Journal of Corporate Finance 46 (2017) 320–341

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

Languages and corporate savings behavior夽 Shimin Chen a , Henrik Cronqvist b , Serene Ni c, * , Frank Zhang d a

China Europe International Business School, China University of Miami, United States SHU-UTS SILC Business School, Shanghai University, China d Murdoch University, Australia b c

A R T I C L E

I N F O

Article history: Received 17 April 2017 Received in revised form 11 July 2017 Accepted 24 July 2017 Available online 2 August 2017 Keywords: Corporate savings behavior Linguistic hypothesis

A B S T R A C T Speakers of strong future time reference (FTR) languages (e.g., English) are required to grammatically distinguish between future and present events, while speakers of weakFTR languages (e.g., Chinese) are not. We hypothesize that speaking about the future in the present tense may result in the belief that adverse credit events are more imminent. Consistent with such a linguistic hypothesis, weak-FTR language firms are found to have higher precautionary cash holdings. We report additional supportive results from changes in the relative importance of different languages in a country’s business domain, evidence from within one country with several distinct languages, and results related to changes following a severe financial crisis. Our evidence introduces a new explanation for heterogeneity in corporate savings behavior, provides insights about belief formation in firms, and adds to research on the effects of languages on economic outcomes. © 2017 Published by Elsevier B.V.

The diversity of languages is not a diversity of signs and sounds but a diversity of views of the world. —Wilhelm von Humboldt, “On the Comparative Study of Languages,” 1820.

1. Introduction The notion that languages may influence decision-making has a long history in social science (e.g., Campbell, 2003). In particular, the linguistic relativity hypothesis, also referred to as the Sapir-Whorf hypothesis, posits that the structure of a

夽 We are thankful for the comments from the seminar participants at Accounting & Finance Association of Australia and New Zealand, China Europe International Business School, China International Conference in Finance, Fudan University, Global Finance Conference, Massey University, Shanghai University of Finance and Economics, Stockholm School of Economics, University of Auckland, University of Miami, Vietnam International Finance Conference, and Xi’an Jiaotong-Liverpool University, and Renée Adams, Nick Barberis, Bo Becker, Henk Berkman, Chen Chen, Stefanos Delikouras, David Denis, Yuan Ding, Anand Goel, Jarrad Harford, Danling Jiang, Kathy Kahle, George Korniotis, Alok Kumar, Andy Leone, Feng Li, Kai Li, Florencio López-de-Silanes, Timothy Lu, Yadong Luo, Ben Marshall, Mili Mormann, Michael Naylor, Mattias Nilsson, Thomas Post, Amit Seru, Paolo Sodini, Per Strömberg, Anh Tran, Mitch Warachka, Bin Xu, Frank Yu, Xiaoyun Yu, and Tim Zhu. We are thankful to Keith Chen for sharing his language structure data. We also thank James Kent, Miranda Sun, and Nancy Yao for their excellent research assistance. We thank Standard & Poor’s China and Hong Kong teams for their help with various data issues. We acknowledge the generous research funding from China Europe International Business School, and appreciate the financial support from the National Natural Science Foundation of China (approval number: 71332004, 71672165, 71202091). * Corresponding author. E-mail addresses: [email protected] (S. Chen), [email protected] (H. Cronqvist), [email protected] (S. Ni), [email protected] (F. Zhang).

http://dx.doi.org/10.1016/j.jcorpfin.2017.07.009 0929-1199/© 2017 Published by Elsevier B.V.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

