Corruption and valuation of multinational corporations

Available online at www.sciencedirect.com

Journal of Empirical Finance 15 (2008) 387 – 417 www.elsevier.com/locate/jempfin

Corruption and valuation of multinational corporations Christos Pantzalis a,1 , Jung Chul Park b,2 , Ninon Sutton a,⁎ a

b

College of Business Administration, Department of Finance, University of South Florida Florida, 4202 E. Fowler Ave., BSN 3403, Tampa, FL 33620-5500, United States Department of Economics and Finance, College of Business, P. O. Box 10318, Louisiana Tech University, Ruston, LA 71272, United States Received 11 March 2006; received in revised form 15 May 2007; accepted 14 September 2007 Available online 26 September 2007

Abstract This paper examines the relationship between U.S. MNCs' valuation and corruption in countries where the MNCs' foreign subsidiaries are located. We uncover that country-level corruption has a multi-dimensional impact on MNCs' valuation. We find that the impact of intangibles is less pronounced for MNCs operating primarily in corrupt countries, consistent with the view that the lack of property rights protection and information asymmetry problems are more prevalent in corrupt environments. We also find that the expansion of a MNC network dominated by corrupt countries negatively affects MNCs' valuation, suggesting that investors may recognize it as an additional risk. However, more importantly, we find that geographic diversification in corrupt countries significantly increases firm value if the MNC has high levels of intangibles such as technological know-how and marketing expertise. Assuming that transactions costs in corrupt countries are higher, our findings are consistent with the notion that the advantages from internalizing the cross-border transfer of intangibles are greater in the presence of corruption. Our findings remain unchanged when we account for endogeneity at the country-and firm-level, when we use alternative corruption measures, and when we re-estimate models by omitting MNCs with operations in locations with big “negative” shocks during the sample period. Moreover, we show that firms with expertise in dealing with corruption enjoy greater benefits from internalization. © 2007 Elsevier B.V. All rights reserved. JEL classification: D73; F23; G30 Keywords: Corruption; Intangible assets; Multinational corporation

1. Introduction The degree of corruption in a country can have a profound impact on the country's macroeconomic development as well as firm valuations within the country. According to Transparency International (TI), corruption – the abuse of public power for private benefit – is still a widely spread phenomenon, even though many national governments have

⁎ Corresponding author. Tel.: +1 813 974 6337. E-mail addresses: [email protected] (C. Pantzalis), [email protected] (J.C. Park), [email protected] (N. Sutton). 1 Tel.: +1 813 974 6326. 2 Tel.: +1 318 257 3571; fax: +1 318 257 4253. 0927-5398/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.jempfin.2007.09.004

388

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

often undertaken efforts to reduce it. Consequently, corruption has become a prominent issue of concern within international institutions and with firms active in foreign markets. In addition, academic researchers in the areas of international finance and economics have recently been investigating the causes and important economic consequences of corruption in world markets.3 In this paper we examine corruption from the perspective of U.S. multinational corporations (hereafter MNCs). In particular, we investigate how the involvement of U.S. MNCs in corrupt countries affects firm value. We develop hypotheses derived from the basic premises of the internalization theory of MNCs and test how involvement in corrupt foreign countries affects the value impact of multinationality, intangibles, and internalization.4 For example, the lack of property rights protection and information asymmetry problems prevalent in corrupt environments (see, for example, Zhao (2006)) may decrease the value of applicable intangibles in a corrupt environment as compared to a “clean” environment. Furthermore, from a theoretical standpoint, corruption may have a negative impact on firm value if investors recognize it as an additional risk to which MNCs are exposed, and consequently adjust firm value lower.5 Thus, holding other things equal, the expansion of a corrupt-countries-dominated MNC foreign operations network should be value decreasing. On the other hand, due to the higher transactions costs in corrupt countries, one can also hypothesize that MNCs with substantial involvement in corrupt countries can benefit from internalizing the transfer of intangibles across borders. That is, for the same firm with the same level of intangible assets, the firm gets more value in the form of transactions cost savings from internalizing markets for the transfer of the intangibles in a corrupt environment than in a clean environment. Overall, the different facets of the relationship between corruption and MNC valuation remain largely unresolved. This paper aims at providing clear, unequivocal answers to these important questions. Our tests are performed on a sample of non-financial U.S. MNCs over the 1995–1998 and the 2002–2005 periods and utilize both OLS regressions and panel data regressions. Our multivariate tests provide evidence that supports the hypotheses developed based on the internalization theory. First, our results show that the general impact of intangibles on the value of MNCs is positive, but smaller for MNCs whose foreign operations networks include many corrupt countries than for MNCs that primarily operate in clean countries. Second, we find that the direct effect of multinationality (i.e. the size of the firm's foreign operations network) on market value is negative for the group of MNCs whose foreign operations are primarily in corrupt countries, while the effect is insignificant for MNCs operating in clean countries. However, the coefficient of the interaction term of intangibles with multinationality is significantly positive for MNCs operating primarily in corrupt environments, and negative and insignificant for MNCs operating in clean environments. This finding implies that, although intangibles have generally a lower impact on MNC valuation in corrupt countries, MNCs can reap greater benefits from internalizing intangibles within a corrupt countries-dominated foreign operations network because transactions costs are higher in corrupt countries. Our methodology also accounts for the possibility of two different endogenous relations between corruption level and unobservable variables. First, country-level corruption is likely to be associated with poverty and the type of legal environment. To account for this statistical problem, we utilize measures of country-level corruption, which are orthogonal to poverty and legal environment measures. Second, firm-level corruption – measured by an index which represents the firm's level of exposure to corruption in its foreign operations – is highly correlated with other firmspecific characteristics. We account for this by using a two-stage least squares (2SLS) model that estimates the firmlevel corruption index in the first-stage. Our evidence remains robust to the use of an alternative country-level measure 3

For example, Beck, Demirguc-Kunt, and Maksimovic (2005), Ehrlich and Liu (1999), Bliss and Di Tella (1997), Mauro (1995), and Shleifer and Vishny (1993), among others. 4 The internalization theory was developed by Buckley and Casson (1976), Casson (1979), Rugman (1981) and Hennart (1982), among others, and borrows its arguments from transactions costs economic theory as described in Williamson (1975). Morck and Yeung (1991, 1992) provide empirical evidence in support of the internalization theory. 5 Under traditional CAPM assumptions, idiosyncratic risk should not be rewarded with higher returns, since it reflects, by definition, diversifiable or unsystematic risk. However, past theoretical studies have shown that, once some of the CAPM assumptions are relaxed, idiosyncratic volatility (hereafter, IV) can be a priced factor. For example, in the CAPM extension of Levy (1978), where investors hold under-diversified portfolios, the relationship between idiosyncratic risk and expected returns is predicted to be positive. Similarly, in the models of Merton (1987) and Xu and Malkiel (2003), where investors can only invest in a subset of all available stocks due to various exogenous reasons, like incomplete information or transaction costs, imperfect diversification leads to investors requiring to be compensated with higher expected returns for holding stocks with high levels of IV. Finally, in the prospect theory model of Barberis and Huang (2001) idiosyncratic risk also produces higher expected returns. Moreover, recent empirical studies have also shown that there is a statistically significant relationship between IV and stock returns (see, for example, Goyal and Santa-Clara (2003), Bali, Cakici, Yan and Zhang (2005), Ang, Hodrick, Xing and Zhang (2006 and 2007), and Spiegel and Wang (2005)).

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

389

of corruption (see Kaufmann, Kraay, and Mastruzzi (2004)). In addition, we recognize that our results could be driven by our inability to correctly estimate chop shop Tobin's qs (see Yeung (2003)).6 To address this concern, we reestimate our model after omitting MNCs with operations in locations with big “negative” shocks, such as in countries affected by the Asian financial crisis in 1997 and the Brazilian Real problem in 1998. Our results are robust to this reestimation. We also provide evidence that prior experience with country-specific corruption helps enhance the benefit from internalizing the transfer of intangibles in the presence of corruption. Finally, we show that our overall results are not specific to the 1995–1998 period. When we repeat our tests using data covering a more recent period (2002–2005), we obtain similar results. This paper provides a contribution to both the international business and finance literature by providing evidence on how corruption affects MNC valuation. Our test results reveal an important facet of internalization theory by pointing to the role of corruption as the link between valuation of intangibles and multinationality. The rest of the paper is organized as follows. In the next section, we develop testable hypotheses of the relationship between corruption and valuation of U.S. MNCs. Section 3 describes the data sources and the sample selection. Section 4 introduces the measures of corruption level, and describes the different regression models. Section 5 reports univariate and multivariate results. Section 6 provides a summary and concluding remarks. 2. Corruption, degree of foreign involvement, and valuation of MNC intangibles: hypothesis development Internalization theory provides an answer to the question of why MNCs exist and how they create value for their shareholders. The theory suggests that firms increase value by internalizing markets for the transfer of intangibles across country borders. Intangible assets include technological know-how, patents, marketing expertise, managerial skills, or consumer goodwill. The value of intangible assets is, in general, positively related to the scale of the firm's markets. A firm possessing high levels of intangible assets can reap the benefit of reduction in transactions costs by creating intra-firm (i.e., internalizing) markets for such assets. How does corruption affect the degree of reliance of an MNC on its intangibles for generating value for its shareholders? It is widely accepted (see, for example, Johnson, McMillan, and Woodruff (2002)) that corrupt environments are associated with worse property rights protection. In addition, it is also shown that agency costs and information asymmetries are more pronounced in the presence of corruption (see Acemoglu and Zilibotti (1999), among others). Therefore, it follows that MNCs would prefer to maintain a high mode of control in corrupt environments and carefully safeguard proprietary knowledge-based advantages.7 However, since property rights protection decreases and agency problems increase with corruption, the value of applicable intangibles per se should also decrease. We formulate this hypothesis below. Hypothesis 1. All else being equal, the value impact of intangibles should be lower for MNCs that primarily operate in corrupt countries than for MNCs that primarily operate in clean countries. Some studies have suggested that corruption may have a positive effect on economic growth since it can act as “speed money (or grease money),” in the sense that it enables individuals to avoid bureaucratic delay (Leff (1964) and Huntington (1968)). However, most theoretical and empirical studies suggest that corruption distorts economic decisions. For example, it has been documented that corruption might (1) lower investment and thereby lead to slower economic growth (Rose-Ackerman (1978), Murphy, Shleifer, and Vishny (1991), Shleifer and Vishny (1993), Mauro (1995)), (2) increase the size of unofficial economic activity (De Soto (1989), Murphy, Shleifer, and Vishny (1993), and Friedman, Johnson, Kaufmann, and Zoido-Lobaton (2000)), (3) result in the misallocation of talent to individuals' occupations due to large opportunities for unproductive but beneficial rentseeking (Baumol (1990), and Murphy et al. (1991)), and (4) increase inequality in distribution of income (Li, Xu, and Zou (2000)). 6

Yeung (2003) provides a detailed description of the chop shop Tobin’s q problem in empirical tests of diversification effects on firm value. Smarzynska (2004), and Smarzynska and Wei (2002) find that foreign investors with more sophisticated technology prefer wholly-owned affiliates over partnerships with local firms, because these MNCs are expected to lose more if their knowledge-based advantages are dissipated to partners or other firms. Smarzynska and Wei (2002) also show that keeping the technological know-how level constant, a foreign investor is more likely to have a local partner in a more corrupt host country. 7

390

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Corruption is not only a key determinant of development and growth at the macroeconomic level, but also at the microeconomic level it may have an impact on MNC valuation. This argument is supported by findings of studies such as Hines (1995), Henisz (2000), and Wei (2000) who show that corruption has a negative impact on foreign direct investment. In addition, investors may associate operations in corrupt countries with exposure to additional risk and, therefore, they may discount MNC market values accordingly if the firm is heavily involved in corrupt environments. At the same time, the effect of operations in corrupt markets on MNC value may differ depending on the level of intangibles the MNC possesses. For example, MNCs with high level of intangibles operating in corrupt environments where external markets are characterized by high transactions costs can create intra-firm markets for the transfer of intangibles-related advantages and thereby lower transaction costs. Therefore, the valuation of MNCs with high level of intangibles can be positively associated with the degree of involvement in corrupt environments. However, MNCs without intangibles will not be able to reap the benefit of lower transactions costs that intra-firm markets provide. Thus, these firms with low internalization potential are likely to have difficulty expanding their operations' network in corrupt environments. Thus, our second hypothesis reflects two different expectations: Hypothesis 2a. All other things equal, and in the absence of intangibles, expanding a corrupt-countries-dominated foreign operations network has a negative impact on market value. Hypothesis 2b. All other things equal, for firms possessing intangibles, expanding a corrupt-countries-dominated foreign operations network has a positive impact on market value. Internalization theory relies heavily on the arguments of transaction costs economics. Williamson (1975) extensively analyzes the nature of the transactions costs involved in using the market mechanism for transferring intangibles. In particular, he summarizes transactions costs associated with organizing economic activity via the external market as 1) the costs of bringing the parties together, 2) the costs of negotiating the terms of the contract, 3) the costs of writing up a contract, and 4) the costs of overseeing contractual terms. Due to the magnitude of the transaction costs associated with the transfer of proprietary knowledge-based advantages across borders, foreign direct investment in wholly-owned (or majority-owned) subsidiaries can be less expensive than organizing the transfer via contracting in the external market. Thus, the greater the MNC's reliance on intangible assets, the greater is the possibility that it will attempt to expand geographically via a network of majority-owned subsidiaries in all foreign countries.8 Begovic (2005) argues that the very specific nature of corruption as an illegal contract generates its substantial transactions costs. Thus, in the presence of corruption, due to the illegal nature of the contract, transactions costs are multiplied relative to the case of standard legal contracts.9 Consequently, we argue that transaction costs in external markets will be higher in corrupt countries relative to clean countries. This reasoning implies that the efficiency benefits (in terms of lower transactions costs) from creating intra-firm markets for the transfer of intangibles-related advantages should be greater within a corrupt environment than within a clean environment. Furthermore, one can conjecture that since corruption is associated with low property rights and poorly developed capital markets, an MNC can reap additional benefits from the process of internalization in corrupt environments. This reasoning is based on Zhao (2006) and Desai, Foley and Hines (2006). Zhao (2006) illustrates that MNCs involved in countries with poor intellectual property rights protection and poor institutional environment are in a unique position to engage in internalization arbitrage. That is, in countries where R&D is discouraged and human capital is undervalued, 8

Williamson (1975) also explains that at least four factors may force a firm to internalize a transaction (i.e., to make rather than buy, or to organize economic activity within the firm rather than using contracts in the market): 1) specificity of assets —when physical and human assets are highly specific, firms will internalize; 2) quality of the product — when the quality of a good or service is hard to assess, a supplier will have an incentive to behave opportunistically because negotiating and enforcing a market contract is difficult. Thus, the firm may choose to produce the good or service within its organization; 3) frequency of transactions — when a repeated market transaction among a small number of participants allows for an immense scope for opportunistic behavior, a firm may internalize the transaction to reduce opportunism; and 4) environmental uncertainty — when a transaction is associated with high uncertainty, searching for relevant information is costly, so the firm’s internal supply may improve the information available. 9 Benham and Benham (2004) and Knack and Keefer (1997) are examples of other studies that have argued that transactions costs are affected by the level of corruption.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

