Labor market institutions and firms’ location choices

Abstract

The paper evaluates the empirical effects of labor market institutions (LMI) on foreign direct investment (FDI) decisions using an individual dataset describing French firms’ expansion strategies in OECD countries over 1992–2002. First, we provide evidence that labor market institutions do matter in location decisions. Precisely, we show that labor market rigidity significantly reduces the country’s attractiveness for foreign investors. Yet, the effect is of limited magnitude compared to FDI determinants related to the country’s market potential or supply access. Second, we go deeper in the precise role of various LMI dimensions. In line with the literature, we find that stringent employment protection laws have a dampening effect on the location probability. Besides, we show that this is not the only dimension that matters. In particular, we find that the generosity of the unemployment benefit system plays a significant negative role on the country’s attractiveness, even once the role of employment protection is controlled for.

Appendices

Appendix 1: Data Appendix: Definitions and sources

We study FDI choices of French firms over the period 1992–2002 in the set of 18 OECD countries. The list of countries is notably reported in Table 9.

Table 9 OECD Labor market institutions: country and year coverage

1.1 Standard explanatory variables

Sources for the standard explanatory variables of FDI choices included in our sample are described hereafter.

The real market potential variable is constructed as in Redding and Venables (2004), relying on importation data taken from the “Dots” database of the IMF and gravity variables taken from the CEPII “Distance” database. Our measure is based on the following definition of country i’s real market potential:

$$ RMP_i=\sum_j I_j P_j^{\sigma-1}\phi_{ij} $$

(3)

where I _j is the nominal income in country j, P _j the price level and ϕ _ij a measure of the various barriers limiting trade between countries i and j. Redding and Venables (2004) show how to build the variable using as proxy for the country-specific determinants of real market potential (the I _j P ^σ−1 _j terms in the definition above) estimates of importer-specific fixed effects obtained from a gravity equation. The extent of trade barriers is then measured using various proxies.

The gravity equation that is first estimated has the following form:

$$ \ln X_{ij}=\theta+\mu_i cty_i +\mu_j cty_j+\delta \Upphi_{ij}+\varepsilon_{ij} $$

(4)

where X _ij is the value of the trade flow between country i and country j, cty _i and cty _j are exporter- and importer-specific fixed effects, and $\Upphi_{ij}$ is a vector containing various measures of bilateral trade barriers (the distance between both countries and a set of binary variables, that indicate the existence of a common border, of past colonial links, the use of a common language and the countries’ involvement into trade agreements and monetary unions).

From the estimation of the gravity equation, one can restore a measure of real market potentials (expressed in current US dollars) as in:

$$ \hat{RMP}_i=\sum_j \left[\exp(cty_j)\right]^ {\hat{\lambda}_j}\left[\exp(\Upphi_{ij})\right]^{\hat{\delta}} $$

(5)

The variable is built annually between 1992 and 2002.

The GDP per capita is obtained by dividing current GDP series (converted at nominal exchange rate in US Dollars) by the population level of the country, based on the “World Developments Indicators”, World Bank. The variable is taken in logarithm.

As an alternative to GDP per capita, we use the unit labor cost, defined as the ratio of total labor costs to real output, or equivalently, as the ratio of average labor costs per hour to labor productivity. The series is expressed in percentage. Data are taken from the OECD’s Main Economic Indicators. Series are available on a yearly basis for all countries of our sample except for Switzerland.

Distance from France (“ln distance”) comes from the CEPII’s “Distance” database.

The supply access variable is built as in Mayer et al. (2010) using data from the French Input/Output (I/O) Tables and the Enquête Annuelle d’Entreprises for employment data. The rationale behind the construction is the following. The incentive for a firm in sector s to locate in country i increases in (i) country i’s supply of intermediate goods, relative to the rest of the world, and (ii) sector s’s use of intermediate inputs. To capture the first element, we use information on input producers in country i which are affiliates of French firms. Namely, the share ω ^m _i of inputs m produced in country i is measured by the share of the overall employment by French affiliates in industry m that is located in country i. The use of intermediate inputs in the sector firm k belongs to is approximated using information from the French I/O tables. This implicitly assumes that foreign affiliates of French firms have the same technological function as French firms in the same sector. The total share of intermediate goods in the production of the affiliate is thus approximated by the share recorded in the French I/O tables for sector s (called β _s hereafter). The same holds true for the technical coefficients a ^m _s measuring the quantity of industry m’s inputs needed to produce one unit of output in industry s.