321

language may shape its speakers’ representations of reality (e.g., Whorf et al., 1956). As a concrete example, it has been wellestablished in linguistics research that languages have different ways of grammatically referencing the future (e.g., Dahl, 1985). For example, an English speaker repeatedly changes the structure of a sentence when referencing a future, as opposed to a present, event (e.g., “It will be ...” as opposed to “It is ...”). In contrast, a Chinese speaker may use the present tense when referencing a future event.1 Do such differences across languages affect the ways individuals view the world, form beliefs, and in the end make corporate decisions?2 It has recently been shown that heterogeneity in languages explains the variation in individual savings behavior (e.g., Chen, 2013 and Sutter et al., 2015). A follow-up question is whether these language effects also carry over to behaviors in the corporate domain. In this paper, we focus on one aspect of corporate financial decision-making, namely the propensity of firms to hold cash. Countries have different savings cultures (e.g., Carroll et al., 1994, 1999), which may be reflected in the behaviors of individuals as well as corporations. In fact, a precautionary motive for savings has been proposed for individuals (e.g., Kimball, 1990 and Hubbard et al., 1995) as well as for corporations (e.g., Keynes, 1934), with supporting evidence in economics and finance (e.g., Lusardi, 1998 and Opler et al., 1999). Indeed, Fig. 1 shows a strong positive correlation (q = 0.47) between individual and corporate savings propensities for the largest countries in the world, suggesting that some of the factors affecting individual behavior may also impact corporations. This positive correlation is intriguing and we believe that this is the first time that it has been documented. In this paper, we examine whether differences in corporate savings behavior may partly be attributable to language heterogeneity. We report a collage of evidence that supports the linguistic hypothesis, starting with a large-sample empirical approach. Language heterogeneity is found to explain variation in corporate savings behavior: Firms in weak future time reference (FTR) language countries (i.e., countries where speakers may reference a future event with the present tense, e.g., China) are found to have substantially higher average cash holdings. This evidence dovetails well with previous evidence in economics on the savings behavior of individuals (e.g., Chen, 2013). The estimated effect is statistically significant, economically sizable, and does not attenuate after we control for differences in industry composition or include an extensive set of firm- and country-level controls that have previously been found to explain variations in cash holdings. Importantly, we find that the language effect is not simply absorbing legal origin, colonization, or alternative culture effects. We also report results from several other empirical approaches. First, analyzing difference-in-differences, we find that Hong Kong based companies, compared to several control groups, have increased their cash holdings as the Chinese language has increased in importance in the business domain after the 1997 handover (i.e., the transfer of sovereignty over Hong Kong from the U.K. to China). Second, within Switzerland, firms in weak-FTR language regions are found to have higher cash holdings compared to those in strong-FTR language regions. Finally, we find that weak-FTR language firms’ cash holdings increased relatively more after the 2008 financial crisis, suggesting that differences in beliefs about the time distance to another adverse credit event is one possible mechanism through which language affects variation in cash holdings. Our study contributes to several research disciplines. First, we introduce language heterogeneity as a novel explanation for variations in corporate financial decision-making. The scientific basis for our linguistic hypothesis is pre-existing evidence in the intersection of economics, linguistics, and psychology (e.g., Aronoff and Rees-Miller, 2003). We show that language explains not only savings behavior at the level of individuals (e.g., Chen, 2013 and Sutter et al., 2015), but also corporate savings propensities. Our results contribute to recently emerging evidence on the effect of languages on corporate decision-making (e.g., Brannen et al., 2014 and Kim et al., 2017). Second, explaining differences in corporate cash holdings has resulted in several studies in international corporate finance (e.g., Dittmar et al., 2003, Kalcheva and Lins, 2007, and Pinkowitz et al., 2016).3 We contribute a new perspective by suggesting that heterogeneous beliefs, based on language structure differences, may explain why some firms exhibit a higher propensity of holding cash. That is, language may explain variation in the precautionary savings motive for cash across firms around the world. The rest of the paper is organized as follows. Section 2 reviews related research and develops our linguistic hypothesis. Section 3 describes the construction of our data set. Section 4 reports our empirical evidence and several robustness checks. Section 5 reports further evidence and extensions. Finally, Section 6 concludes. 2. Related research 2.1. Linguistic relativity hypothesis The Sapir-Whorf hypothesis (SWH) posits that the structure of a language systematically affects its speakers’ representations of reality (e.g., Whorf et al., 1956 and Boroditsky, 2003). In spite of the possibility of translating from one language to another,

1 Anecdotal evidence from Chinglish, i.e., English that is influenced by the Chinese language, emphasizes these differences in that Chinglish speakers often ignore verb conjugation and tense. 2 Evolutionary linguists explain the origins of languages and heterogeneity across different languages based on environmental constraints related to, e.g., the division of labor and reproduction in ancient agro-societies (e.g., Johansson, 2005). Characteristics of languages, e.g., the ways of grammatically referencing the future, have remained stable over thousands of years. That is, language structure may be expected to be exogenous to current economic outcomes (e.g., Tabellini, 2008), reducing endogeneity concerns in an empirical analysis of a linguistic hypothesis. 3 For a more comprehensive review of research related to corporate cash holdings, we refer to Almeida et al. (2014).