391

MNCs can arbitrage the difference in factor prices across national borders using their internal organizations as a substitute for inadequate external institutions.10 Desai et al. (2006) show that MNCs employ internal capital markets opportunistically to overcome imperfections in external capital markets. Thus, based on the above, our third hypothesis can now be expressed as follows. Hypothesis 3. Internalization of intangibles provides greater value enhancement for MNCs operating primarily in corrupt countries than for MNCs operating primarily in clean countries. We proceed to the empirical design and testing of the above three hypotheses in the sections following the description of our sample. 3. Data The country-level scores of corruption are extracted from Transparency International (TI), which ranks countries based on the Corruption Perceptions Index (CPI). The CPI represents the degree of corruption among a particular country's public officials and politicians, derived from surveys of public perceptions. Survey results reflect the views of broad range of people because surveys are carried out among businesspeople and country analysts, including experts who are resident in countries evaluated by TI. The surveys include different questions about the misuse of public power for private benefit, such as bribe-taking by public officials in public procurement. CPI scores range from 0 (highly corrupt) to 10 (not corrupt).11 We re-define the corruption score of country j at year t (Cj,t) by subtracting CPIj,t from 11, so that a high value of this measure indicates a high level of corruption in country j. Next we use the corruption score Cj,t to construct a measure of the overall exposure of an MNC to corruption. This measure is the corruption index, CIi,t, which is computed as the sum of the weighted country-level corruption scores. CIi;t ¼

X

ð1Þ

Wi; j Cj;t

j

where Cj,t is country j's corruption score in year t. Wi,j is MNC i's weight for country j and is computed as the ratio of foreign subsidiaries located in country j divided by the total number of foreign subsidiaries of MNC i. Wi; j ¼ NFSi; j =

X

NFSi; j

ð2Þ

j

Thus, MNCs with a large number of subsidiaries in corrupt countries and only a few subsidiaries in noncorrupt countries should have a high corruption index, indicating high exposure to corruption in its foreign operations.12 The MNCs in our study meet the following criteria. First, the firms are publicly traded in the U.S. and are in the manufacturing and mining industries. They have at least US$10 million in foreign sales, and have paid some foreign taxes in the years covered in the study. In addition, the firms have at least one foreign affiliate recorded in the Dun and Bradstreet's 1998/99 issue of Who Owns Whom, which provides foreign affiliate information for the 1996/97 period. Finally, we require that the MNCs in the sample have at least one subsidiary in a country with corruption score (C) available. Using the above criteria, we identified 528 MNCs. Our study spans the 1995–1998 period.13

10

Harris, Morck, Slemrod, and Yeung (1993) also show how MNCs can utilize intangibles to transfer resources out of a country. Readers who are interested in detailed information about CPI and its construction can visit the official web-site of Transparency International at http://www.transparency.org/. 12 Note that subscripts i and j represent particular firm and country, respectively. 13 We assume that firms in the sample experience no change in subsidiary locations over a 4-year period, 1995–1998. While this may not be true in every case, it is reasonable to assume that the location and structure of subsidiaries is fairly stable. 11

392

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Financial data are collected from COMPUSTAT, and information related to stock price is extracted from the Center for Research in Securities Prices (CRSP) database. We require that accounting data be available for each MNC in every year. This requirement reduced the number of firms in the sample to 296 and the total number of firm-year observations used in the tests is 1184.14 4. Empirical methodology After constructing the corruption index, CIi,t, for each firm, we rank CIi,t in every year and create an indicator variable, HCIi,t, which takes the value of one if the MNC is ranked in the top half of CI rankings in each year, and the value of zero, otherwise. Thus, a HCIi,t value of one (zero) indicates a firm with a foreign operations network that is concentrated in countries with high (low) corruption scores. In the first stage of our empirical investigation, we employ univariate tests designed to illustrate whether the degree of corruption a MNC faces is correlated with a) valuation, b) intangibles' intensity, c) measures of the multinational network, and d) other firm characteristics. We also conduct univariate tests on subsamples created after sorting based on the level of intangibles MNCs possess as well as on the degree of multinationality, which is proxied by the total number of foreign subsidiaries. Our multivariate tests include both ordinary least squares (OLS) and panel data regressions. The valuation measure for the individual MNC firm is regressed on the level of geographical diversification, intangibles' intensity, and their interaction term. We also control for other firm-specific characteristics. The general form of the regression equation is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t

ð3Þ

where VALUEi,t is the valuation measure for the individual MNC i at time t. We measure firm's valuation in three ways. First, we use the Chung and Pruitt (1994) measure of TOBINQ1i,t, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets). Second, we adjust the Chung and Pruitt (1994) 's Tobin's q by adding intangibles to total assets in the denominator, TOBINQ2i,t. We make this adjustment to account for the possibility that MNCs with high corruption indexes may have a higher stock of intangibles, causing them to have Tobin's q measures that are upwardly biased when compared to lower corruption index firms.15 Last, we utilize firm's excess value, EXVi,t, by computing the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in Rhodes-Kropf, Robinson, and Viswanathan (2005). Fundamental value, V, is estimated by decomposing the market-to-book into two components: a measure of price to fundamentals (ln(M/V)), and a measure of fundamentals to book value (ln(V/B)). The first component captures the part of book-to-market associated with mispricing. In extreme cases where markets perfectly price stocks, this component would be equal to zero, otherwise positive (over-valuation) or negative (under-valuation). This component is further decomposed into firm-specific and industry-specific mispricing. In our tests, we use the firm-specific mispricing component based on Model III of Rhodes-Kropf et al. (2005) that also accounts for net income and leverage effects.       þ ln Mi;t ¼ a0j;t þ a1j;t ln Bi;t þ a2j;t lnðNIÞþ i;t þa3j;t Iðb0Þ lnðNIÞi;t þa4j;t ln LEVi;t þ 1i;t

ð4Þ

where M is firm value, B is book value, NI+ is absolute value of net income, I(b 0)ln(NI)+ is an indicator function for negative net income observations, and LEV is the leverage ratio. We obtain the fundamental value (V) by computing the fitted value from Eq. (4). INTi,t is a measure of intangibles' intensity, calculated as the sum of RDi,t and ADi,t where RDi,t and ADi,t are research and development expenditures and advertising expenditures, respectively, scaled

14 We restrict our panel data to be balanced. This is because our multivariate tests are based on OLS and random-effects regressions. Panel data sets that are either balanced or unbalanced are not problematic in panel regressions. For the OLS-based models, however, balanced panel data may provide more reliable results. 15 Given that Tobin’s q may also capture growth opportunities, and firms with high R&D tend to be high growth firms, a spurious correlation may affect the relationship between intangibles and Tobin’s q. Thus, we also employ alternative valuation measures to alleviate this concern.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

393

by sales. MNi is the multinational network variable as measured by the number of foreign subsidiaries (NFSi).16 The interaction between INTi,t and MNi captures the combined effect of intangibles and geographic diversification on valuation as implied by internalization theory. Xi,t is a vector of firm characteristics. Following past studies (e.g., Morck and Yeung (1991) and Berger and Ofek (1995)), the vector Xi,t contains variables capturing financial leverage (long-term debt ratio, measured as long-term debt divided by total assets, LEVi,t), capital expenditures' intensity (capital expenditures over sales, CAPXSi,t), size (natural log of total assets, SIZEi,t), industry diversification (number of business segments, NSEGi,t), and the degree of involvement in developing countries (the number of developing countries where foreign subsidiaries locate divided by the total number of countries, DEVEi,t). INDi (YEARt) is a vector of industry (year) indicator variables, which control for industry (year) fixed effects.17 A detailed description of all variables used in the tests is provided in the Appendix. Since we are interested in identifying whether the impact of intangibles, multinationality, and internalization on valuation varies with the degree of involvement in corrupt countries, we run regressions separately for the groups of highcorruption-index firms (HCIi,t = 1) and low-corruption-index firms (HCIi,t = 0). Then, the difference in the coefficients between the two groups of regressions is captured by including all firms and using interaction terms between HCIi,t and each variable in the original model. Thus, the coefficients, δ, represent differences in the coefficients for respective variables. VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ d0 HCIi;t þ d1 INTi;t ⁎HCIi;t þ d2 MNi 4HCIi;t þ d3 INTi;t ⁎MNi ⁎HCIi;t þ d4 Xi;t ⁎HCIi;t þ d5 INDi ⁎HCIi;t þ d6 YEARt ⁎HCIi;t þ ei;t

ð5Þ

The first hypothesis addresses that the value impact of intangibles should be lower for MNCs that primarily operate in corrupt countries than for MNCs that primarily operate in clean countries. Based on this hypothesis, we expect that the value impact of intangibles is positive for both MNC groups, but that the positive impact is stronger for MNCs whose multinational networks are concentrated more in clean countries. To test the value impact of intangibles, we need to consider the combined effect of intangibles, (β1 + β3MN), in Eq. (3).18 Then, we compare the sum of β1 and β3MN obtained from estimating the regression for low-corruption-index firms to that for high-corruption-index firms. If the first hypothesis is true, the regressions would show substantially larger and positive (β1 + β3MN) for lowcorruption-index firms than for high-corrupt-index firms. This is equivalent to testing whether the sum of the coefficients δ1 and δ3MN obtained from estimating model (5) is zero. The second hypothesis has two tiers. First, when MNCs do not possess intangible assets, multinationality's value impact should be negative when it entails expanding a corrupt-countries-dominated foreign operations network (Hypothesis 2a). Second, when MNCs have intangibles, an expansion of a corrupt countries-dominated network may have a positive impact on firm value (Hypothesis 2b). The coefficient of β2 obtained from estimating model (3) provides the empirical evidence to test Hypothesis 2a because in this case we test MNCs assumed to have no intangibles, which implies that the interaction term is zero and there is no coefficient β3. We also expect that β2 for low-corrupt-index firms (i.e., MNCs with clean-countries-dominated network) will be positive and larger than β2 for high-corrupt-index firms (i.e., MNCs with corrupt-countries-dominated network). If Hypothesis 2b is true, when MNCs possess intangibles and expand into corrupt countries, the positive effect of the interaction between multinationality and intangibles will exceed the negative “pure” effect of multinationality. That is, the sum 16

We also measure multinationality using the number of foreign countries (NFCi) and number of foreign regions (NFGi). The subsidiaries are assigned to eight major geographic regions, depending on the location of each subsidiary’s host country. The eight geographic regions are 1) the North American Free Trade Association, NAFTA, area that includes Canada and Mexico, 2) central and south America, 3) western Europe, 4) eastern Europe, 5) advanced Asia countries which consist of Hong Kong, Japan, South Korea, Singapore, and Taiwan, 6) remaining Asian countries, 7) Oceania area including Australia and New Zealand, and 8) Africa. The results based on these alternative multinationality measures, (NFCi) and (NFGi), are qualitatively similar to the ones reported in the paper. They are left out of the paper for the sake of brevity, but are available upon request. 17 We test this model with several different industry classifications and found that our results are not materially affected by the specific type of classification. 18 Because our dataset includes only MNCs, which by construction have at least one foreign subsidiary in their multinational network, it is not technically possible to observe the “direct” impact of intangibles (β1 only) by assuming that there is no multinationality (MN = 0) and excluding the impact of internalization (β3).

394

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

of β2 and β3INT will be positive. We will also examine the break-point percentage of intangibles' intensity (intangible assets /sales). Most importantly, the last hypothesis argues that the value impact from internalization, which concerns the combined impact from intangibles and multinationality, is greater for MNCs operating primarily in corrupt countries than for MNCs operating primarily in clean countries. The coefficient of the interaction term between intangibles' intensity (INT) and multinationality (MN) captures the internalization effect on firm value. Thus, we will test whether β3 for high-corrupt-index firms is significantly larger than β3 for low-corrupt-index firms. Our dataset comprises a panel of 296 firms spanning four years. Therefore the typical OLS regression model may yield biased coefficients because it does not account for time-invariant, unobservable firm characteristics. Therefore, we also estimate the model using the random effects approach. We estimate the random rather than the fixed effects version of model (3) due to couple of reasons. First, Greene (1993) indicates that the fixedeffects model is a solution to this problem when the researcher is confident that differences between units (in our case, firms) can be regarded as parametric shifts of the regression function, which is a reasonable assumption when the sample consists of the full population of units. If this is not the case, i.e. when the sample is drawn from a large population as in our study, then it might be more appropriate to view constant terms as randomly distributed across cross-sectional units. Because our data represents a panel of 296 firms over 4 years from a large population, it appears that the random effects model is a better fit. Second, we are technically unable to use a fixed effects model because the multinational network variable (MN), by construction, is time-invariant in the sample period. As addressed above, we use the foreign affiliate information available for the 1996/97 period as measure of MN, assuming that the multinational network does not change over the four-year sample period, 1995–1998. We recognize two different endogenous relations that may exist between corruption and unobservable variables at the country-and firm-level. First, country-level corruption (the corruption score C) is likely to be associated with a country's poverty level and the country's legal environment. In general, poor countries are more likely to suffer from the high level of corruption. We use GDP per capita as a proxy for the country's level of poverty.19 La Porta, Lopez-DeSilanes, Shleifer and Vishny (1997a, 1997b, 1998, 1999, 2000, hereafter LLSV) show the connection between the origin of a country's legal environment, corporate finance practices, firm valuation, and the country's level of corruption. Following LLSV, we divided law origins into five groups: BRITISH common law (BRITISH), FRENCH civil law (FRENCH), German civil law (GERMAN), SCANDINAVIAN civil law (SCANDINAVIAN), and SOCIALIST law (SOCIALIST). We regress country-level corruption scores Cj,t on GDPj,t and a vector of dummy variables indicating type of legal origin. 20 Cj;t ¼ g0 þ g1 GDPj;t þ g2 FRENCHj;t þ g3 GERMANj;t þ g4 SCANDINAVIANj;t þ g5 SOCIALISTj;t þ gj;t ð6Þ In a statistical sense, the residual, ηj,t, in Eq. (6) excludes all effects of poverty and legal system origins on corruption level in particular country j. Therefore, it is regarded as a cleaner, orthogonal country-level corruption score. Our tests are also performed using regression models that utilize the orthogonal corruption index (OCIi,t) which is constructed in a same fashion as CIi,t, but uses orthogonal corruption (OCj,t) instead of original corruption scores (Cj,t). The second type of endogeneity problem that our tests account for is the possibility that the firm-level corruption index can be endogenously associated with firm-specific characteristics. There are some firm characteristics, which might be correlated with the corruption index. Consider, for example, the firms that depend highly on foreign markets or have extensive operations in foreign countries. Such firms, i.e., MNCs with high foreign sales ratios or large number of foreign subsidiaries (countries, or regions), also tend to have higher corruption index values. Thus, MNCs that maintain high foreign sales might be more likely to have higher corruption indexes than other MNCs. If our model fails to control for the relationship between foreign market dependence and the corruption index, the empirical results on

19

We collect GDP information from three different sources: International Monetary Fund, World Bank, and Penn World Table Version 6.1. Gross Domestic Product (GDP) is a widely-used measure of a country’s economic growth (see Levine and Zervos (1998) and Borensztein, DeGregorio, and Lee (1998), among many others). 20

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

395

firm value could be mistakenly attributed simply to the corruption index itself rather than to the real reason, which in this case would be the dependency on foreign markets. To reduce the potential problem of endogeneity in the multiple regression models, we use a two-stage least squares (2SLS) model which involves the simultaneous estimation of regression models (7) and (8). The 2SLS procedure requires that the first stage equation contain at least one instrumental variable that is unrelated to firm value and therefore is not included in the second stage model. Here, we use the foreign sales ratio (TFSALEPi,t) as an instrument.21 Thus, our full 2SLS model is structured as: First-stage model: HCIi;t ¼ a0 þ a1 INTi;t þ a2 MNi þ a3 NSEGi;t þ a4 TFSALEPi;t þ a5 INDi þ a6 YEARt þ wi;t