Based on these data, it is possible to measure the availability of inputs within country i that are used by an affiliate operating in industry s as:

$$ SA_{i}^{s}=\frac{\beta^{s}}{d_{ii}}\left[\sum_{m=1}^{S} a_{s}^{m} \omega_{i}^{m}\right] $$

(6)

The supply access variable thus measures the average share of world intermediate goods produced in country i. In the average, each industry is weighted by the technical coefficient measuring the reliance of sector s to this particular input (a ^m _s ): Affiliates benefit more of the proximity to local suppliers producing intermediate goods they use intensively. The supply access variable is also higher if intermediate inputs are a large component of costs in industry s (as measured by β^s). Finally, the measure is divided by the internal distance of country i, d _ii , in order to account for the ease of access to suppliers inside i. Using I/O tables for each year of the sample, we obtain time-series of sector-specific supply access. In the estimates, the explanatory variable is supply access in the year preceding the investment, in order to limit endogeneity and avoid double-counting the firm’s own investment.

The governance indicator is built using the governance indicators defined and measured by Kaufman et al. (2005). Data are available on the World Bank web site. ²⁰ The indicators measure six dimensions of governance: (1) Voice and Accountability measures political, civil and human rights; (2) Political Instability and Violence measures the likelihood of violent threats to, or changes in, government, including terrorism; (3) Government Effectiveness measures the competence of the bureaucracy and the quality of public service delivery; (4) Regulatory Burden measures the incidence of market-unfriendly policies; (5) Rule of Law measures the quality of contract enforcement, the police and the courts, as well as the likelihood of crime and violence; (6) Control of Corruption measures the exercise of public power for private gain, including both petty and grand corruption as well as state capture.

Data are available for the years 1996, 1998, 2000 and 2002. All countries in the sample are covered. For the years 1992–1995, we use the same value than in 1996. For the year 1997, 1999 and 2001, we take the average of the two yearly adjacent values. All variables are transformed so that they take values between 0 and 100, increasing with the quality of governance. The average indicator is built as a simple arithmetic mean of the 6 dimensions of governance. The larger the variable, the better the quality of governance.

The corporate tax rate series are taken from the OECD Tax database. Precisely, we use the “combined corporate income tax rate”. Series are built as a percentage share, taking values within a [0;100] interval. The corporate tax rate variable is thus introduced in level in the regression. It is denoted “Corporate tax rate (%)”. Series are available on a yearly basis over the period 1992–2002 for all countries in the choiceset.

Aggregate FDI inflows series are taken from the OECD’s International direct investment database. They correspond to the gross FDI inflows for the total economy. Series are available on a yearly basis over the period 1992–2002 for all countries in the choiceset, except for Switzerland (in 1993), Germany (in 1992) and New Zealand (in 2001).

1.2 Labor market institutions

Economic Freedom database: The database is provided by the Fraser Institute, available online, http://www.freetheworld.com. We use the 2005 edition of the Economic Freedom of the World annual report. LMI indicators are defined as follows:

The Synthetic LMI Index sums up various sub-indices that are related to different dimensions of the labor market functioning: (1) the “minimum wage impact”, (2) the “unemployment benefits” variable, (3) the “Hiring and firing practices” index, (4) the “Centralization” index, and (5) an indicator of the use of conscripts to obtain military personnel.

The Minimum Wage Impact variable is based on two survey responses obtained from the Global Competitiveness Report, asking about (1) the overall “impact of the minimum wage”, and (2) the strength of enforcement of the minimum wage law. Countries receive lower ratings if the survey respondents indicate the minimum wage has a large impact and/or is strongly enforced.

The Unemployment Benefits variable indicates whether the unemployment benefits system preserves the incentive to work, with low values associated to pernicious effects.

The Hiring and Firing Practices variable indicates whether hiring and firing practices of companies are determined by private contract, with low values meaning that firing and hiring laws are more constraining.

The Centralization Index measures the share of labor force whose wages are set by centralized collective bargaining.

Original data take values over the range [0,10] but have been rescaled over [0,100] before introducing this variable in level in the conditional logit. This allows interpreting coefficients as the probability change attributable to a one percentage point increase in the indicator. Besides, original EF variables are increasing with the degree of labor market flexibility. To homogenize the interpretation of coefficients with OECD LMI variables, we rebuilt the variables from Economic Freedom for them to be increasing with the degree of labor market rigidity. Precisely, we take 100 minus the original value. This preserves the cross-country distribution of the variables, while making them take values over the range [0;100] as OECD LMI variables.

We use raw data that are given for the years 1990, 1995, 2000, 2001 and 2002, and we rely on interpolation for missing years. The only missing values in this dataset are for the Unemployment Benefit Index, that is missing for Australia in year 1992–1993 and 1994.