322

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

Fig. 1. Corporate savings propensity vs. individual savings propensity. The figure shows the relationship between corporate savings propensity and individual savings propensity across different countries and regions. The data on corporate savings are from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The data on individual savings are from the World Bank and cover the 1989–2013 period.

“there will always be an incommensurable residue of untranslatable culture associated with the linguistic structures of any given language” (e.g., Kramsch, 1998, p. 12).4 The strong form of the SWH (“linguistic determinism”) argues that language structure may affect beliefs and behaviors by explicitly controlling cognitive processes. While there is little evidence to support the strong form (e.g., Chomsky, 1957 and Pinker, 1994), a weaker form argues that language heterogeneity may affect non-linguistic behaviors without controlling cognitive processes. For example, Slobin (1987) asserts that language affects thought because a language may grammatically require speakers to attend to and encode different aspects of their experiences when speaking, e.g., future-referencing of events, gender-referencing. Such variations in grammatical systems may affect its speakers’ representations of reality, based on recent psychology research (e.g., Boroditsky, 2001 and Evans and Levinson, 2009). 2.2. Language structure heterogeneity and future time reference It has been well-established in linguistics research that languages have different ways of grammatically referencing the future. Specifically, Dahl (2000) characterizes languages as “futureless” if they do not require “obligatory [future time reference] use in (main clause) prediction-based contexts.” Thieroff (2000) refers to this subset as “weakly-grammaticalized future” languages. The difference is the requirement to use a specific grammatical construct when speaking about the future. In some languages, e.g., English, future time reference (FTR) is used explicitly even in prediction-based contexts, i.e., even if speaking about future events that are not under the speaker’s control. In other languages, e.g., Chinese, this is not the case, and the future is referred to with the present tense. This paper focuses on such language structure heterogeneity in referencing the future. A concrete example may be used to explain such linguistic heterogeneity. A speaker of the English language has to grammatically refer to the future with constructs such as “will” or “be going to.” In contrast, weak-FTR languages do not require the speaker to encode any difference between the future and the present. For example, Chinese has no tenses, so a speaker of the Chinese language is not required to use any equivalent of “will” in the English language.5 As a result, if discussing expectations about the profitability of a firm over the next couple of years, English and Chinese speakers will use different grammatical future references: English: Earnings will be lower Chinese (pinyin): Lìrùn xiàjiàng Chinese (translated): Earnings are lower

4 5

For a review of research related to linguistics, culture, and economics, we refer to Ginsburgh and Weber (2014). There exists a large number of different dialects in China (e.g., Chang et al., 2015), but these differences are not with respect to FTR language structure.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