ð7Þ

Second-stage model: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ d0 b HCIi;t ⁎ b ⁎b ⁎ ⁎ b þ d1 INTi;t HCIi;t þ d2 MNi HCIi;t þ d3 INTi;t MNi HCIi;t þ 1i;t

ð8Þ

The first stage equation is a probit model with the high-corruption-index dummy (HCIi,t) as the dependent variable. The following set of variables are included as explanatory variables: intangibles' intensity (INTi,t), multinationality (MNi, number of foreign subsidiaries), industry diversification (NSEGi,t), foreign sales ratio (TFSALEPi,t), industry dummies (INDi), and year dummies (YEARt).22 The second stage model estimates the firm valuation as a function of b interactions with the fitted value of HCIi,t from the first stage equation, HCI i;t , and several firm characteristics. As an alternative specification, we use a continuous variable (CIi,t) for the dependent variable in the first stage equation, and b use the fitted value( CI i;t ) to estimate firm value in the second stage equation. At the end of our analysis, we address concerns about certain factors that could be driving our results. First, we recognize that the results of our tests can be attributed to the use of a specific corruption measure, i.e. the one based on country-level corruption scores provided by Transparency International. As a robustness check we repeat our tests using an alternative country-level measure of corruption, KCj,t, constructed by Kaufmann, Kraay, and Mastruzzi (2004) and widely used in other papers. In their paper Kaufmann et al. (2004) define corruption as the exercise of public power for private gain and provide “Control of Corruption,” which measures perceptions of corruption.23 The Control of Corruption indicator aggregates more data from various sources and covers more countries than the Corruption Perception Index (CPI) of Transparency International. The Control of Corruption indicator ranges from − 1.12 to 2.58, with low values representing high levels of corruption. As we did with the CPI, we subtract the Control of Corruption indicator from 3 to create a measure that increases with corruption. Second, we also recognize that our results could be driven by our inability to correctly estimate chop shop Tobin's qs as addressed by Yeung (2003). In a statistical sense, this problem cannot be solved by simply accounting for yearly fixed effects in the regressions. In fact, this problem cannot be completely rectified because we simply do not have the information necessary to estimate chop shop Tobin's qs for MNCs. This problem characterizes all studies that do not utilize an event study methodology, but it is more severe when the sample spans time periods and locations that experienced valuation shocks. To minimize this problem, we re-estimate our model after omitting from our sample MNCs with operations in locations with big negative shocks. In particular, 21 Because we employ one instrumental variable, TFSALEP, and have two endogenous variables, our IV-regression models meet the order and rank conditions for identification. For TFSALEP to be used as an instrumental variable, TFSALEP should have a significant relationship with the firm’s corruption index, but should not be correlated with the error term of the second-stage model. The foreign sales ratio is highly correlated with the high-corruption-index dummy (HCI). The correlation coefficient is 0.099 and statistically significant at the 1% level. This indicates that MNCs with high foreign sales ratios are likely to be involved in corrupt countries. In our sample, however, the coefficient of correlation between TFSALEP and the error term of the second-stage model is just 0.043 and not statistically significant. 22 Based on the nature of their operations, different types of companies may have different motivations for operating in a corrupt country. We attempt to address this point by controlling for different industry types, using the Fama-French industry categorizations. The industries are broken down into 5 basic categories, as follows: (1) consumer products, (2) manufacturing, (3) high tech, (4) health, and (5) other. The industry classification is available at Kenneth R. French’s website, http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. 23 This measurement, Control of Corruption, was first constructed in Kaufmann, Kraay, and Zoido-Lobaton (1999a, 1999b).

396

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

the corrupt locations likely include certain Asian countries and Latin American countries where financial market crises occurred. For example, Hong Kong, Indonesia, Malaysia, Philippines, South Korea, and Thailand experienced the Asian Financial Crisis in 1997, and Brazil had a similar crisis in 1998. These big negative shocks, Table 1 Variable description and summary statistics

Panel A: summary statistics for firm-level variables Corruption index (CI) Tobin's q (TOBINQ1) Adjusted Tobin's q (TOBINQ2) Excess value (EXV) R&D intensity (RD) Advertising intensity (AD) Intangibles' intensity (INT) Leverage (LEV) Size (SIZE) Capital expenditure (CAPXS) Industrial diversification (DIVERS) Number of segments (NSEG) Foreign sales ratio (TFSALEP) Involvement in developing countries (DEVE) Number of foreign subsidiaries (NFS) Number of foreign countries (NFC) Number of foreign regions (NFR)

N

Mean

Std. Dev.

Median

10th percentile

90th percentile

1184 1184 1184 1110 1184 1184 1184 1184 1184 1184 1184 1184 1184 1184 1184 1184 1184

3.496 1.543 1.427 0.173 0.048 0.012 0.060 0.170 20.48 0.079 0.438 1.934 0.383 0.255 20.59 9.182 3.267

0.916 1.455 1.260 0.513 0.062 0.037 0.069 0.145 1.734 0.144 – 1.344 0.231 0.234 29.83 8.751 2.014

3.500 1.120 1.067 0.124 0.025 0 0.035 0.149 20.49 0.052 – 1 0.332 0 9 6 3

2.300 0.431 0.407 − 0.404 0 0 0 0 18.17 0.022 – 1 0.147 0.220 1 1 1

4.575 3.020 2.743 0.820 0.125 0.044 0.149 0.431 22.80 0.135 – 4 0.653 0.500 47 21 6

6.270 10,489 0.315 0.435 0.065 0.043 0.141

2.305 8752 – – – – –

7.000 7185 – – – – –

2.300 1470 – – – – –

8.700 23.5 – – – – –

Panel B: summary statistics for country-level variables Corruption score (C) 368 GDP per capita (GDP) 368 Legal origin dummy —BRITISH 368 Legal origin dummy —FRENCH 368 Legal origin dummy —GERMAN 368 Legal origin dummy —SCANDINAVIAN 368 Legal origin dummy —SOCIALIST 368

The sample contains 296 firms (totally 1184 firm observations) and 92 countries (totally 368 country observations) over the period 1995–1998. Panel A reports descriptive statistics for firm-level variables. The corruption index, CIi is computed as the sum over the weighted corruption scores. CIi;t ¼

X

Wi; j Cj;t

j

where Cj,t is country j's corruption score in year t, which is computed by subtracting Transparency International's Corruption Perception Index (CPI) from 11, and Wi,j is MNC i's weight for country j and is computed as the ratio of foreign subsidiaries located in country j divided by the total number of foreign subsidiaries of MNC i. We use three measures (TOBINQ1, TOBINQ2, and EXV) for firm's valuation. TOBINQ1 is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). TOBINQ2 is the adjusted version of Tobin's q by adding intangibles to total assets in the denominator of TOBINQ1. EXV is the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in RhodesKropf, Robinson, and Viswanathan (2005). RD, AD, and CAPXS are R&D expenditures, advertising expenditures, and capital expenditures, respectively, scaled by sales. INT is intangibles' intensity and computed as a sum of RD and AD. LEV is a measure of financial leverage, measured as long-term debt / total assets. SIZE is the natural log of total assets. DIVERS is an indicator variable for industrial diversification, which takes the value of one if the firm reports two or more business segments and the value of zero if it reports only one segment. NSEG is the number of business segments. TFSALEP is the foreign sales ratio, computed as total foreign sales as percentage of total firm sales. DEVE is the degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. Three variables are used to measure multinationality; NFS = number of foreign subsidiaries; NFC = number of foreign countries; NFR = number of foreign regions. Panel B reports descriptive statistics for country-level variables. Matching information from two data sources, Dun and Bradstreet's Who owns Whom and Transparency International's annual CPI reports, results in a set of 92 countries with complete information, so that the number of total observations is 368 over our four-year sample period. Cj,t is country j's corruption score in year t. GDP per capita in US dollar (GDPj,t) is a proxy for the country's poverty level and is collected from three sources: International Monetary Fund, World Bank, and Penn World Table Version 6.1. Legal origins (Lj) are divided into five groups and indicated by five indicator variables: BRITISH common law (BRITISH), FRENCH civil law (FRENCH), German civil law (GERMAN), SCANDINAVIAN civil law (SCANDINAVIAN), and SOCIALIST law (SOCIALIST). Legal origin data is employed from La Porta, Lopez-De-Silanes, Shleifer and Vishny (1997a, 1997b, 1998, 1999, 2000).

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

397

if not taken out, make comparisons of Tobin's qs among firms that invest and do not invest in these corrupt countries less credible. Third, Shleifer and Vishny (1993) and Wei (2000), among others, document that experience with corruption matters in MNCs' operations in foreign countries. Therefore, we re-estimate our regression models to examine whether the value impact of intangibles differs based on the MNCs level of expertise with foreign country corruption. To conduct this test, we define an MNC as having expertise with corruption if it has at least two subsidiaries in any one corrupt country. The rationale of this measure is that a firm that increases its presence in a particular corrupt country beyond the initial stage of a single subsidiary must have gathered enough experience on how to deal with country-specific corruption issues. For example, consider firm A that has two of its subsidiaries in one corrupt country and firm B that has a number of subsidiaries in non-corrupt countries and only one subsidiary in each corrupt country of its foreign operations network. Based on the aforementioned criterion, firm A is considered as a firm that possesses expertise with corruption. In contrast, firm B has little expertise with corruption, even though it has many subsidiaries. This example represents an extreme case. In our sample most expert firms have many subsidiaries in many corrupt countries. A corrupt country is defined as one with a corruption score greater than or equal to 6 (i.e. lower than or equal to 5 in TI's CPI). Using this classification, we find 295 observations of firms with corruption expertise and 889 observations without corruption expertise. Last, we recognize that our sample period (1995–1998) may exhibit a specific time trend on financial markets. For example, Tobin's q ratios were strongly trending upward during this period. To generalize our empirical evidence, we test the models over a more recent period. In sum, we start our analysis from OLS and random-effects models. To account for the endogeneity problem, we compute orthogonal corruption scores and also use 2SLS. In our robustness tests, we 1) use an alternative corruption measurement, 2) exclude sample firms whose subsidiaries are located in countries experiencing a financial crisis, 3) re-group the sample using different criteria, and 4) analyze a different time period. 5. Empirical results 5.1. Univariate tests' results Panel A of Table 1 reports descriptive statistics for firm-level variables in the pooled sample. The average corruption index is 3.5, which implies that most MNCs have the majority of their foreign subsidiaries in less corrupt countries. This is consistent with the evidence of Hines (1995). Generally, MNCs in our sample are highly valued; the average Tobin's' q is 1.5, the adjusted Tobin's q is 1.4, and the excess value is 0.17. These companies spend on average six percent of their total sales revenues for R&D and advertising. The mean book value of total assets is 785 million dollars (exponential of 20.481). Companies diversifying in multiple industries account for forty-four percent of the total number of observations in our sample. Ninety percent of firms in our sample operate in less than four different four-digit SIC industries, while the average firm reports fewer than two business segments. MNCs in our sample generate a high percentage of sales in foreign markets (38%). On average, these companies operate in more than three foreign regions (nine foreign countries) and have more than twenty foreign subsidiaries. Panel B shows summary statistics for country-level variables. Average GDP per capita is 10,489 in U.S. dollars. Ninety-two countries in our sample consist of twenty-nine BRITISH common law countries (31.5%), forty FRENCH civil law countries (43.5%), six GERMAN civil law countries (6.5%), four SCANDINAVIAN civil law countries (4.3%), and thirteen SOCIALIST law countries (14.1%). In Table 2, we present univariate tests using three different sorting procedures to classify firms into different subsamples. Panel A reports mean values of all variables used in the study for the two sub-samples consisting of MNCs with high and low concentration of foreign subsidiaries in corrupt countries, respectively. Because high-and low-corruption-index firms are classified based on the median value of the corruption index, the number of observations in the two groups is identical. Also reported are the mean differences across the two groups and the corresponding t-statistics for the mean difference test. On average, MNCs with high corruption indexes are more highly valued than MNCs with low corruption indexes. Both groups on average spend almost the same percentage of total sales revenues for R&D but MNCs with high corruption indexes spend more for advertising. MNCs with high concentration of subsidiaries in corrupt countries are substantially bigger and also more diversified across

398

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table 2 Mean difference tests for sub-groups of MNCs Panel A: comparison of mean values of sub-groups based on corruption index

Corruption index (CI) Tobin's q (TOBINQ1) Adjusted Tobin's q (TOBINQ2) Excess value (EXV) R&D intensity (RD) Advertising intensity (AD) Intangibles' intensity (INT) Leverage (LEV) Size (SIZE) Capital expenditure (CAPXS) Industrial diversification (DIVERS) Number of segments (NSEG) Foreign sales ratio (TFSALEP) Involvement in developing countries (DEVE) Number of foreign subsidiaries (NFS) Number of foreign countries (NFC) Number of foreign regions (NFR)

[1]

[2]

High-CI-ranked MNCs

Low-CI-ranked MNCs

(i.e., MNCs with corruptcountries-dominated network MNCs)

(i.e., MNCs with cleancountries-dominated network MNCs)

Mean difference

t-statistics of test for null hypothesis:

N = 592

N = 592

[1] − [2]

Mean difference = 0

2.797 1.452 1.349 0.122 0.047 0.006 0.053 0.171 19.829 0.083 0.380 1.698 0.360 0.265 12.632 5.414 2.340

1.398⁎⁎⁎ 0.182⁎⁎ 0.157⁎⁎ 0.101⁎⁎⁎ 0.001 0.012⁎⁎⁎ 0.013⁎⁎⁎ − 0.002 1.304⁎⁎⁎ − 0.009 0.115⁎⁎⁎ 0.473⁎⁎⁎ 0.046⁎⁎⁎ − 0.020 15.912⁎⁎⁎ 7.537⁎⁎⁎ 1.855⁎⁎⁎

40.63 2.16 2.14 3.31 0.39 5.54 3.28 − 0.24 13.96 − 1.07 4.01 6.15 3.43 − 1.47 9.52 16.41 17.85

4.195 1.634 1.506 0.224 0.048 0.018 0.066 0.169 21.133 0.074 0.495 2.171 0.405 0.245 28.544 12.951 4.194

Panel B: comparison of mean values of sub-groups based on the level of intangibles (INT) MNCs possess

Corruption index (CI) Tobin's q (TOBINQ1) Adjusted Tobin's q (TOBINQ2) Excess value (EXV) R&D intensity (RD) Advertising intensity (AD) Intangibles' intensity (INT) Leverage (LEV) Size (SIZE) Capital expenditure (CAPXS) Industrial diversification (DIVERS) Number of segments (NSEG) Foreign sales ratio (TFSALEP) Involvement in developing countries (DEVE) Number of foreign subsidiaries (NFS) Number of foreign countries (NFC) Number of foreign regions (NFR)

[1]

[2]

MNCs with high level of intangibles

MNCs with low level of intangibles

Mean difference

t-statistics of test for null hypothesis:

N = 592

N = 592

[1] − [2]

Mean difference = 0

3.389 1.162 1.143 0.114 0.012 0.001 0.013 0.201 20.459 0.091 0.546 2.073 0.342 0.262 16.655 7.439 3.118

0.215⁎⁎⁎ 0.763⁎⁎⁎ 0.568⁎⁎⁎ 0.116⁎⁎⁎ 0.072⁎⁎⁎ 0.022⁎⁎⁎ 0.094⁎⁎⁎ − 0.062⁎⁎⁎ 0.045 − 0.024⁎⁎⁎ − 0.216⁎⁎⁎ − 0.277⁎⁎⁎ 0.081⁎⁎⁎ − 0.014 7.865⁎⁎⁎ 3.486⁎⁎⁎ 0.297⁎⁎

