.