323

It is important to emphasize several points related to language structure heterogeneity and future time reference. First, languages differ with respect to what future reference they require of their speakers, not with respect to what they may use (e.g., Jakobson and Halle, 1956). Speakers of weak-FTR languages are of course still able to distinguish between the future and the present using their languages, but they are not required to do so each time they speak. That is, we do not argue that it is impossible to change a sentence in a weak-FTR language to refer to the future, but simply that speakers of weak-FTR languages are more likely to not mark the future in their daily conversations, and thus may be less precise about the difference between the present and the future. Second, while there exists a number of language structure differences, we focus on one specific source of heterogeneity, i.e., future time reference, because it is well-established in linguistics research (e.g., Dahl, 1985, 2000). Recent research has reported that heterogeneity in language structure explains differences in economic behaviors across individuals from different countries. For example, Chen (2013) finds that an individual’s savings behavior is partly explained by the structure of the language that an individual speaks. Sutter et al. (2015) confirm this conclusion using an experimental approach. In organizational behavior, there also exists some emerging and related evidence. For example, Liang et al. (2014) find that language heterogeneity with respect to future time reference affects firms’ social responsibility policies. 2.3. A linguistic hypothesis in corporate finance Following the Sapir-Whorf hypothesis, we develop a linguistic hypothesis. Specifically, we apply the hypothesis to corporate savings behavior, i.e., the propensity of firms to hold cash. If a firm engages in value maximization, the cash holdings should be set so that the marginal benefit equals the marginal cost. If financial markets were perfect, in the sense of no frictions, firms may raise external cash for any positive NPV investment at any time (e.g., Modigliani and Miller, 1958). In reality, firms may be concerned about their future ability to finance ongoing projects and new investments with external funds, resulting in a precautionary motive for cash holdings (e.g., Keynes, 1934). In other words, firms may use the saved cash if other sources of funding are not available or are very costly. Prior research has found empirical evidence of a precautionary motive for corporate cash holdings (e.g., Opler et al., 1999, Han and Qiu, 2007, and Bates et al., 2009). The precautionary motive for corporate cash holdings is contingent on bad credit events occurring in the future. We hypothesize that languages may affect time-distance beliefs, i.e., whether an adverse credit event is imminent or not. For example, repeatedly speaking about the future in the present tense may result in the belief that adverse credit events are more imminent. In addition, if speakers do not use different tenses, then it may also make past bad credit events seem more vivid. For centuries, speakers of strong-FTR languages have used what linguists call the “historical present,” i.e., describing past events in the present tense, to make them appear more vivid and immediate (e.g., Schiffrin, 1981). A belief that an adverse credit event is more immediate may result in larger precautionary cash holdings. We hypothesize that firms in weak-FTR countries exhibit a higher propensity of holding cash. First, Lins et al. (2010) conduct a survey and report that firms use additional cash holdings to hedge against future adverse events, while credit lines are used as an option to exploit opportunities in cases of positive shocks. As a result, our focus in this paper is on cash holdings, and not credit lines.6 Second, managers of firms are likely to be risk averse, i.e., care relatively more about bad events. This has previously been found to affect firms’ financial decisions (e.g., Berger et al., 1997 and Graham et al., 2013). Finally, firms may focus more on adverse credit events compared to positive shocks because of costly financial distress. As a result, we expect that weak-FTR language firms hold more cash. 3. Data In this section, we describe the construction of our data set, and we also explain the language structure classification and the measures used in our empirical analysis. 3.1. Language structure classification and measures The explanatory variable of main interest is Weak-FTR Language, a language structure measure that we did not develop for this paper, but adopt from the European Science Foundation’s Typology of Languages in Europe (EUROTYP) project. This project is the most extensive research program to study the cross-linguistic grammaticalization of future time referencing. The project was funded by the European Science Foundation, involved about 100 linguistic scholars over five years (1990–1994), and resulted in an 800-page report on Tense and Aspect, edited by Östen Dahl, Professor of General Linguistics at Stockholm University, and published in 2000. In this study, we use “weak-FTR” languages for what Dahl (2000) calls “futureless” and Thieroff (2000) calls “weaklygrammaticalized future” languages. We classify other languages as “strong-FTR” languages. As a result, Weak-FTR Language is an indicator variable that is one for firms in weak-FTR language countries, and zero otherwise (e.g., Chen, 2013). For most of the countries in our data set, Chen (2013) provides the dominant languages used in those countries. For other countries, we use the

6 If speakers of weak-FTR languages believe that positive shocks are closer in time, this may affect firms’ credit lines, but this hypothesis is challenging to examine across countries because of lack of data on credit lines.