4.07 9.34 7.96 3.79 24.64 10.95 32.02 − 7.53 0.44 − 2.83 − 7.68 − 3.56 6.12 − 1.00 4.57 6.99 2.55

3.604 1.925 1.712 0.230 0.083 0.023 0.107 0.139 20.504 0.067 0.329 1.796 0.423 0.248 24.520 10.926 3.416

business segments and countries than MNCs with low concentration of subsidiaries in corrupt countries. In Panel B, we sort MNCs based on the level of intangibles they possess. MNCs with high levels of technological know-how and marketing and managerial skills tend to diversify in many countries, even corrupt ones. Thus, their degree of multinationality and their corruption indexes are substantially higher than for MNCs with low intangibles. As

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

399

Table 2 (continued ) Panel C: comparison of mean values of sub-groups based on the degree of multinationality (MN)

Corruption index (CI) Tobin's q (TOBINQ1) Adjusted Tobin's q (TOBINQ2) Excess value (EXV) R&D intensity (RD) Advertising intensity (AD) Intangibles' intensity (INT) Leverage (LEV) Size (SIZE) Capital expenditure (CAPXS) Industrial diversification (DIVERS) Number of segments (NSEG) Foreign sales ratio (TFSALEP) Involvement in developing countries (DEVE) Number of foreign subsidiaries (NFS) Number of foreign countries (NFC) Number of foreign regions (NFR)

[1]

[2]

MNCs with high degree of multinationality

MNCs with low degree of multinationality

Mean difference

t-statistics of test for null hypothesis:

N = 616

N = 568

[1] − [2]

Mean difference = 0

3.226 1.424 1.315 0.062 0.054 0.008 0.062 0.160 19.337 0.089 0.312 1.519 0.350 0.267 3.606 2.951 1.746

0.520⁎⁎⁎ 0.230⁎⁎⁎ 0.216⁎⁎⁎ 0.215⁎⁎⁎ − 0.013⁎⁎⁎ 0.009⁎⁎⁎ − 0.004 0.018⁎⁎ 2.200⁎⁎⁎ − 0.019⁎⁎ 0.242⁎⁎⁎ 0.797⁎⁎⁎ 0.062⁎⁎⁎ − 0.024⁎ 32.641⁎⁎⁎ 11.987⁎⁎⁎ 2.922⁎⁎⁎

10.17 2.73 2.95 7.12 −3.54 4.00 −1.02 2.15 28.19 −2.33 8.64 10.67 4.63 −1.75 22.45 32.24 36.20

3.745 1.654 1.531 0.277 0.041 0.016 0.058 0.179 21.537 0.069 0.554 2.317 0.412 0.243 36.247 14.929 4.669

This table compares mean values of variables based on three classifications. In Panel A, sample firms are divided into two groups based on corruption index. High-CI-ranked MNCs (i.e. HCI = 1) are firms ranked in the highest 1/2 of CI rankings in each year, while the remaining firms are classified as the low-CI-MNCs (i.e. HCI = 0). Corruption index (CI) is computed as the sum over the weighted corruption scores as defined in Table 1. In Panel B, sample firms are divided into two groups based on the level of intangibles (INT) MNCs possess. MNCs with high intangibles are those whose intangibles ranked in the upper half of sample firms, while MNCs with low intangibles ranked in the lower half of sample firms. In Panel C, sample firms are classified into two groups in terms of multinationality (MN) which is proxied by the number of foreign subsidiaries (NFS). MNC's with a high degree of multinationality are those whose number of foreign subsidiaries ranked in the top half of sample firms, while MNCs with a low degree of multinationality are those ranking in the bottom half of sample firms. The last two columns report the mean differences and corresponding tstatistic. The variables examined are defined as follows. We use three measures (TOBINQ1, TOBINQ2, and EXV) for firm's valuation. TOBINQ1 is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). TOBINQ2 is the adjusted version of Tobin's q by adding intangibles to total assets in the denominator of TOBINQ1. EXV is the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in Rhodes-Kropf, Robinson, and Viswanathan (2005). RD, AD, and CAPXS are R&D expenditures, advertising expenditures, and capital expenditures, respectively, scaled by sales. INT is intangibles' intensity and computed as a sum of RD and AD. LEV is a measure of financial leverage, measured as long-term debt/total assets. SIZE is the natural log of total assets. DIVERS is an indicator variable for industrial diversification, which takes the value of one if the firm reports two or more business segments and the value of zero if it reports only one segment. NSEG is the number of business segments. TFSALEP is the foreign sales ratio, computed as total foreign sales as percentage of total firm sales. DEVE is the degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. Three variables are used to measure multinationality; NFS = number of foreign subsidiaries; NFC = number of foreign countries; NFR = number of foreign regions. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

expected, the intangibles MNCs possess are positively related to MNCs' value, based on all three measures of valuation. For example, the mean of Tobin's q for MNCs with high level of intangibles is 1.925, which is 0.763 higher than that for MNCs with low level of intangibles. The difference is statistically significant at the one percent level with a t-statistic of 9.34. Panel C shows that the results from univariate tests based on subsamples created when sorting on multinationality are also consistent with our expectations. High-multinationality group firms have, on average, thirty-six foreign subsidiaries, compared with only three subsidiaries for low-multinationality group firms. The degree of geographical diversification is positively associated with firm valuation and exposure to corruption. In the following test (Table 3), we examine the joint effect of corruption exposure and intangibles intensity on firm value by analyzing subsamples based on firms' exposure to corruption in foreign operations and on the firms' level of intangibles. Specifically, we independently sort MNCs based on both corruption index and intangibles. Firms belonging to the highest (lowest) 33rd percentile of CI are assigned to the highest (lowest) CI group, while the

400

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

remaining firms are assigned to the middle group. Firms are also classified into a low, medium, or high groups based on their levels of intangible assets (INT). The averages of the valuation measures show that higher levels of intangible assets are associated with higher firm value. This is shown in the fourth column of Table 3, which presents a monotonic increase in firm value. If we compare firm values in terms of the level of corruption index, we find that the firm values are in general highest when their corruption index is moderate. In other words, firm values are enhanced when their subsidiaries locate in moderately corrupt countries, or when countries where their subsidiaries locate are evenly mixed so that on average the corruption index is moderate. However, if we look at firms with the highest level of intangibles, the average firm value is highest when these firms are also greatly exposed to corruption. In the fifth row, for example,

Table 3 Tobin's q of groups of firms formed after sorting on corruption index and the intangibles' intensity [2]

[3] highest CI

All MNCs

[3] − [1] (t-statistics)

Panel A: comparison of TOBINQ1 [1] 1.076 Low INTS [0.820] [2] 1.201 [0.982] [3] 1.672 High INTS [1.800] All MNCs 1.252 [1.185] [3] − [1] 0.596⁎⁎⁎ (t-statistics) (3.74)

1.142 [0.808] 1.558 [0.992] 2.272 [2.301] 1.742 [1.697] 1.129⁎⁎⁎ (4.86)

1.090 [0.615] 1.227 [0.727] 2.480 [1.883] 1.626 [1.386] 1.391⁎⁎⁎ (7.17)

1.098 [0.767] 1.327 [0.909] 2.207 [2.071] 1.543 [1.455]

0.014 (0.15) 0.026 (0.25) 0.808⁎⁎⁎ (3.21)

Panel B: comparison of TOBINQ2 [1] 1.069 Low INTS [0.809] [2] 1.146 [0.934] [3] 1.429 High INTS [1.397] All MNCs 1.176 [1.018] [3] − [1] 0.360⁎⁎⁎ (t-statistics) (2.68)

1.132 [0.800] 1.491 [0.946] 1.975 [1.967] 1.596 [1.469] 0.842⁎⁎⁎ (4.19)

1.081 [0.609] 1.180 [0.697] 2.185 [1.634] 1.503 [1.207] 1.104⁎⁎⁎ (6.49)

1.089 [0.758] 1.271 [0.867] 1.923 [1.755] 1.427 [1.260]

0.012 (0.13) 0.034 (0.34) 0.756⁎⁎⁎ (3.60)

Panel C: comparison of EXV [1] 0.043 Low INTS [0.420] [2] 0.010 [0.427] [3] 0.118 High INTS [0.759] All MNCs 0.051 [0.527] [3] − [1] 0.075 (t-statistics) (0.99)

0.184 [0.350] 0.205 [0.411] 0.284 [0.628] 0.233 [0.504] 0.100 (1.46)

0.133 [0.358] 0.152 [0.348] 0.376 [0.639] 0.229 [0.489] 0.242⁎⁎⁎ (3.26)

0.108 [0.389] 0.126 [0.402] 0.277 [0.669] 0.173 [0.513]

0.090⁎ (1.71) 0.142⁎⁎⁎ (2.92) 0.258⁎⁎⁎ (2.72)

[1] lowest CI

This table reports the mean and the standard deviation [in brackets] of Tobin's q for groups of firms formed after sorting independently on the corruption index (CI) and the intangibles intensity (INT). Corruption index (CI) is computed as the sum over the weighted corruption scores as defined in Table 1. We use three measures (TOBINQ1, TOBINQ2, and EXV) for firm's valuation. TOBINQ1 is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). TOBINQ2 is the adjusted version of Tobin's q by adding intangibles to total assets in the denominator of TOBINQ1. EXV is the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in Rhodes-Kropf, Robinson, and Viswanathan (2005). INT is intangibles' intensity and computed as a sum of RD (R&D expenditures/sales) and AD (advertising expenditures/sales). Firms belonging to the highest (lowest) 33rd percentile of CI are classified in the highest (lowest) CI group, while the remaining firms are classified into the [2] group. Then firms are independently sorted on the level of intangibles intensity using the same method. Also reported are mean differences between sub-groups and corresponding t-statistic (in parentheses). ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

401

the difference of TOBINQ1 between the high and low intangibles' firms is only 0.6 for the lowest-CI group, but it increases up to 1.4 for the highest-CI group. Overall, the univariate findings from Tables 2 and 3 provide preliminary evidence to support the prediction that internalizing the transfer of intangibles has greater advantages in the case of corrupt countries than in the case of clean countries. Next we examine this evidence further in a multivariate setting. 5.2. Multivariate tests' results In Table 4, we present our multivariate tests using OLS and random-effects regression models, which allow us to directly test our three hypotheses. We use three different valuation measures (TOBINQ1, TOBINQ2, and EXV) as the dependent variable in Panels A, B, and C, and estimate the regressions separately for high-and low-CI groups of firms. Starting with the first hypothesis, regressions in Table 4 show that intangibles' intensity (INT) is positively associated with firm value. However, the relationship of INT with VALUE is much stronger for low-corruption-index firms, based on all three measures of valuation. These results support the notion that the value of applicable intangibles is lower in a corrupt environment than in a clean environment, as expressed in Hypothesis 1. This can be explained by the fact that lack of property rights protection and information asymmetry problems are likely to be exacerbated in corrupt environments (Zhao (2006)). Hypothesis 2a is tested by the coefficient β2. On one hand, in the low-CI sample regressions, multinationality does not have a significant impact on value, which is consistent with the evidence of Morck and Yeung (1991). On the other hand, for the high-CI sample, the coefficient of the multinationality variable's coefficient, β2, is negative and significant, which implies that, without accounting for the effect of intangibles, MNCs experience a reduction in value when they expand into corrupt countries. These results support the notion that, in the absence of intangibles, an expansion into corrupt countries entails additional risks, which lead to lower valuation. However, consistent with Hypothesis 2b, for MNCs with intangibles, the overall impact of multinationality is positive. More specifically, the coefficients of β2 and β3 in OLS model (1) of Panel A, are −0.009 and 0.097, respectively. Thus, although the “pure” impact of multinationality in corrupt countries is −0.009, there is a strong positive impact from internalization (i.e. β3 is 0.097) so that the total impact of multinationality can be positive if a MNC possess high levels of intangibles. Based on these findings, the break-even intangibles intensity is at 9%. In other words, if a MNC spends more than 9% of sales revenues in R&D and advertising and expands its operations network in corrupt environments, then the total impact of multinationality on its value becomes positive.24 Note that the mean (median) value of intangibles' intensity for our sample of MNCs is 6% (3.5%). The coefficient of the interaction term between intangibles' intensity and multinationality is positive and significant for high-corruption-index firms, while it is negative and insignificant for low-corruption-index firms. In the fifth column, we report the difference between the coefficients of the two sub-samples regressions, and the corresponding t-statistics for the test of the null hypothesis that there is no difference in the coefficients. The difference in the coefficients of the interaction terms is statistically significant, in support of Hypothesis 3. This finding is consistent with the notion that gains from internalization of intangibles are substantially larger in the presence of corruption. The coefficients of the other control variables show the expected signs. Market valuation is negatively related to leverage and positively related to size. The coefficient of industry diversification (number of segments (NSEG)) is negative and significant, consistent with prior evidence documenting a diversification discount (see Lang and Stulz (1994), Comment and Jarrell (1995), and Berger and Ofek (1995), among others). As discussed earlier, there is a possibility that the OLS model coefficients are biased. Therefore, we report results of random-effects regressions in the third and fourth columns. These results are consistent with those obtained from the OLS regressions.25 Furthermore, the empirical results are not sensitive to alternative measurements of firm valuation, as evidenced by the similar results in Panels A, B, and C. This consistency in results using the different valuation measures holds in the following robustness tests as well. Therefore, to save space, we only report the results for the regressions using TOBINQ1 as a dependent variable (i.e., Panel A) in subsequent tables. The empirical findings above are graphically presented in Fig. 1. This figure illustrates the magnitude of the value impact of intangibles' intensity, multinationality, and internalization. The combined value impact is computed based on the coefficients of INT, MN and INT ⁎ MN obtained from estimating OLS models (1) (for high-CI MNCs) The combined value impact of multinationality is − 0.009 × MN + 0.097 × INT × MN = (− 0.009 + 0.097 × INT)MN = 0. Thus, the break-point of INT = 0.009/0.097 = 0.09 (or 9%). If INT is greater than 9%, the total value impact of multinationality, (− 0.009 + 0.097 × INT)MN, becomes positive. 25 We alternatively test our models based on year by year cross-sectional estimations, and find that regression results in every year are qualitatively similar to the ones in Table 4. 24

402

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table 4 Value impact of intangibles, multinationality, and internalization Expected signs

OLS

For [1] or [3]: highCI-ranked MNCs

[1] high-CIranked MNCs (i.e., MNCs with corrupt-countriesdominated network)

[2] low-CIranked MNCs (i.e., MNCs with cleancountries-dominated network)

[3] high-CIranked MNCs (i.e., MNCs with corruptcountries-dominated network)

[4] low-CIranked MNCs (i.e., MNCs with cleancountries-dominated network)

− 2.772⁎⁎⁎ (−2.81) 1.259 (1.11) − 0.009⁎⁎⁎ (−3.18) 0.097⁎⁎⁎ (3.94) − 2.230⁎⁎⁎ (−5.26) 0.243⁎⁎⁎ (5.57) 0.525 (0.69) − 0.177⁎⁎⁎ (−4.15) 0.716⁎⁎ (2.26) Yes Yes 592 28.48% 15.29⁎⁎⁎ [0.000]

− 2.861⁎⁎⁎ (− 2.81) 7.944⁎⁎⁎ (7.83) − 0.001 (− 0.22) − 0.035 (− 0.52) − 1.722⁎⁎⁎ (− 4.39) 0.201⁎⁎⁎ (4.11) 0.581⁎ (1.93) − 0.101⁎ (− 1.81) 0.022 (0.11) Yes Yes 592 21.07% 11.52⁎⁎⁎ [0.000]

