Using propensity matching estimators to evaluate the impact of privatization on wages

Abstract

Whether the transfer of ownership rights to the private sector leads to a decline or increase in wage growth is theoretically ambiguous, given that the outcome depends on the uncertain interaction between firms and workers. Using propensity matching techniques, this article investigates the effects of privatization on wages in the Portuguese banking industry. The empirical results, obtained from Quadros de Pessoal for the period between 1989 and 1997, generally show a negative (positive) short-run (long-run) effect of privatization on relative wage growth for both men and women retained in the privatized firms. Moreover, the results show that the most educated and experienced (oldest) workers, as well as those in the high skill occupational categories, were more likely to experience a negative wage effect.

Acknowledgements

I thank two anonymous referees, Mark Stewart, Odd Rune Straume, Ian Walker and participants at several seminars and conferences for valuable comments. I am indebted to the Ministério do Trabalho e da Solidariedade for allowing the availability of data from Quadros de Pessoal. Financial support was provided by the Ministério da Ciência e Tecnologia under the grant BD/SFRH/2000/1291.

Notes

¹ Some notable exceptions include Brainerd (2002), Haskel and Szymanski (1993), Ho et al. (2002), La Porta and Silanes (1999), Monteiro (2004), Parker and Martin (1996) and Peoples and Talley (2001). Megginson and Netter (2001) survey the empirical literature on privatization.

² Megginson et al. (1994) provide an excellent historical overview of postwar privatizations in developed countries. For a study of privatization effects in a large number of developing countries, see Al-Obaidan (2002).

³ For details, see the original papers of Heckman et al. (1997, 1998a) or the discussions in, e.g. Smith and Todd (2005) and Blundell and Dias (2000).

⁴ Sousa and Cruz (1995) describe and discuss the economic and financial situation of public enterprises.

⁵ This total number (10) of firms privatized in the banking industry does not coincide with the 11 privatized firms reported by the OECD 1999 survey, due to the absence of one bank in the data.

⁶ According to the privatization literature, the date of the first tranche sale of each firm is considered the date of effective privatization.

⁷ International investors could buy a limited share of the equity, ranging from 2 to 40% of sales.

⁸ A contrary conclusion is reached by Kraft et al. (2006), with data from the Croatian banking sector, showing that privatizations did not have an immediate effect on improved efficiency.

⁹ The unionization density includes unionized, both active and retired, employees.

¹⁰ Source: Greves Anual, Informação Estatística (Síntese), various years, MSST.

¹¹ Source: Own computations based on Colecção Relatórios e Análises, Série Regulamentação do Trabalho. Aumento Médio Ponderado Intertabelas e Aumentos Intertabelas Deflacionados, MSST and on bargaining contract data supplied by Sindicato Bancário do Norte (1981–1997).

¹² Unit of currency = escudos (PTE). 1 Euro = 200.482 PTE

¹³ In some cases, the corporate economic restructuring involves the adoption of less secure job (human resource) practices, including either temporary or partial employment, in order to achieve more flexible industrial relations. Cam (1999), for example, reports significant jumps in the number of temporary posts in the Turkish cement industry.

¹⁴ The T-test for the estimated wage difference between the treatment and the control group is statistically significant at the 1% level.

¹⁵ We are implicitly adopting the stable unit-treatment value assumption (SUTVA) first expressed by Rubin (1980). This assumption requires that an individual's potential outcome is independent of the treatment status of other individuals, ruling out any eventual within-group or spillover (general equilibrium) effect.

¹⁶‘Ignorable treatment assignment’, in the terminology of Rubin (1977) and Rosenbaum and Rubin (1983).

¹⁷ This assumption together with CIA are the ‘strong ignorability treatment assignment conditions’ in the terminology of Rosenbaum and Rubin (1983).

¹⁸ See Heckman et al. (1999) and Smith and Todd (2005) for a detailed description of a variety of different matching estimators.

¹⁹ This contrasts with the active labour market policies, in which both the policy and evaluation object targets coincide.

²⁰ Variables relating labour force status of treated individuals were found to be very significant (even more than earnings) in explaining the participation decision in training programs.

²¹ This post-treatment period choice is also conditioned by the first merger wave in 1998 in the banking industry, which involved recently privatized firms.

²² We also tried to include the size of the firm (for the banking sector, this is the only available firm characteristics variable in the data set) in order to control for observable selection of the firms being privatized. Perhaps because all banks are similar in size (with the exception of one bank that is substantially larger than the average), the overall impacts remained unaffected by the inclusion of this variable, affecting (worsening) only the quality of matching.

²³ We also tried different specifications for the propensity score, including these and other variables, such as the monthly wage before privatization, as suggested by the work of Heckman et al. (1998). Once again, the magnitude of the impacts remains unaffected by the inclusion of these variables in the specification, affecting (worsening) only the quality of matching.

²⁴ The matching estimates were obtained using command PSMATCH2 in Stata, written by Leuven and Sianesi (2003).

²⁵ Table 12 in the appendix shows that lack of common support is not an issue in the present evaluation.

²⁶ The results remain qualitatively unchanged if we adopt different neighbourhood sizes for each participant.

²⁷ See Rosen (1992) or Wolfram (1998) for a theoretical and empirical survey on executive pay levels.