324

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

Fig. 2. Languages and corporate savings behavior. The figure shows the average cash to total assets among firms in different countries and regions. Dark-color bars represent weak-FTR countries and regions, while light-color bars represent strong-FTR countries and regions. The classification of weak-FTR (“future time reference”) versus strong-FTR language countries and regions is based on Chen (2013). The Appendix contains the definitions of all the variables.

approach developed by Chen (2013) to classify the countries. We refer to Lewis (2009) and the Ethnologue database for further details.7 3.2. Construction of data set We construct an unbalanced panel data set of public firms from 1989 to 2013. We obtain these data from Standard & Poor’s Compustat Global and CapitalIQ databases. We exclude financial (SIC codes 6000–6999) and utility (SIC codes 4900–4999) companies because they may choose policies to meet some regulatory requirements. We drop firms with sales of less than $10 million, i.e., the very smallest firms in the database. We also drop firms in the top and bottom 1% of the distributions to reduce the influence of extreme values. The final sample consists of 224,788 (31,061) firm-year (firm) observations in 44 different countries, each of which has at least 300 firm-year observations in total across all years. About 43.3% of the firm-years are in 17 weak-FTR countries and 56.7% are in 27 strong-FTR countries. The Appendix contains the definitions of all of the variables analyzed in this study. 4. Empirical evidence In this section, we first report descriptive evidence, followed by several sets of regressions where we follow pre-existing research and control for industry, firm, and country heterogeneity, and finally we report several robustness checks. 4.1. Descriptive evidence Several pieces of descriptive evidence are suggestive of a relationship between languages and corporate savings behavior. Fig. 2 reports the relationship between language structure and corporate cash holdings for the countries in our data set. The countries are displayed in descending order with respect to average Cash/Assets. Without controlling for any industry-, firm-, or country-level heterogeneity, the figure suggests that there is a strong relationship between language and cash holdings. Darkcolor bars represent weak-FTR countries, while light-color bars represent strong-FTR countries. The figure shows that average

7 A common question is why Portugal and Brazil have different classifications. In both European Portuguese (EP) and Brazilian Portuguese (BP), there are two forms for the future tense, cantarei (which is strong-FTR) and vou cantar (which is weak-FTR). Both forms are common in EP, but the first one is very formal in BP. That is, individuals who speak BP generally use the second one. As a result, Chen (2013) classifies BP as a weak-FTR language.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

325

Table 1 Summary statistics. N (Firms) Panel A: Weak-FTR language countries and regions Austria 99 Belgium 122 Brazil 217 China 2085 Denmark 133 Finland 134 Germany 749 Hong Kong 896 Indonesia 218 Japan 3422 Malaysia 843 Netherlands 221 Norway 201 Singapore 587 Sweden 344 Switzerland 234 Taiwan 1415 Mean Median Std. dev. Panel B: Strong-FTR language countries and regions Australia 797 Canada 1389 Chile 95 France 805 Greece 229 India 1621 Ireland 95 Israel 244 Italy 273 Jordan 53 Mexico 72 New Zealand 91 Pakistan 167 Peru 56 Philippines 96 Poland 288 Portugal 67 Russia 85 South Africa 293 South Korea 1163 Spain 145 Sri Lanka 111 Thailand 400 Turkey 126 United Kingdom 1762 United States 8463 Vietnam 155 Mean Median Std. dev.

N (Firm-years)

Cash/Assets

852 1092 1103 9620 1117 1372 6182 5995 1328 40,612 6981 1885 1157 4382 2369 2389 8966

0.098 0.114 0.137 0.173 0.111 0.108 0.114 0.186 0.099 0.158 0.104 0.095 0.118 0.159 0.104 0.141 0.166 0.148 0.121 0.114

4697 7605 560 6820 1584 8004 808 1344 2141 197 495 590 1058 313 603 1530 511 365 2089 5702 1344 615 3244 705 12,158 61,916 388

0.088 0.093 0.064 0.131 0.080 0.073 0.137 0.180 0.105 0.082 0.082 0.063 0.079 0.056 0.105 0.083 0.055 0.084 0.115 0.133 0.085 0.085 0.078 0.094 0.104 0.113 0.106 0.106 0.063 0.120

The table reports summary statistics for the data set used in this study. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The classifications of weak-FTR (“future time reference”) language and strong-FTR language countries is based on Chen (2013). The Appendix contains the definitions of all the variables.