−0.889 (− 0.53) 1.012 (0.73) −0.007 (− 1.48) 0.094⁎⁎ (2.50) −2.347⁎⁎⁎ (− 4.88) 0.156⁎⁎ (2.13) 0.513 (0.68) −0.090⁎⁎ (− 2.13) 0.507 (0.89) Yes Yes 592 27.47% 97.59⁎⁎⁎ [0.000]

− 1.865 (− 1.14) 6.105⁎⁎⁎ (6.28) 0.003 (0.32) − 0.087 (− 0.91) − 2.270⁎⁎⁎ (− 5.64) 0.150⁎⁎ (1.96) 0.128 (0.46) − 0.019 (− 0.33) − 0.116 (0.30) Yes Yes 592 20.95% 97.80⁎⁎⁎ [0.000]

0.089 (0.06) − 6.685⁎⁎⁎ (− 4.38) − 0.008 (− 1.18) 0.131⁎ (1.84) − 0.508 (− 0.88) 0.042 (0.65) − 0.056 (− 0.07) − 0.076 (− 1.08) 0.694⁎ (1.84)

− 2.474⁎⁎⁎ (−2.82) 0.304 (0.30) − 0.007⁎⁎⁎ (−2.82) 0.072⁎⁎⁎ (3.31) − 2.032⁎⁎⁎ (−5.39) 0.221⁎⁎⁎ (5.71) 0.719 (1.07) − 0.163⁎⁎⁎ (−4.30) 0.662⁎⁎⁎ (2.35) Yes Yes 592

− 2.530⁎⁎⁎ (− 2.82) 5.309⁎⁎⁎ (5.94) − 0.001 (− 0.10) − 0.037 (− 0.62) − 1.527⁎⁎⁎ (− 4.42) 0.182⁎⁎⁎ (4.24) 0.536⁎⁎ (2.02) − 0.094⁎ (− 1.91) 0.022 (0.12) Yes Yes 592

−0.674 (− 0.45) 0.257 (0.21) −0.005 (− 1.24) 0.069⁎⁎ (2.05) −2.143⁎⁎⁎ (− 5.09) 0.137⁎⁎ (2.10) 0.639 (0.97) −0.075⁎⁎ (− 2.01) 0.498 (0.98) Yes Yes 592

− 1.716 (− 1.20) 3.405⁎⁎⁎ (3.96) 0.003 (0.33) − 0.076 (− 0.91) − 1.981⁎⁎⁎ (− 5.57) 0.141⁎⁎ (2.11) 0.123 (0.51) − 0.022 (− 0.44) − 0.092 (− 0.28) Yes Yes 592

0.056 (0.05) − 5.005⁎⁎⁎ (− 3.71) − 0.006 (− 1.14) 0.109⁎ (1.72) − 0.505 (− 0.99) 0.039 (0.68) 0.184 (0.25) − 0.069 (− 1.11) 0.640⁎ (1.91)

For [2] or [4]: lowCI-ranked MNCs

Panel A: dependent variable is TOBINQ1 Intercept INT

0

+

MN

–

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F(or X2)] Panel B: dependent variable is TOBINQ2 Intercept INT 0

+

−

0

+

0

MN INT ⁎ MN LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N

Random effects

Test for coefficient differences: [1] − [2]

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

403

Table 4 (continued) Expected signs

OLS

For [1] or [3]: highCI-ranked MNCs

[1] high-CIranked MNCs (i.e., MNCs with corrupt-countriesdominated network)

[2] low-CIranked MNCs (i.e., MNCs with cleancountries-dominated network)

[3] high-CIranked MNCs (i.e., MNCs with corruptcountries-dominated network)

[4] low-CIranked MNCs (i.e., MNCs with cleancountries-dominated network)

25.95% 13.46⁎⁎⁎ [0.000]

19.03% 9.02⁎⁎⁎ [0.000]

24.67% 89.58⁎⁎⁎ [0.000]

16.66% 70.28⁎⁎⁎ [0.000]

–1.408⁎⁎⁎ (− 4.03) 0.542 (1.29) −0.0004 (− 0.40) 0.020⁎⁎ (2.28) −0.094 (− 0.57) 0.086⁎⁎⁎ (5.53) −0.091 (− 0.35) −0.054⁎⁎⁎ (− 3.57) 0.341⁎⁎⁎ (3.03) Yes Yes 556 23.12% 10.83⁎⁎⁎ [0.000]

– 1.224⁎⁎⁎ (−2.89) 2.641⁎⁎⁎ (6.49) 0.003 (1.10) − 0.029 (−1.08) − 0.011 (−0.07) 0.076⁎⁎⁎ (3.88) 0.196 (1.05) − 0.032 (−1.38) − 0.126 (−1.52) Yes Yes 554 13.83% 5.76⁎⁎⁎ [0.000]

−0.837 (− 1.57) 1.027⁎ (1.83) 0.001 (0.40) 0.014 (1.07) −0.123 (− 0.65) 0.058⁎⁎ (2.48) −0.073 (− 0.25) −0.038⁎⁎ (− 2.31) 0.306⁎ (1.74) Yes Yes 556 22.31% 62.35⁎⁎⁎ [0.000]

− 0.148 (− 0.23) 2.046⁎⁎⁎ (4.84) 0.003 (1.04) − 0.022 (− 0.60) − 0.148 (− 0.81) 0.023 (0.80) 0.062 (0.29) 0.007 (0.26) − 0.163 (− 1.18) Yes Yes 554 11.48% 41.43⁎⁎⁎ [0.000]

For [2] or [4]: lowCI-ranked MNCs

Panel B: dependent variable is TOBINQ2 R2 F (or X2) [Prob. N F(or X2)] Panel C: dependent variable is EXV Intercept INT

0

+

MN

–

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F(or X2)]

Random effects

Test for coefficient differences: [1] − [2]

− 0.184 (−0.34) − 2.099⁎⁎⁎ (−3.55) − 0.003 (−1.22) 0.049⁎ (1.82) − 0.083 (−0.35) 0.009 (0.38) − 0.287 (−0.86) − 0.022 (−0.80) 0.467⁎⁎⁎ (3.24)

This table reports coefficients for OLS and random effects regressions. Also reported are the corresponding t-statistics in parentheses for OLS and zstatistics for random effects model. The expected signs of the coefficients implied by the hypotheses outlined in Section 2 are shown in the first two columns. We estimate the regressions separately for the high-and low-CI-ranked MNC groups. The corruption index (CI) is computed as the sum over the weighted corruption scores as defined in Table 1. High-CI-ranked MNCs (i.e. HCI = 1) are firms ranked in the highest 1/2 of CI rankings in each year, while the remaining firms are classified as the low-CI-MNCs (i.e. HCI = 0). The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t 4MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t where VALUE is alternatively TOBINQ1, TOBINQ2, or EXV. TOBINQ1 is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)]/ (total assets) as defined in Chung and Pruitt (1994). TOBINQ2 is the adjusted version of Tobin's q by adding intangibles to total assets in the denominator of TOBINQ1. EXV is the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in Rhodes-Kropf, Robinson, and Viswanathan (2005). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and yearfixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-CI indicator variable, HCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

404 C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Fig. 1. Value impact of intangibles, multinationality, and internalization. This figure shows the magnitude of the value impact of intangibles' intensity (INT), multinationality (MN), and their combination (INT ⁎ MN, i.e. internalization). The vertical axis represents the magnitude of the value impact measured by substituting values for INT, MN and INT ⁎ MN into a model that uses the coefficients of INT, MN and INT ⁎ MN obtained from estimating OLS models (1) and (2) in Table 4. Tobin's q is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). Intangibles' intensity is the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

405

Table 5 Construction of an orthogonal measure of country-level corruption Dependent variable: corruption score (C) Intercept GDP per capita Legal origin — FRENCH Legal origin — GERMAN Legal origin — SCANDINAVIAN Legal origin — SOCIALIST N R2 F [Prob. N F]

8.208⁎⁎⁎ (63.90) − 0.0002⁎⁎⁎ (− 28.47) 0.487⁎⁎⁎ (3.68) 0.128 (0.51) − 1.721⁎⁎⁎ (− 5.76) 0.610⁎⁎⁎ (3.35) 368 78.55% 265.15⁎⁎⁎ [0.000]

This table shows the method used to obtain country-level corruption scores that are orthogonal to country-level poverty and legal origin and the correlation between the corruption scores and country level variables. The original corruption scores are regressed on GDP per capita and legal origins. Cj;t ¼ g0 þ g1 GDPj;t þ g2 FRENCHj;t þ g3 GERMANj;t þ g4 SCANDINAVIANj;t þ g5 SOCIALISTj;t þ gj;t Orthogonal corruption (OC) is the residual (ηj,t) obtained from the estimation of the above regression model. ηj,t is therefore the country-level corruption score after excluding any correlation with country-level poverty and with legal origin. Matching countries from two data sources, Dun and Bradstreet's Who owns Whom and Transparency International's annual CPI reports, results in a set of 92 countries with all necessary information available. Thus, the number of total observations is 368 over the four-year sample period. Cj,t is country j's corruption score in year t, which is computed by subtracting Transparency International's Corruption Perception Index (CPI) from 11. GDP per capita in US dollar (GDPj,t) is a proxy for the level of a country's poverty and is collected from three sources: International Monetary Fund, World Bank, and Penn World Table Version 6.1. Legal origins are divided into five groups and indicated by five dummies: BRITISH common law (BRITISH), FRENCH civil law (FRENCH), German civil law (GERMAN), SCANDINAVIAN civil law (SCANDINAVIAN), and SOCIALIST law (SOCIALIST). Legal origin data are obtained from the La Porta, Lopez-De-Silanes, Shleifer and Vishny (1997a, 1997b, 1998, 1999, 2000) studies. The orthogonal corruption scores are subsequently used to obtain new corruption index, which we use to classify MNCs into high-and low-corruption groups in the next table. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

and (2) (for low-CI MNCs) in Panel A. The left-hand side graph is for MNCs with networks concentrated in corrupt countries and the right-hand side graph is for MNCs with networks concentrated in clean countries. In general, the combined value impact shown by the height of the three-dimensional plane in the graph is higher for MNCs with clean environment-dominated networks. For MNCs with networks concentrated in corrupt countries, the slope of the plane gets steeper as the degree of multinationality increases. This pattern confirms the evidence that MNCs without intangibles experience decrease in value when they expand their geographic operations in corrupt countries. However, they benefit when they possess high level of intangibles and expand their network in corrupt countries. This is shown by the sharp increase in the slope of the plane when one moves along the multinationality axis. For the MNCs with networks in clean countries, the value impact decreases as the level of intangibles decreases regardless of the degree of multinationality. This illustrates our multivariate tests' finding that these MNCs display a positive value-impact of intangibles but a non-positive internalization effect. Therefore, the graphs clearly illustrate the different patterns of the internalization effect on value for MNCs that operate in clean environments and corrupt environments, respectively. The results presented thus far provide strong support for our hypotheses. In particular we show 1) that intangibles are worth more in clean countries than in corrupt countries, 2a) multinationality has a negative value impact when the MNCs do not possess intangibles but nevertheless choose to expand their network in corrupt countries, 2b) that MNCs add value through expanding in corrupt countries once they have sufficient amount of intangibles (i.e., a intangibles-tosales ratio of greater than 8%), and 3) that internalization benefits are enhanced in the presence of corruption. However, these findings can be attributed to the fact that the corruption measures we use are highly correlated with other country characteristics, such as poverty level and legal environment. We address this issue next. To ensure that our results can be attributed purely to the level of corruption in a country and not to other factors that are correlated with it, we devise an orthogonal country-level corruption score (OCj,t) and use it to create a new firm-specific corruption index. We then re-classify firms into high-and low orthogonal corruption index groups (high OCI and low OCI

406

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table 6 Value impact of intangibles, multinationality, and internalization: using the orthogonal corruption index (OCI) Expected signs

OLS

For [1] or [3]: highOCI-ranked MNCs

[1] high-OCI-ranked MNCs (i.e., MNCs with corrupt-countriesdominated network)

[2] low-OCI-ranked MNCs (i.e., MNCs with clean-countriesdominated network)

[3] high-OCI-ranked MNCs (i.e., MNCs with corrupt-countriesdominated network)

[4] low-OCI-ranked MNCs (i.e., MNCs with clean-countriesdominated network)

− 1.617⁎ (− 1.67) 0.074 (0.08) − 0.007⁎⁎ (− 2.54) 0.110⁎⁎⁎ (5.10) − 2.013⁎⁎⁎ (− 5.34) 0.194⁎⁎⁎ (5.01) 0.020 (0.03) − 0.160⁎⁎⁎ (− 3.92) 1.266⁎⁎⁎ (4.16) Yes

− 3.510⁎⁎⁎ (− 3.43) 9.491⁎⁎⁎ (8.74) − 0.002 (− 0.53) − 0.068 (− 1.27) − 1.791⁎⁎⁎ (− 4.35) 0.249⁎⁎⁎ (5.09) 0.606⁎ (1.87) − 0.126⁎⁎ (− 2.42) − 0.286 (− 1.34) Yes

−0.014 (− 0.01) −1.461 (− 1.25) −0.006 (− 1.49) 0.129⁎⁎⁎ (3.93) −1.794⁎⁎⁎ (− 4.15) 0.110⁎ (1.79) 0.358 (0.52) −0.077⁎ (− 1.82) 1.030⁎⁎ (2.06) Yes

− 2.307 (−1.50) 6.531⁎⁎⁎ (6.70) − 0.004 (−0.66) − 0.027 (−0.38) − 2.566⁎⁎⁎ (−6.37) 0.195⁎⁎⁎ (2.71) 0.219 (0.78) − 0.055 (−1.09) − 0.289 (−0.82) Yes

Yes

Yes

Yes

Yes

592 30.81% 17.10⁎⁎⁎ [0.000]

592 27.65% 14.68⁎⁎⁎ [0.000]

592 29.39% 98.97⁎⁎⁎ [0.000]

592 25.43% 124.93⁎⁎⁎ [0.000]

For [2] or [4]: lowOCI-ranked MNCs

Intercept INT

0

+

MN

–

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F (or X2)]

Random effects

Test for coefficient differences: [1] − [2]

1.893 (1.32) − 9.417⁎⁎⁎ (−6.35) − 0.004 (−0.83) 0.178⁎⁎⁎ (3.27) − 0.222 (−0.39) − 0.055 (−0.88) − 0.586 (−0.72) − 0.034 (−0.52) 1.553⁎⁎⁎ (3.94)

This table reports coefficients for OLS random effects regressions. Also reported are the corresponding t-statistics in parentheses for OLS and zstatistics for random effects model. Based on the hypotheses in Section 2, the expected signs of coefficients are shown in the first two columns. We conduct regressions for two groups, high-and low-OCI-ranked MNCs, where orthogonal corruption index (OCI) is computed as the sum over the weighted orthogonal corruption scores (OC), which are residuals from the regression estimated in Panel A of Table 4. High-OCI-ranked MNCs (i.e. HOCI = 1) are firms ranked in the highest 1/2 of OCI rankings in each year, while the remaining firms are classified as the low-OCI-MNCs (i.e. HOCI = 0). The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t 4MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t For the dependent variable (VALUE), we useTOBINQ1, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and year-fixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-OCI indicator variable, HOCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

respectively) and re-estimate our OLS and random-effects models. The results of the regression used to orthogonalize the country-level corruption scores are reported in Table 5. The results from re-estimating the multivariate regressions using the orthogonal corruption index as a basis for classification into high-and low-OCI groups are documented in Table 6.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