OECD sources: We collect data on various LMI for OECD countries over the period 1992–2002, using data provided by the OECD. We focus on the following set of LMI:

Employment Protection Laws: We consider the EPL indicator provided by the OECD, for all workers. ²¹ The original index takes values in the range [0;5], increasing with strictness of employment protection. We rescaled it over [0;100] and introduce it in level in the regressions. Data are available for 1990, 1998 and 2003. They are interpolated over the period 1992-2002.

Centralization Degree of Bargaining is a discrete variable of bargaining centralization taken from OECD (2004). It ranges between 1 and 5 and is increasing in the degree of centralization: 1 = Company and plant level predominant, 2 = Combination of industry and company/plant level, with an important share of employees covered by company bargains, 3 = Industry level predominant, 4 = Predominantly industrial bargaining, but also recurrent central-level agreements, 5 = Central-level agreements of overriding importance. Information on this variable covers a 5-year period, on 1980–84, 1985–89, 1990–1994, 1995–2000. We conserve the most recent value for 2001 and 2002.

Benefit Replacement Ratio: We consider the gross replacement rates provided by the OECD’s Social and Welfare Statistics database. ²² It is defined as the average of the gross unemployment benefit replacement rates for two earnings levels, three family situations and three durations of unemployment. Raw data have one observation every 2 years, starting in 1985. We rely on interpolation for missing years.

Minimum Wage Legislation: The ratio of minimum wage to median wage is taken from the OECD’s Labor Force Statistics database. It corresponds to the minimum relative to median wages for full-time workers. It is available on a yearly basis for 14 OECD countries. Notice that Ireland and the United Kingdom had no legal minimum wage policy before 2000 and 1999 respectively. We complete these pieces of information using data from ILO Bureau of Statistics (LABORSTA database). This database contains legal and negotiated minimum wages in national currency and international US$ in 2003. This information is used to reconstitute series of minimum wages for countries in which minimum wages are negotiated at the sector level, that are not included in OECD data (precisely, Switzerland, Germany, Finland and Italy). For these countries, we build the series of minimum-to-median wage ratio as follows. First, as the ILO data have no time dimension, it has been assumed that negotiated minimum wages only adjust to inflation. Under this assumption, time series can be rebuilt using inflation series, calculated on consumption-price indices obtained from national sources. Second, we calculate the ratio of minimum to median wages using OECD Earnings data on gross median wages.

Table 9 displays the country coverage for the various LMI dimensions coming from OECD sources.

Appendix 2: More on the role of labor market institutions

To detect potential collinearity problems when simultaneously including various LMI in the regression, we calculate the correlation coefficients between our LMIs. They are reported in Table 10. Precisely, we report here the “between-country” correlation coefficient, i.e., considering the mean value of the LMI over the sample period for each country.

Table 10 Correlation between LMI dimensions

First, it is worth noticing that all correlations are positive, notably with the synthetic LMI indicator, and whatever the source of LMI dataset (EF or OECD). In line with expectations, a higher value of each labor market variable can be associated with a more rigid labor market functioning. Second, if positive and non-negligible, the correlations between the specific LMI variables of the same dataset (i.e, either EF LMI variables or OECD LMI variables) reported in Table 10 are not that strong to make us suspect serious multicollinearity problems when all included in the regression (Table 6). However, to ensure this point, we also run collinearity tests on the set of LMI variables on the set of LMI variables as reported by Economic Freedom (Table 11) and the OECD (Table 12).

Table 11 Collinearity diagnostics—Economic Freedom

Table 12 Collinearity diagnostics—OECD

These collinearity diagnostics are obtained from the SAS software collinearity test routine and derived from Belsley et al. (1980). In each table, the numbers in Column (1) correspond to different linear combinations of the LMI variables. The eigenvalues in Column (2) correspond to the variance of that combination. The condition index, obtained from those eigenvalues and reported in Column (3), is used to detect potential multicollinearity issues. Namely, it indicates whether the inversion of the matrix is numerically unstable with finite-precision numbers. A high condition index associated with the linear combination that explains most of the variance of a given variable (reported in columns (4) to (7)) can be interpreted as indicative of potential multicollinearity issues. Belsey et al. (1980) suggest that, when this number is around 10, weak dependencies may be starting to affect the regression estimates. When this number is larger than 100, the estimates may have a fair amount of numerical error. In our case, for both LMI datasets, the highest condition index is lower than 10, we thus conclude on the absence of collinearity issue between the regressors.