cash holdings in weak-FTR language countries are 12.9% compared to 9.4% in strong-FTR language countries. In other words, the unconditional difference in cash holdings of weak- and strong-FTR language firms is substantial.8 In Table 1, we compare the difference in mean (median) cash holdings among all the firm-year observations in weak-FTR versus strong-FTR countries. We find that the mean (median) Cash/Assets among weak-FTR firms is 14.8% (12.1%) compared

8 We note that the cash holdings data do not seem to exhibit significant heteroskedasticity because the standard deviation of corporate savings behavior is very similar in weak-FTR countries (0.114) and strong-FTR countries (0.120). This reduces the possibility of biased statistical inference. In addition, all reported standard errors in all relevant tables in the paper are White (1980) heteroskedasticity-robust.

326

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

to 10.6% (6.3%) for strong-FTR countries. That is, comparing the means (medians), we find that public companies in weak-FTR countries have 4.2 (5.8) percentage points higher cash holdings. These differences are all statistically significant at the 1%-level. 4.2. Controlling for industry and firm heterogeneity Because heterogeneity at the industry- or firm-levels may be partially responsible for the previously reported difference in corporate savings behavior of weak- and strong-FTR companies, we turn to regressions to control for such sources of variation across the firms in our data set. We first estimate the following model specification: yijkt = b0 + b1 Weak-FTR Languagekt + b2 Xit + 0t + 0j + 4ijkt

(1)

where y is cash holdings, i indexes firms, j industries, k countries, and t years. X is a vector of time-varying firm control variables, 0 are sets of fixed effects, and 4 is an error term. All reported standard errors in all relevant tables in the paper are (White, 1980) heteroskedasticity-robust and clustered by country. Table 2 reports the relationship between languages and corporate savings behavior. The dependent variable is Cash/Assets, and the explanatory variable of interest is Weak-FTR Language. Column (1) includes Weak-FTR Language as the only explanatory variable in addition to year fixed effects, in order to characterize the average difference in cash of firms in weak- and strong-FTR language countries and controlling for aggregate year-to-year variation. We find that the point estimate on Weak-FTR Language is positive (0.041) and statistically significant at the 1%-level. That is, the size of the coefficient is similar to the unconditional difference in cash holdings between weak- and strong-FTR firms which we reported previously. One concern with this model specification is that weak- and strong-FTR language countries may have different industry structures. For example, the language in a country may in part explain which industries are created in that country, and industry may in turn correlate with the marginal benefits and costs of cash holdings. In column (2), we therefore add industry fixed effects based on 2-digit SIC codes. We find that industry differences increase the explanatory power but do not explain the language effect because the size of the point estimate on Weak-FTR Language is essentially unchanged (0.042 versus 0.041) after we add industry fixed effects. A related concern is that companies in weak- and strong-FTR language countries have different firm characteristics, which may also correlate with corporate behavior and cause an omitted variables problem. In columns (3) to (6), we add several sets Table 2 Controlling for industry and firm heterogeneity.

Weak-FTR language

(1)

(2)

(3)

(4)

(5)

(6)

(7)

0.041*** (0.010)

0.042*** (0.009)

0.046*** (0.010) 0.001 (0.002) 0.030*** (0.004)

0.044*** (0.009)

0.039*** (0.007)

0.042*** (0.010)

0.034*** (0.007) 0.001 (0.001) 0.021*** (0.002) −0.036* (0.019) 0.011 (0.008) −0.188*** (0.036) 0.573*** (0.081) −0.138*** (0.010) −0.233*** (0.022) −0.220*** (0.010) −0.006 (0.005) 0.152*** (0.008) 224,788 0.301 Yes Yes

Firm size Market-to-book ratio

−0.013 (0.038) 0.037*** (0.008)

Cash Flow/Assets Cash Flow volatility

−0.148*** (0.034) 0.774*** (0.108) −0.071*** (0.018) −0.246*** (0.025)

Capex/Assets R&D/Sales Net working capital/Assets Acquisitions/Assets Leverage Dividend indicator Constant N Adjusted R-squared Year fixed effects Industry fixed effects