407

Combining corruption data from Transparency International and multinational network data from Dun and Bradstreet's Who Owns Whom provides 92 country observations in each year, i.e. a total of 362 observations spanning four years. The results in Table 5 show that the model's R-squared is over 78%, which indicates that poverty and legal origin provide strong explanatory power in the country's corruption level regression. As expected, GDP per capita is strongly negatively related to corruption (t-statistic of − 28.47), indicating that rich countries are less corrupt. This effect is also economically significant; $1000 more in GDP per capita is associated with a reduction of 0.2 in the corruption score. Following LLSV's studies, we use dummy variables for FRENCH, German, SCANDINAVIAN, and SOCIALIST legal environments and drop the BRITISH legal origin dummy (BRITISH) from the regression. Thus, a positive (negative) coefficient of a legal origin dummy variable implies that countries with the particular legal environment are more (less) corrupt relative to countries with legal environment based on the BRITISH common law system. We find that most civil law countries except for those that have adopted the SCANDINAVIAN civil law system are more corrupt than BRITISH common law countries. Indeed, SCANDINAVIAN countries (Denmark, Finland, Norway, and Sweden) are always top-ranked in the CPI of TI's surveys over the sample period (i.e. they display the lowest scores in corruption level).26 SOCIALIST countries are also more corrupt than BRITISH common law countries. We extract the residuals from this regression and define them as orthogonal corruption scores (OCj,t). In Table 6, we re-estimate the OLS and random-effects models for the high-and low-orthogonal-corruption-index groups. Similar to the findings in Table 4, the pure impact of intangibles, β1, is positive and significant at the 1% level for MNCs with high concentration in clean countries, but is not significant for MNCs with high concentration in corrupt countries. In addition, the findings show that expansion of the multinational network per se destroys value when MNCs without intangibles expand their foreign operations' network. However, this negative value impact becomes positive once intangibles' intensity reaches the 6% level (as compared to 9% in Table 4). The coefficient for the interaction term between multinationality and intangibles intensity is positive for high-OCI firms, consistent with prior results. The interaction term is slightly negative for low-OCI firms. The coefficients of interaction terms in the high-OCI regressions are economically as well as statistically stronger than the corresponding regressions in Table 4. Also, the difference in the interaction terms' coefficients between the high-and lowOCI regressions is even stronger than the corresponding differences shown in Table 4. Overall, the use of orthogonal corruption index measure has not altered our previous findings. Corruption is still found to be an important link between intangibles' intensity, multinationality, and firm valuation, in the spirit of the internalization theory. Next, we consider the possibility that the firm-level corruption index can also be endogenously associated with firmspecific characteristics. We use a two-stage least squares (2SLS) model involving the simultaneous estimation of the first-and second-stage models. When we estimate the high corruption index dummy (HCIi,t) in the first-stage model, we involve a probit model. In the second stage, fitted values of the high corruption index are interacted with the variables of interest. b Interaction terms with HCI i;t represent the coefficient differences between the high-and low-corruption index groups, similar b b to the ones reported in the last columns of Table 4. So, the interaction terms, HCI i;t ⁎ INT i,t ( HCI i;t ⁎ MN i) capture the difference in the value impact of intangibles (multinationality) between the high-and low-corruption index groups, b and the triple interaction term, HCI i;t * INT i,t * MN i, documents the difference in the magnitude of the interaction ⁎ terms, INTi,t MNi between the two groups.27 The results we obtain from the 2SLS model, shown in Table 7, confirm our previous results. Consistent with the findings based on the coefficient differences reported in the last column of Tables 4 and 6, the coefficient δ1 (δ2) representing the difference in value impact is negative. The coefficient of the triple interaction term is positive and statistically significant, consistent with the notion that the benefits of internalization of intangibles are greater for high-corruption-index MNCs than for low-corruptionindex MNCs. 28 Next, the third and fourth columns report the results of two-stage models using a continuous index variable (CIi,t) instead of a dummy variable (HCIi,t). The findings based on the use of the continuous variable are similar to those based on the dummy variable. Overall, as the models in Table 7 show, even after controlling for 26

The average of corruption scores of these four countries is 1.76, compared with the average of 6.3 in pooled sample. b However, we do not interact HCI i;t with other control variables such as leverage, size, capital expenditure, and number of segments because 1) their differences are not the primary focus of our paper, and 2) because using a smaller number of interaction variables reduces the multicollinearity problem. 28 We also accounted for the possibility that good performing MNCs (i.e. MNCs with high valuations) are more likely to have high levels of intangibles and, thus, have the “know-how” to expand in corrupt countries. In other words, high valuation leads to MNC networks, which are concentrated in corrupt countries. In this case, it is difficult to identify the cause and effect between corruption level in the foreign operations network and firm value. To control for the endogenous relation between MNC value (TOBINQ) and firm-level corruption index (HCI), we also estimated HCI by including Tobin’s q in the first stage regression. However, we find that the coefficient of TOBINQ in the first-stage model is insignificant (i.e. Tobin’s q does not explain HCI), and that the coefficients in the second-stage model are very similar to ones reported in Table 7. These results are available from the authors upon request. 27

408

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

the endogeneity between corruption index and other firm characteristics, the results are consistent with the OLS and random effects findings previously discussed. 5.3. Robustness tests' results Our first robustness check involves employing a different measure of corruption. In Table 8, we use the measure constructed by Kaufmann et al. (2004) and report the results obtained after re-estimating the regressions in Table 4. Consistent with our previous evidence in support of the internalization theory, the results in Table 8 indicate that internalization of intangibles provides greater value enhancement for MNCs with an extensive network in corrupt countries than for MNCs with a strong presence in clean countries. The second robustness check concerns the possibility that our results are driven by the inability to correctly estimate chop shop Tobin's qs for MNCs (see Yeung (2003)). The problem with the estimation of chop shop Tobin's qs arises because it is impossible to collect the information on country-level Tobin's q for each MNC network. This problem could be severe in the case of MNC networks that include countries experiencing large negative shocks. Therefore, we attempt to provide a partial solution to the problem by omitting from our sample MNCs operating in any country which

Table 7 Value impact of intangibles, multinationality, and internalization: controlling for endogeneity of firm's corruption index Using dummy HCI

Intercept INT MN INT ⁎ MN

Second-stage

First-stage

Second-stage

Dependent variable = HCI

Dependent variable = TOBINQ

Dependent variable = CI

Dependent variable = TOBINQ

− 1.105⁎⁎⁎ (−4.01) 2.376⁎⁎⁎ (3.64) 0.011⁎⁎⁎ (5.99)

−3.094⁎⁎⁎ (− 3.50) 11.244⁎⁎⁎ (3.31) 0.018 (1.21) −0.296⁎ (− 1.65) −1.929⁎⁎⁎ (− 5.74) 0.237⁎⁎⁎ (5.67) 0.646⁎⁎ (2.02) −0.086 (− 1.51) 0.273 (1.41) −1.715 (− 1.34) −6.468 (− 1.23) −0.020 (− 1.36) 0.391⁎⁎ (1.99)

3.144⁎⁎⁎ (17.07) 1.553⁎⁎⁎ (3.59) 0.005⁎⁎⁎ (5.22)

− 0.337 (− 0.14) 22.48⁎⁎ (2.01) 0.040 (1.30) − 0.868⁎⁎ (− 2.25) − 1.925⁎⁎⁎ (− 5.77) 0.229⁎⁎⁎ (5.71) 0.662⁎⁎ (2.09) − 0.136⁎⁎⁎ (− 3.42) 0.270 (1.40) − 0.860 (− 1.20) − 4.162 (− 1.38) − 0.010 (− 1.32) 0.222⁎⁎ (2.42)

LEV SIZE CAPXS NSEG

0.087⁎⁎ (2.47)

DEVE b HCI b ⁎ INT HCI b ⁎ MN HCI b ⁎ INT ⁎ MN HCI TFSALEP Industry dummies Year dummies N R2 (or pseudo R2) F (or X2) [Prob. N F (or X2)]

Using continuous variable CI

First-stage

0.378⁎⁎ (2.28) Yes Yes 1184 8.05% 132.08⁎⁎⁎ [0.000]

Yes Yes 1184 24.42% 19.80⁎⁎⁎ [0.000]

0.010 (0.45)

0.274⁎⁎ (2.38) Yes Yes 1184 7.25% 8.33⁎⁎⁎ [0.000]

Yes Yes 1184 24.41% 19.79⁎⁎⁎ [0.000]

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

409

had experienced a large shock during the sample period. In particular, in the computation of the corruption index (CI), we exclude firms with operations in any of the countries of the Asian Financial Crisis in 1997 (i.e. Hong Kong, Indonesia, Malaysia, Philippines, South Korea, and Thailand) or in Brazil, which experienced a crisis in 1998. The resulting subsample consists of 620 observations. The regression results are reported in Table 9. Consistent with the previous findings, the coefficients δ1 and δ2 on the last column are negative suggesting that intangibles and multinationality are less valuable when MNCs have operations network concentrated in corrupt countries. The positive and significant δ3 supports internalization theory. Thus far, our analyses are conducted based on two sub-samples classified based on high-or low-corruption-index. This classification provides a useful experimental environment for understanding internalization theory. We consider that the theory is also supported and explained by the degree of expertise in operating in corrupt countries. MNCs with prior experience in a particular country may be more likely to possess the know-how necessary to transform countryspecific corruption threats into opportunities. Thus, we test whether the benefit from internalizing intangibles' advantages is greater for firms that have prior experience with operations in corrupt countries versus firms that do not possess such experience. Due to the lack of detailed information regarding firms' experience in particular countries, we proxy MNCs' expertise by the depth of its operations in corrupt countries. Thus, we define an MNC as having corruption expertise if it has at least two subsidiaries in any one corrupt country. The rationale for this proxy is that having multiple subsidiaries in a corrupt country implies that the firm has already acquired the know-how necessary to deal with local corruption. As shown in Table 10, the regression models using this corruption expertise measure also support the internalization theory. Specifically, we find that the coefficient for the interaction between multinationality

Notes to Table 7 This table reports coefficients obtained from estimating a two-stage least squares (2SLS) regression model. Also reported are the corresponding z-statistics for the first-stage probit model and t-statistics for the second-stage model. The first step involves a probit model wherein high corruption b ) is used in the second-stage model to create interaction terms with all independent index dummy (HCI) is estimated. The predicted value ( HCI variables of the Tobin's q regression model. The coefficients of the interaction terms, δ, capture the differences in value impacts between high-and lowCI-ranked MNCs. Positive (negative) δ coefficients indicate that the effect on MNC value for high-CI-ranked MNC group is larger (smaller) than that for low-CI-ranked MNC group. The following two equations describe the first-and second-stage regression models. First-stage model: HCIi;t ¼ a0 þ a1 INTi;t þ a2 MNi þ a3 NSEGi;t þ a4 TFSALEPi;t þ a5 INDi þ a6 YEARt þ wi;t Second-stage model: HCIi;t þ d1 INTi;t ⁎b HCIi;t þ d2 MNi ⁎b HCIi;t VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ d0 b þ d3 INTi;t ⁎MNi ⁎b HCIi;t þ 1i;t Alternatively, First-stage model: CIi;t ¼ a0 þ a1 INTi;t þ a2 MNi þ a3 NSEGi;t þ a4 TFSALEPi;t þ a5 INDi þ a6 YEARt þ wi;t Second-stage model: CIi;t þ d1 INTi;t ⁎ b CIi;t þ d2 MNi ⁎ b CIi;t VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ d0 b þ d3 INTi;t ⁎MNi ⁎ b CIi;t þ 1i;t The variables examined are defined as follows. The corruption index, CI is computed as the sum over the weighted corruption scores as defined in Table 1. High-CI-ranked MNCs (i.e., HCI= 1) are firms ranked in the highest 1/2 of CI rankings in each year, while the remaining firms are classified as the low-CIMNCs (i.e., HCI = 0). For the dependent variable (VALUE) in the second-stage model, we useTOBINQ1, which is computed as [market value of common equity+ preferred stock liquidating value+ long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as a sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. Number of foreign subsidiaries (NFS) is used to measure the degree of multinationality (MN). NSEG is the number of business segments. TFSALEP is the foreign sales ratio, computed as total foreign sales as percentage of total firm sales. X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. Industry-(IND) and year (YEAR) indicator variables are included. To determine whether the relation between firm's value and corruption index is endogenous, we use the Hausman test which provides statistical significance in difference of coefficients between OLS and 2SLS. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

410

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table 8 Value impact of intangibles, multinationality, and internalization: using a corruption INDEX based on country-level corruption scores constructed by Kaufmann et al. (2004) Expected signs

OLS

INT

0

+

MN

–

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F(or X2)]

Test for coefficient differences: [1] − [2]

[2] low-KCI-ranked MNCs (i.e., MNCs with clean-countriesdominated network)

[3] high-KCIranked MNCs (i.e., MNCs with corrupt-countriesdominated network)

[4] low-KCIranked MNCs (i.e., MNCs with clean-countriesdominated network)

− 3.794⁎⁎⁎ (− 3.89) 1.574 (1.29) − 0.011⁎⁎⁎ (− 3.68) 0.097⁎⁎⁎ (3.73) − 2.529⁎⁎⁎ (− 5.63) 0.288⁎⁎⁎ (6.38) 0.270 (0.35) − 0.166⁎⁎⁎ (− 3.75) 0.960⁎⁎⁎ (2.88) Yes

− 1.300 (− 1.30) 8.214⁎⁎⁎ (8.71) 0.004 (0.68) − 0.082 (− 1.14) − 1.396⁎⁎⁎ (− 3.83) 0.121⁎⁎⁎ (2.60) 0.500⁎ (1.76) − 0.099⁎ (− 1.81) − 0.133 (− 0.69) Yes

− 1.312 (− 0.83) 0.185 (0.14) − 0.009⁎⁎ (− 2.00) 0.121⁎⁎⁎ (3.25) − 2.221⁎⁎⁎ (− 4.74) 0.162⁎⁎ (2.26) 0.346 (0.49) − 0.086⁎⁎ (− 2.11) 0.869 (1.51) Yes

−0.312 (− 0.20) 6.384⁎⁎⁎ (6.55) 0.006 (0.62) −0.105 (− 1.07) −2.113⁎⁎⁎ (− 5.36) 0.072 (1.01) 0.256 (0.92) 0.002 (0.04) −0.195 (− 0.61) Yes

Yes

Yes

Yes

Yes

592 28.25% 15.12⁎⁎⁎ [0.000]

592 24.56% 12.50⁎⁎⁎ [0.000]

592 26.89% 95.67⁎⁎⁎ [0.000]

592 22.40% 102.30⁎⁎⁎ [0.000]

For [1] or [3]: For [2] or [4]: [1] high-KCI-ranked high-KCIlow-KCIMNCs (i.e., MNCs ranked MNCs ranked MNCs with corrupt-countriesdominated network) Intercept

Random effects

− 2.495⁎ (− 1.77) − 6.640⁎⁎⁎ (− 4.35) − 0.015⁎⁎ (− 2.02) 0.178⁎⁎ (2.23) − 1.132⁎⁎ (− 1.97) 0.167⁎⁎ (2.54) − 0.226 (− 0.28) − 0.067 (− 0.94) 1.092⁎⁎⁎ (2.91)

