Determining the existence of Amakudari in publicly-traded Japanese companies: a Poisson regression approach

Abstract

Amakudari is a political-economic phenomenon frequently observed in the Japanese economy and other modern societies. Retired high-ranking bureaucrats are appointed as top executives in lucrative private firms that are controlled by the government ministry or main banks that Amakudari once regulated. These high-ranking ex-government officials use their influence for favourable legislation and treatment to help their companies. Due to the fact that Japanese publicly traded companies hire mostly a smaller portion of directors through Amakudari and sometimes none of them are hired, the most commonly used classical linear regression framework cannot be applied due to violating an assumption on the continuous dependent variable. Instead, this study adopts a counting-process regression approach in order to explain the existence of Amakudari using 2003 data of all publicly traded companies in Japan.

Acknowledgements

I would like to thank Prof. Yongil Jeon for his helpful discussions and advice. Min-ji Park and Chan-hee Yoon provided a valuable contribution to the compilation of the database. This study was partly supported by research fund of the Yonsei Center for Global Studies, Yonsei University.

Notes

¹ The Japanese meaning of Amakudari is ‘descent from heaven’, which is known as the revolving door in the US. Colignon and Usui (2003) define Amakudari in a narrow sense as ‘elite bureaucrats in the heavenly world moving down to the earthy private corporations upon retirements to take the lucrative boards of director positions’. (p. 20) In moving to their new positions, the former bureaucrats take with them invaluable insider knowledge about a ministry's policies and procedures, as well as the contracts and personal relationships developed during their careers (p. 3).

² See Schaede (1995), Calder (1998, 1999), Suzuki (2001), Horiuchi and Shimizu (2001) and Asano and Eto (2003).

³ A count variable takes on nonnegative integer values, and its examples include the number of children born to a woman, the number of times someone is arrested in a year, or the number of patents applied for by a firm in a year. See more on Wooldridge (2006, pp. 604–5).

⁴ The narrower definition of Amakudari is commonly used in the existing literature (Nakano, 1998; Colignon and Usui, 2003). We prefer using a broader concept of Amakudari because the appointment of directors in Japanese companies has not been determined mainly by the host companies' initiatives based on purely economic and managerial factors. Due to rigid cross-share holding relations and socio-cultural transactions between main banks and firms, many directors tend to be dispatched by the main banks and not recruited by the host firms. Therefore, the patterns of Amakudari from the three different subgroups display more similarities than differences.

Fig. 1. The number of board members vs. the number of Amakudari

⁵ Since Table 1 was made of two-dimensional tables, it is hard to make a three-dimension discussion based on these three subgroups. For example, it is hard to count the number of Amakudari within each company that come from government sectors, from financial sectors or from quasi-government sectors.

⁶ The regression experiments of this study indicate that industry, last year's sales and sales increase rate were not consistently significant.

⁷ See Cameron and Trivedi (1998) and Wooldridge (2006) for more information.

⁸ Wooldridge (2006, p. 606) indicates that, although Poisson analysis is a natural first step for count data, it is often much too restrictive since all probabilities and higher moments of the Poisson distribution are determined entirely by the mean. Wooldridge (2006, p. 609) says, ‘If we are interested in the effect of an independent variable on the mean response, there is little reason to go beyond Poisson regression’, which is of interest. Also considering more rigorous (fully or partially) parametric count regression models such as negative binomial model, simulated maximum likelihood, finite mixture models (in Cameron and Trivedi (2005)) will be a future work.