0.107*** (0.006) 224,788 0.036 Yes No

0.095*** (0.001) 224,788 0.114 Yes Yes

0.050*** (0.006) 224,788 0.154 Yes Yes

0.092*** (0.003) 224,788 0.116 Yes Yes

0.102*** (0.003) 224,788 0.189 Yes Yes

−0.204*** (0.012) −0.012 (0.008) 0.162*** (0.007) 224,788 0.197 Yes Yes

The table reports regressions of corporate cash holdings in which we control for industry and firm heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

327

of firm characteristics that capture a variety of previously studied economic mechanisms of relevance for the marginal benefits and costs of cash, e.g., economies of scale, growth opportunities, operating cash flows and volatility, investment intensity, corporate financial structures. We first control for Firm Size and Market-to-Book Ratio. We also control for Cash Flow/Assets and Cash Flow Volatility. In addition, we control for Capex/Assets, Net Working Capital/Assets, R&D/Sales, and Acquisitions/Assets. Finally, we control for Leverage and also include a Dividend Indicator. Column (7) includes all the firm characteristics simultaneously. We find that the point estimate on Weak-FTR Language is reduced somewhat when including all firm-level controls, but remains statistically significant at the 1%-level. The results reported so far show an economically sizable and statistically significant relationship between languages and corporate savings behavior. Specifically, firms in weak-FTR language countries are found to have higher cash holdings. The estimated effect is found to attenuate only about 17% (= (0.041 − 0.034)/0.041) after we control for differences in industry composition of weak- and strong-FTR country firms and the same extensive set of firm-level controls that have been found in pre-existing work to explain heterogeneity in cash holdings. 4.3. Controlling for country heterogeneity A potentially significant concern with the evidence reported so far is that the regressions may omit variables that capture important economic differences across countries which are related to economic growth and frictions in financial markets, and therefore the marginal benefits and costs of cash holdings. That is, language structure may correlate with a variable that our model specification does not control for. As a result, we estimate the following model specification: yijkt = b0 + b1 Weak-FTR Languagekt + b2 Xit + b3 Xkt + 0t + 0j + 4ijkt

(2)

where y is cash holdings, i indexes firms, j industries, k countries, and t years. X are vectors of time-varying firm and country control variables, 0 are sets of fixed effects, and 4 is an error term. Table 3 reports the relationship between languages and corporate savings behavior, controlling for several important sources of country variation. Note that all of these regressions include the full set of firm-level controls from the previously reported

Table 3 Controlling for country heterogeneity.

Weak-FTR language Log (GDP/Capita) GDP growth Real interest rate

(1)

(2)

(3)

(4)

(5)

0.039*** (0.008) 0.002 (0.002) −0.019 (0.019) 0.000 (0.000)

0.032*** (0.012)

0.035*** (0.008)

0.034*** (0.008)

0.036** (0.014) 0.003 (0.002) −0.017 (0.017) 0.001 (0.001) −0.017 (0.014) −0.003 (0.021) −0.027 (0.019) 0.002 (0.004) 0.001 (0.003) −0.005* (0.003) 0.175*** (0.036) 194,036 0.310 Yes Yes Yes

−0.002 (0.009) 0.009 (0.014) −0.034** (0.014)

French legal origin German legal origin Scandinavian legal origin

−0.001 (0.004) 0.001 (0.003)

Creditor rights index Shareholder rights index Legal rights index Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls

0.152*** (0.031) 196,131 0.307 Yes Yes Yes

0.153*** (0.009) 224,788 0.304 Yes Yes Yes

0.150*** (0.009) 221,156 0.301 Yes Yes Yes

−0.002 (0.002) 0.172*** (0.016) 213,549 0.300 Yes Yes Yes

The table reports regressions of corporate cash holdings in which we control for country heterogeneity, in addition to industry and firm differences. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The dependent variable is Cash/Assets. Weak-FTR Language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