This table reports coefficients obtained from estimating OLS and random effects regressions models. Also reported are the corresponding t-statistics in parentheses for OLS and z-statistics for random effects model. The first two columns show the expected signs of the coefficients as predicted by the hypotheses outlined in Section 2. We estimate the regressions separately for the high-and low-KCI-ranked MNC groups, where the Kaufmann et al. corruption index (KCI) is computed as the sum over the weighted Kaufmann et al. (2004) ' measure of country corruption. High-KCI-ranked MNCs (i.e. HKCI = 1) are firms ranked in the highest 1/2 of KCI rankings in each year, while the remaining firms are classified as the low-KCI-MNCs (i.e. HKCI = 0). The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t For the dependent variable (VALUE), we useTOBINQ1, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and year-fixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-KCI indicator variable, HKCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

and intangibles intensity is significant only for the MNCs that have expertise in corrupt countries. Thus, the benefit of internalization is great for MNCs with expertise in operating in corrupt countries. In contrast, the value of internalization for MNCs that lack the expertise to deal with corruption is insignificant.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

411

Table 9 Value impact of intangibles, multinationality, and internalization: after excluding MNCs with operations in locations with big negative shocks Expected signs

OLS

For [1] or [3]: For [2] or [4]: [1] high-CI-ranked high-CIlow-CI-ranked MNCs (i.e., MNCs ranked MNCs MNCs with corruptcountries-dominated network) Intercept INT 0

+

−

0

+

0

MN INT ⁎ MN LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2 [Prob. N F(or X2

Random effects

Test for coefficient differences: [1]− [2]

[2] low-CI-ranked MNCs (i.e., MNCs with cleancountries-dominated network)

[3] high-CI-ranked MNCs (i.e., MNCs with corruptcountries-dominated network)

[4] low-CIranked MNCs (i.e., MNCs with clean-countriesdominated network)

− 9.609⁎⁎⁎ (− 4.59) 1.035 (0.47) − 0.057⁎⁎⁎ (− 2.85) 0.789⁎⁎⁎ (3.24) −2.353⁎⁎⁎ (− 3.60) 0.557⁎⁎⁎ (6.69) − 0.582 (− 0.44) − 0.241⁎⁎ (− 2.11) 0.380 (0.74) Yes

− 1.209 (− 1.09) 8.584⁎⁎⁎ (8.47) − 0.015 (− 1.35) − 0.025 (− 0.17) − 1.305⁎⁎⁎ (− 3.35) 0.111⁎⁎ (2.03) 0.627⁎⁎⁎ (2.43) − 0.016 (− 0.25) − 0.032 (− 0.17) Yes

− 9.092⁎⁎⁎ (− 2.83) 0.230 (0.10) − 0.036 (− 1.24) 0.350 (1.45) − 2.609⁎⁎⁎ (− 3.70) 0.499⁎⁎⁎ (3.86) − 0.346 (− 0.30) − 0.038 (− 0.39) 0.193 (0.23) Yes

− 1.147 (−0.66) 6.231⁎⁎⁎ (6.04) − 0.018 (−1.07) 0.020 (0.09) − 1.873⁎⁎⁎ (−4.36) 0.113 (1.35) 0.221 (0.92) − 0.014 (−0.20) − 0.070 (−0.22) Yes

Yes

Yes

Yes

Yes

310 31.57% 9.04⁎⁎⁎ [0.000]

310 35.74% 10.90⁎⁎⁎ [0.000]

310 28.81% 42.38⁎⁎⁎ [0.000]

310 32.59% 96.83⁎⁎⁎ [0.000]

−8.400⁎⁎⁎ (− 3.67) −7.549⁎⁎⁎ (− 3.67) −0.042⁎ (− 1.86) 0.814⁎⁎⁎ (2.89) −1.048 (− 1.39) 0.447⁎⁎⁎ (4.44) −1.209 (− 1.05) −0.225⁎ (− 1.76) 0.412 (0.83)

This table reports the coefficients obtained from estimating OLS and random effects regressions. Also reported are the corresponding t-statistics in parentheses for OLS and z-statistics for random effects model. The first two columns show the expected signs of the coefficients as predicted by the hypotheses outlined in Section 2. We exclude all MNCs with subsidiaries located in any country where a financial crisis occurred during the sample period, 1995–1998. We identify eight such countries: Hong Kong, Indonesia, Malaysia, Philippines, South Korea, and Thailand which experienced the Asian Financial Crisis in 1997, and Brazil which experienced a Real crisis in 1998. Our sample after is reduced to 620 MNC-year observations from the 1184 MNC-year observations used in the multivariate tests in Tables 4, 6, 7 and 8. We estimate the regressions separately for the high-and low-CI MNC groups. The corruption index (CI) is computed as the sum over the weighted corruption scores as defined in Table 1. High-CI-ranked MNCs (i.e. HCI = 1) are firms ranked in the highest 1/2 of CI rankings in each year, while the remaining firms are classified as the low-CI-MNCs (i.e. HCI = 0). The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t For the dependent variable (VALUE), we useTOBINQ1, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and year-fixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-CI indicator variable, HCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

412

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table 10 Value impact of intangibles, multinationality, and internalization: based on expertise of mncs in operating in corrupt countries Expected signs

For [1] or [3]: MNCs with corruption expertise

OLS

For [2] or [4]: MNCs without corruption expertise

Intercept INT

0

+

MN

−

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F(or X2)]

Random effects

Test for coefficient differences: [1] − [2]

[1] MNCs with corruption expertise

[2] MNCs without corruption expertise

[3] MNCs with corruption expertise

[4] MNCs without corruption expertise

0.082 (0.07) − 0.298 (− 0.16) − 0.006⁎⁎ (− 2.44) 0.108⁎⁎⁎ (4.13) − 1.855⁎⁎⁎ (− 3.18) 0.097 (1.55) 1.292 (0.97) − 0.090⁎⁎ (− 2.16) 0.881⁎ (1.76) Yes Yes 295 46.14% 15.93⁎⁎⁎ [0.000]

−4.105⁎⁎⁎ (− 4.51) 5.951⁎⁎⁎ (6.44) 0.001 (0.19) −0.086 (− 1.15) −1.998⁎⁎⁎ (− 6.09) 0.266⁎⁎⁎ (6.32) 0.587⁎⁎ (1.97) −0.145⁎⁎⁎ (− 3.06) 0.279 (1.51) Yes Yes 889 20.38% 14.90⁎⁎⁎ [0.000]

1.687 (0.92) 0.505 (0.20) − 0.004 (− 1.04) 0.093⁎⁎⁎ (2.40) − 2.068⁎⁎⁎ (− 3.11) 0.009 (0.10) 2.097 (1.39) − 0.081⁎ (− 1.85) 1.139 (1.42) Yes Yes 295 45.55% 118.10⁎⁎⁎ [0.000]

−2.811⁎⁎ (− 2.03) 4.376⁎⁎⁎ (4.93) −0.001 (− 0.08) −0.002 (− 0.03) −2.294⁎⁎⁎ (− 6.81) 0.195⁎⁎⁎ (3.13) 0.213 (0.79) −0.022 (− 0.48) 0.234 (0.73) Yes Yes 889 18.99% 116.27⁎⁎⁎ [0.000]

4.188⁎⁎ (2.26) − 6.249⁎⁎ (− 2.46) − 0.008 (− 1.08) 0.194⁎⁎ (2.48) 0.142 (0.17) − 0.169⁎ (− 1.86) 0.705 (0.40) 0.055 (0.78) 0.602 (0.88)

This table reports the coefficients obtained from estimating OLS and random effects regressions. Also reported are the corresponding t-statistics in parentheses for OLS and z-statistics for random effects model. We estimate the regressions separately for two groups of MNCs: MNCs with corruption expertise and MNCs without corruption expertise. MNC is regarded as having expertise in dealing with corruption if it has at least two subsidiaries in any corrupt country. Otherwise, it is classified as MNC without corruption expertise. The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t For the dependent variable (VALUE), we use TOBINQ1, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and year-fixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-CI indicator variable, HCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

Last, we analyze a different time period to see whether or not our results hold more generally or whether they are specific to the 1995–1998 time period. This concern is particularly driven by the earlier-mentioned observation that Tobin's q ratios are trending up during our 1995-1998 sample period. We collected additional data from the 2006/ 2007 Dun and Bradstreet's issue of Who Owns Whom, which provides foreign affiliate information for the years 2004/2005. Using the same methodology described in the sample selection section, we obtain 312 MNCs and 1248

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

413

Table 11 Value impact of intangibles, multinationality, and internalization: testing a different time period Expected signs

OLS

For [1] or [3]: For [2] or [4]: [1] high-CI-ranked high-CI-ranked low-CI-ranked MNCs (i.e., MNCs MNCs MNCs with corruptcountries-dominated network) Intercept INT

0

+

MN

−

0

INT ⁎ MN

+

0

LEV SIZE CAPXS NSEG DEVE Industry dummies Year dummies N R2 F (or X2) [Prob. N F(or X2)]

Random effects

Test for coefficient differences: [1]− [2]

[2] low-CI-ranked MNCs (i.e., MNCs with cleancountries-dominated network)

[3] high-CI-ranked MNCs (i.e., MNCs with corruptcountries-dominated network)

[4] low-CI-ranked MNCs (i.e., MNCs with clean-countriesdominated network)

− 2.591⁎⁎⁎ (− 3.54) − 0.457 (− 1.48) − 0.007⁎⁎ (− 2.49) 0.097⁎⁎⁎ (2.87) − 1.108⁎⁎⁎ (− 3.91) 0.200⁎⁎⁎ (5.48) 0.309 (0.51) − 0.049⁎ (− 1.94) 0.068 (0.43) Yes

− 2.327⁎⁎ (−2.30) − 0.140 (−0.21) − 0.009 (−1.56) 0.029 (0.76) − 1.885⁎⁎⁎ (−4.69) 0.184⁎⁎⁎ (3.86) 1.085 (1.09) − 0.088⁎⁎ (−2.25) 0.341 (1.11) Yes

−1.541 (− 1.42) −0.008 (− 0.02) −0.005 (− 1.16) 0.079⁎ (1.67) −0.767⁎⁎ (− 2.45) 0.143⁎⁎⁎ (2.63) −0.067 (− 0.09) −0.031 (− 0.96) 0.112 (0.43) Yes

− 1.049 (− 0.66) − 0.560 (− 0.65) − 0.007 (− 0.79) 0.023 (0.42) − 1.442⁎⁎⁎ (− 3.31) 0.121⁎ (1.65) 0.783 (0.91) − 0.058 (− 1.17) 0.333 (0.63) Yes

Yes 624 15.38% 7.37⁎⁎⁎ [0.000]

Yes 624 15.43% 7.39⁎⁎⁎ [0.000]

Yes 624 14.64% 48.08⁎⁎⁎ [0.000]

Yes 624 14.83% 46.23⁎⁎⁎ [0.000]

−0.264 (− 0.21) −0.316 (− 0.45) 0.002 (0.37) 0.068 (1.29) 0.777 (1.58) 0.016 (0.26) −0.775 (− 0.68) 0.039 (0.85) −0.272 (− 0.83)

This table reports coefficients for OLS random effects regressions. Also reported are the corresponding t-statistics in parentheses for OLS and z-statistics for random effects model. Based on the hypotheses in Section 2, the expected signs of coefficients are shown in the first two columns. We estimate the regressions using data covering the 2002–2005 period. The model estimated in this table is: VALUEi;t ¼ b0 þ b1 INTi;t þ b2 MNi þ b3 INTi;t ⁎MNi þ b4 Xi;t þ b5 INDi þ b6 YEARt þ ei;t For the dependent variable (VALUE), we useTOBINQ1, which is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets) as defined in Chung and Pruitt (1994). INT is intangibles' intensity and computed as the sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. The degree of multinationality (MN) is measured by the number of foreign subsidiaries (NFS). X is a vector of firm characteristic variables consisting of 1) LEV: financial leverage, measured as long-term debt/total assets, 2) SIZE: natural log of total assets, 3) CAPXS: ratio of capital expenditures to sales, 4) NSEG: number of business segments, and 5) DEVE: degree of involvement in developing countries, measured as the number of developing countries where subsidiaries locate divided by the total number of countries. We include industry and year indicator variables to control for the industry-and year-fixed effects. The difference in the coefficients between the two groups of regressions is captured by including all MNCs and using interaction terms between the high-CI indicator variable, HCIi,t, and each independent variable. The tests for the difference in the coefficients between the two groups' regressions are presented in the last column. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at the 10%, 5% and 1% level, respectively.

firm-year observations during the period 2002–2005. Table 11 shows that the results obtained from the tests performed over this more recent time period are qualitatively similar to the ones reported in Table 4, which uses the original time period, 1995–1998.29 29

The results shown in Table 10 are similar to those we obtained when we used TOBINQ2 and EXV as dependent variables. These results are not reported here for the sake of brevity.

414

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Overall, the findings from several robustness tests suggest that our empirical results are not sensitive to the type of corruption measure used, to the selection of countries with financial crisis, to the method of classifying the degree of firms' involvement in corrupt areas, or to a specific time period. 6. Conclusions According to internalization theory, MNCs' gain market power from their ability to arrange intra-firm cross-border transfers of knowledge-based intangible assets such as technological know-how and marketing expertise, rather than relying exclusively on external markets that constitute a more expensive mode of transaction. This paper provides new evidence in support of the internalization theory by examining the effect of foreign country corruption on the benefits MNCs derive from internalizing intangible assets' advantages. First, we find that the value impact of intangibles is lower for MNCs that primarily operate in corrupt countries than for MNCs that primarily operate in clean countries. Second, we show that expanding a corrupt countries-dominated foreign operations network in the absence of intangibles has a negative impact on market value. On the other hand, the value impact of intangibles becomes positive if MNCs possess a certain level of intangibles. Finally, we also present evidence that the value impact from internalization of intangibles is greater for MNCs operating primarily in corrupt countries than for MNCs operating primarily in clean countries. That is, MNC geographic diversification into more corrupt rather than less corrupt countries, coupled with the presence of intangibles, has an economically as well as statistically more positive effect on firm valuation. The last finding is consistent with the notion that, assuming transaction costs are higher in corrupt markets, the advantage of transferring intangibles by internalizing markets is greater in corrupt rather than in clean markets. In our multivariate tests, we start our analysis by conducting OLS and panel data regressions. We consider the potential problems in the statistical methods. We account for the potential endogeneity problem in the corruption measure at the country-level as well as in the corruption index value at the firm-level. To remedy the correlation between country-level corruption and other variables such as poverty and legal environments, we construct a countrylevel corruption score that is orthogonal to poverty level and legal environment variables. At the firm level, the endogeneity concern is addressed via the utilization of a 2SLS model. In the subsequent robustness tests, we use an alternative corruption measurement to examine whether our results are sensitive to our corruption classification. Our tests also control for MNCs with operations networks in countries experiencing big negative shocks. As a further robustness test, we re-classify firms based on a proxy of firms' expertise with corruption at the country level. Finally, we re-estimate the regression models using a different time period. Our findings remain consistent when we conduct all robustness tests. Therefore, these findings provide strong evidence in support of the importance of corruption in the link between intangibles, multinationality and valuation, as implied by the internalization theory. Appendix A Table A1 Variable descriptions Panel A: country-level corruption scores C

OC

KC

Corruption. This represents the country-level corruption score. Since 1995, the Transparency International has ranked countries in terms of Corruption Perceptions Index (CPI), which ranges between 0 (highly corrupt) and 10 (least corrupt), representing the degree of a country's “cleanness”. Therefore, C = 11 − CPI. Orthogonal corruption. This corruption indicator is computed as residual (ηj,t) from the regression, Cj;t ¼ g0 þ g1 GDPj;t þ g2 FRENCHj;t þ g3 GERMANj;t þ g4 SCANDINAVIANj;t þ g5 SOCIALISTj;t þ gj;t , where Cj,t is country j's corruption score in year t, which is computed by subtracting Transparency International's Corruption Perception Index (CPI) from 11. GDPj,t is GDP per capita in US dollar and used as a proxy for the level of country's poverty. We collect GDP data from three sources: International Monetary Fund, World Bank, and Penn World Table Version 6.1. Legal origins are classified into five groups and indicated by five dummies: BRITISH common law (BRITISH), FRENCH civil law (FRENCH), German civil law (GERMAN), SCANDINAVIAN civil law (SCANDINAVIAN), and SOCIALIST law (SOCIALIST). In the regression, the BRITISH common law is based and the other legal origin dummy variables are controlled in the regression. Legal origin data is employed from La Porta, Lopez-De-Silanes, Shleifer and Vishny (1997a, 1997b, 1998, 1999, 2000). Kaufmann et al. corruption. KC is a corruption indicator based on Kaufmann, Kraay, and Mastruzzi (2004). They define corruption as the exercise of public power for private gain and provide “Control of Corruption,” which measures perceptions of corruption. KC = 3 − Control of corruption.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

415

Table A1 (continued ) Panel B: firm-level corruption index CI

OCI

KCI

HCI HOCI HKCI b HCI

 P P P Corruption index. CIi;t ¼ j Wi;j Cj;t ¼ j NFSi;j I j NFSi;j 4Cj;t , where Cj,t is country j's corruption score in year t, which is computed by subtracting Transparency International's Corruption Perception Index (CPI) from 11, and Wi,j is MNC i's weight for country j and is computed as the ratio of foreign subsidiaries located in country j divided by the total number of foreign subsidiaries of MNC i.  P P P Orthogonal corruption index. OCIi;t ¼ j Wi;j OCj;t ¼ j NFSi;j = j NFSi;j 4OCj;t , where OCj,t is the orthogonal corruption measure, and Wi,j is MNC i's weight for country j and is computed as the ratio of foreign subsidiaries located in country j divided by the total number of foreign subsidiaries of MNC i.   P P P Kaufmann et al. corruption index. KCIi;t ¼ j Wi;j KCj;t ¼ j NFSi;j = j NFSi;j 4KCj;t , where KCj,t is Kaufmann et al. corruption, and Wi,j is MNC i's weight for country j and is computed as the ratio of foreign subsidiaries located in country j divided by the total number of foreign subsidiaries of MNC i. High-corruption-index dummy. HCIi,t, = 1 if a MNC is ranked in the highest 1/2 rankings of CIi,t in year t and HCIi,t, = 0 if it is ranked in the lowest 1/2 rankings, where CIj,t is corruption index. High-orthogonal corruption-index dummy. HOCIi,t, = 1 if a MNC is ranked in the highest 1/2 rankings of OCIi,t in year t and HOCIi,t, = 0 if it is ranked in the lowest 1/2 rankings, where OCIj,t is orthogonal corruption index. High-Kaufmann et al. corruption-index dummy. HKCIi,t, = 1 if a MNC is ranked in the highest 1/2 rankings of KCIi,t in year t and HKCIi,t, = 0 if it is ranked in the lowest 1/2 rankings, where KCIj,t is Kaufmann et al. corruption index. Predicted HCI. Using probit model, we estimate HCIi,t, to use it in the second-stage of 2SLS. HCIi;t ¼ a0 þ a1 INTi;t þ a2 MNi þ a3 NSEGi;t þ a4 TFSALEPi;t þ a5 YEARt þ a6 INDi;t þ wi;t , where HCI is Highcorruption-index dummy. INT is intangibles' intensity and computed as a sum of RD and AD, where RD and AD are R&D expenditures and advertising expenditures, respectively, scaled by sales. Number of foreign subsidiaries (NFS) is used to measure the degree of multinationality (MN). NSEG is the number of business segments. TFSALEP is the foreign sales ratio, computed as total foreign sales as percentage of total firm sales.

Panel C: country-level variables GDP

GDP per capita. This is used as a proxy for the level of country's poverty and is collected from three sources: International Monetary Fund, World Bank, and Penn World Table Version 6.1. BRITISH BRITISH common law environment. BRITISH = 1 if a country's legal environment is BRITISH common law and BRITISH = 0 otherwise. FRENCH FRENCH civil law environment. FRENCH = 1 if a country's legal environment is FRENCH civil law and FRENCH = 0 otherwise. GERMAN German civil law environment. GERMAN = 1 if a country's legal environment is German civil law and GERMAN = 0 otherwise. SCANDINAVIAN SCANDINAVIAN civil law environment. SCANDINAVIAN = 1 if a country's legal environment is SCANDINAVIAN civil law and SCANDINAVIAN = 0 otherwise. SOCIALIST SOCIALIST law environment. SOCIALIST = 1 if a country's legal environment is SOCIALIST civil law and SOCIALIST = 0 otherwise. Panel D: Firm characteristics and other variables TOBINQ1 TOBINQ2 EXV

RD AD INT LEV SIZE CAPXS DIVERS

Based on Chung and Pruitt (1994), Tobin's q is computed as [market value of common equity + preferred stock liquidating value + long-term debt − (short-term assets − short-term liabilities)] / (total assets). Adjusted Tobin's q. TOBINQ2 = [market value of common equity + preferred stock liquidating value + long-term debt − (shortterm assets − short-term liabilities)] / (total assets + R&D expenditures + advertising expenditures). Excess value. EXV is the value of the firm-specific component of the difference between market value and fundamental value, based on the procedure outlined in Rhodes-Kropf, Robinson, and Viswanathan (2005). Fundamental value, V is estimated by decomposing the market-to-book into two components: a measure of price to fundamentals (ln(M/V)), and a measure of fundamentals to book value (ln(V/B)). The first component captures the part of book-to-market associated with mispricing. This component is further decomposed into firm-specific and industry-specific mispricing. In our tests, we use the firm-specific mispricing component based on Model III of Rhodes-Kropf et al. (2005) that also accounts for net income and leverage effects. ln(Mi,t) = α0j,t + α1j,t ln(Bi,t) + α2j,t ln(NI)+i,t + α3j,t I(b 0)ln(NI)+i,t + α4j,t ln(LEVi,t) + ζi,t, where M is firm value, B is book value, NI+is absolute value of net income, I(b 0)ln(NI)+is an indicator function for negative net income observations, and LEV is the leverage ratio. Research and development intensity. RD = research and development (R&D) expenditures/sales. Advertising intensity. AD = advertising expenditures / sales. Intangibles' intensity. INT = research and development expenditures / sales + advertising expenditures / sales. Measure of financial leverage. LEV = long-term debt / total assets. SIZE = ln(total assets). CAPXS = capital expenditures / sales. Dummy for industrial diversification. DIVERS = 1 it the firm reports two or more business segments and DIVERS = 0 if it reports only one segment. (continued on next page)

416

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

Table A1 (continued ) Panel D: Firm characteristics and other variables NSEG Number of business segments. Number of 4-digit SIC industries. TFSALEP Foreign sales ratio. TFSALEP = total foreign sales / total firm sales. DEVE The degree of involvement in developing countries. DEVE = the number of developing countries where foreign subsidiaries locate / the total number of countries (NFC). NFS Number of foreign (i.e., non-US) subsidiaries. NFC Number of foreign (i.e., non-US) countries where subsidiaries locate. NFR Number of foreign (i.e. non-US) regions where subsidiaries locate. The subsidiaries are assigned to eight major geographic regions, depending on the location of each subsidiary's host country. The eight geographic regions are 1) the North American Free Trade Association, NAFTA, area that includes the Canada and Mexico, 2) the central and south America, 3) the western Europe, 4) the eastern Europe, 5) the advanced Asia countries which consist of Hong Kong, Japan, South Korea, Singapore, and Taiwan, 6) the remaining Asian countries, 7) Oceania area including Australia and New Zealand, and 8) Africa. YEAR Year dummies. YEAR95–YEAR98. IND 5 industry dummies. IND1–IND5. Fama and FRENCH's 5 industries classification is used. The detail information about the classification is available at Kenneth R. FRENCH's website, http://mba.tuck.dartmouth.edu/pages/faculty/ken.FRENCH/ data_library.html.

References Acemoglu, D., Zilibotti, F., 1999. Information accumulation in development. Journal of Economic Growth 4, 5–38. Ang, A., Hodrick, R., Xing, Y., Zhang, X., 2006. The cross-section of volatility and expected returns. Journal of Finance 61, 259–299. Ang, A., Hodrick, R., Xing, Y., Zhang, X., 2007. High idiosyncratic volatility and low returns: International and further U.S. evidence, Working Paper, Columbia University. Bali, T., Cakici, N., Yan , X., Zhang, Z., 2005. Does idiosyncratic volatility really matter? Journal of Finance 60, 905–929. Barberis, N., Huang, M., 2001. Mental accounting, loss aversion and individual stock returns. Journal of Finance 56, 1247–1292. Baumol, W., 1990. Entrepreneurship: productive, unproductive, and destructive. Journal of Political Economy 98, 893–921. Beck, T., Demirguc-Kunt, A., Maksimovic, V., 2005. Financial and legal constraints to growth: does firm size matter? Journal of Finance 60, 137–177. Benham, A., Benham, L., 2004. The costs of exchange: An approach to measuring transactions costs. Working paper, The Ronald Coase Institute and Washington University in St. Louis. Bliss, C., Di Tella, R., 1997. Does competition kill corruption? Journal of Political Economy 105, 1001–1023. Borensztein, E., DeGregorio, J., Lee, J., 1998. How does foreign direct investment affect economic growth? Journal of International Economics 45, 115–135. Begovic, B., 2005. Corruption: concepts, types, causes, and consequences. The Center for International Private Enterprise (U.S. Chamber of Commerce). Berger, P.G., Ofek, E., 1995. Diversification's effect on firm value. Journal of Financial Economics 37, 39–65. Buckley, P., Casson, M., 1976. The future of the multinational enterprise. Holmes & Meier, London. Casson, M., 1979. Alternatives to the multinational enterprise. McMillan, London. Chung, K.H., Pruitt, S.W., 1994. A simple approximation of Tobin's q. Financial Management 23, 70–74. Comment, R., Jarrell, G.A., 1995. Corporate focus and stock returns. Journal of Financial Economics 37, 67–87. Desai, M.A., Foley, C.F., Hines, J.R., 2006. Capital controls, liberalizations, and foreign direct investment. Review of Financial Studies 19, 1433–1464. De Soto, H., 1989. The other path: The invisible revolution in the third world. Harper and Row, New York. Ehrlich, I., Liu, F., 1999. Bureaucratic corruption and endogenous economic growth. Journal of Political Economy 107, 270–293. Friedman, E., Johnson, S., Kaufmann, D., Zoido-Lobaton, P., 2000. Dodging the grabbing hand: the determinants of unofficial activity in 69 countries. Journal of Public Economics 76, 459–493. Goyal, A., Santa-Clara, P., 2003. Idiosyncratic risk matters! Journal of Finance 58, 975–1007. Greene, W.H., 1993. Econometric analysis. Macmillan. Harris, D., Morck, R., Slemrod, J., Yeung, B., 1993. Income shifting in U.S. multinational corporations. Working paper No 3924. NBER. Henisz, W.J., 2000. The institutional environment for multinational investment. Journal of Law, Economics, and Organization 16, 334–364. Hennart, J.-F., 1982. A theory of multinational enterprise. University of Michigan Press, Ann Arbor. Hines, J.R., 1995. Forbidden payment: Foreign bribery and American business after 1977. Working paper No 5266. NBER. Huntington, S.P., 1968. Political order in changing societies. Yale University Press, New Haven. Johnson, S., McMillan, J., Woodruff, C., 2002. Property rights and finance. American Economic Review 92, 1335–1356. Kaufmann, D., Kraay, A., Zoido-Lobaton, P., 1999a. Governance matters. Working paper. The World Bank. Kaufmann, D., Kraay, A., Zoido-Lobaton, P., 1999b. Aggregating governance indicators. Working paper. The World Bank. Kaufmann, D., Kraay, A., Mastruzzi, M., 2004. Governance matters III: governance indicators for 1996, 1998, 2000, and 2002. The World Bank Economic Review 18, 253–287. Knack, S., Keefer, P., 1997. Does social capital have an economic payoff? A cross-country investigation. Quarterly Journal of Economics 112, 1251–1288.

C. Pantzalis et al. / Journal of Empirical Finance 15 (2008) 387–417

417

Lang, L.H.P., Stulz, R.M., 1994. Tobin's q corporate diversification and firm performance. Journal of Political Economy 102, 1248–1280. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997a. Legal determinants of external finance. Journal of Finance 52, 1131–1150. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997b. Trust in large organizations. American Economic Review 87, 333–338. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1998. Law and finance. Journal of Political Economy 106, 1113–1155. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1999. Investor protection and corporate valuation. Journal of Financial Economics 58, 3–27. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 2000. Agency problems and dividend policies around the world. Journal of Finance 55, 1–33. Leff, N., 1964. Economic development through bureaucratic corruption. American Behavioral Scientist 8, 8–14. Levine, R., Zervos, S., 1998. Stock markets, banks, and economic growth. American Economic Review 88, 537–558. Levy, H., 1978. Equilibrium in an imperfect market: a constraint on the number of securities in a portfolio. American Economic Review 68, 643–658. Li, H., Xu, L.C., Zou, H.-F., 2000. Corruption, income distribution, and growth. Economics and Politics 12, 155–182. Mauro, P., 1995. Corruption and growth. Quarterly Journal of Economics 110, 681–712. Merton, R., 1987. Presidential address: a simple model of capital market equilibrium with incomplete information. Journal of Finance 42, 483–510. Morck, R., Yeung, B., 1991. Why investors value multinationality. Journal of Business 64, 165–187. Morck, R., Yeung, B., 1992. Internalization: an event study test. Journal of International Economics 33, 41–46. Murphy, K.M., Shleifer, A., Vishny, R.W., 1991. The allocation of talent: implication for growth. Quarterly Journal of Economics 106, 503–530. Murphy, K.M., Shleifer, A., Vishny, R.W., 1993. Why is rent-seeking so costly to growth? American Economic Review 83, 409–414. Rhodes-Kropf, M., Robinson, D.T., Viswanathan, S., 2005. Valuation waves and merger activity: the empirical evidence. Journal of Financial Economics 77, 561–603. Rose-Ackerman, S., 1978. Corruption: A study in political economy. Academic Press, New York. Rugman, A., 1981. A test of internalization theory. Managerial and Decision Economics 2, 211–220. Shleifer, A., Vishny, R., 1993. Corruption. Quarterly Journal of Economics 108, 599–617. Smarzynska, B.K., 2004. The composition of foreign direct investment and protection of intellectual property rights: evidence from transition economies. European Economic Review 48, 39–62. Smarzynska, B.K., Wei, S.-J., 2002. Corruption and cross-border investment: Firm-level evidence. Working paper, William Davidson Institute at the University of Michigan Business School. Spiegel, M., Wang, X., 2005. Cross-sectional variation in stock returns: liquidity and idiosyncratic risk. Working paper. Yale University. Wei, S.-J., 2000. How taxing is corruption on international investors? The Review of Economics and Statistics 82, 1–11. Williamson, O.E., 1975. Markets and hierarchies: Analysis and antitrust implications. Free Press, New York. Yeung, B., 2003. Introduction to the JIBS symposium honoring the 25th anniversary of the future of the multinational enterprise. Journal of International Business Studies 14, 101–107. Xu, Y., Malkiel, B., 2003. Investigating the behavior of idiosyncratic volatility. Journal of Business 58, 2321–2349. Zhao, M., 2006. Conducting R&D in countries with weak intellectual property rights protection. Management Science 52, 1185–1199.