328

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

analysis. First, we control for a country’s economic development. To the extent that weak-FTR is related to a country’s GDP and growth, the previously reported effect may simply represent heterogeneity in economic development among the sample of countries that we analyze. It is worth emphasizing that among the largest emerging economies in the world, the so-called BRICS countries, there are both weak- and strong-FTR countries represented. Thus, it is not obvious that economic development correlates with language structure, at least not based on this casual observation. In column (1), we find that controlling for standard measures of economic development, e.g., the log of GDP/Capita, GDP Growth, and Real Interest Rate, does not alter our results.9 Second, it is also potentially important to control for a country’s legal origin and governance (e.g., Pinkowitz et al., 2006, Kalcheva and Lins, 2007, and Lins et al., 2010).10 Indeed, legal origin is the most extensively studied country characteristic in international corporate finance research in the past decades (e.g., La Porta et al., 1997a, 1998).11 We want to emphasize that our linguistic hypothesis provides a very different prediction than an agency hypothesis, i.e., these are not competing hypotheses. For example, French legal origin is associated with the weakest corporate governance, which in turn may correlate with higher corporate cash holdings because of more acute agency problems. In contrast, French is a strong-FTR language, which suggests lower cash holdings based on the linguistic hypothesis. We find that the correlation between language structure and legal origin is weak (0.071). That is, our language measure captures different heterogeneity than a legal origin measure. In column (2), we find that controlling for French Legal Origin, German Legal Origin, and Scandinavian Legal Origin does not change our previous conclusion. In column (3), we include several measures capturing cross-country variation in Creditor Rights Index (e.g., Djankov et al., 2007) and Shareholder Rights Index (e.g., Djankov et al., 2008) to further rule out that our language measure is capturing more acute agency problems. In column (4), we include the Legal Rights Index, i.e., the World Development Indicator (WDI) developed by the World Bank. Column (5) includes all country characteristics simultaneously. We find that these standard measures of differences in legal origin and governance cannot explain the effect of language on corporate behavior.12 The evidence reported so far supports our linguistic hypothesis that language structure is related to corporate financial decision-making. Specifically, firms in weak-FTR language countries are found to have significantly higher cash holdings for reasons that cannot be attributable to industry-, firm-, or country-level characteristics. The estimated effect of weak-FTR language is very sizable. Firms in weak-FTR countries are found to have about 3.6 percentage points higher corporate cash holdings compared to firms in the same industry and with similar firm and country characteristics as strong-FTR firms. This effect is substantial, considering that the average cash holding as a percentage of assets is 10.6% in strong-FTR countries. In other words, the average weak-FTR language firm holds about 34.0% (= 0.036/0.106) more cash than the average strong-FTR language firm, controlling for important sources of variation across the firms in our data set. 4.4. Alternative measures of country culture Language is an integral component of a country’s culture (e.g., Cavalli-Sforza, 2001 and Stulz and Williamson, 2003), but we want to rule out that it is another measure of a country’s culture that is in fact responsible for our findings.13 That is, we explore whether weak-FTR is simply captured by other deeper cultural characteristics of a country. In Table 4, we therefore control for several alternative country culture measures. 4.4.1. Religion Researchers in corporate finance have previously analyzed the effects of a broad measure of culture based on religion (e.g., Stulz and Williamson, 2003).14 In column (1), we include a set of measures of religion at the country-level: Catholic, Protestant, Muslim, and Buddhist. We find that the point estimate on Weak-FTR Language is still statistically significant at the 5%-level, and economicallysizable. Thus, language effects do not simply capture religion effects.15 4.4.2. Corruption and trust Several studies report a significant relation between corruption, trust, and economic outcomes (e.g., La Porta et al., 1997b, Guiso et al., 2004). For example, Guiso et al. (2006) show that trust between inhabitants of different countries is partly explained by the commonality in languages. In column (2), we therefore include Corruption and Trust (e.g., La Porta et al., 1997b), but we do not find that these culture measures explain the previously reported language effect.

9 The results are unaffected if we also add controls for a country’s marginal corporate tax rate and the development of financial markets (i.e., market capitalization of public companies/GDP). 10 Corporate governance has been found to impact cash holdings also within the U.S. (e.g., Dittmar and Mahrt-Smith, 2007, Harford et al., 2008, and Gao et al., 2013). 11 For a review of the economic effects of legal origins, we refer to La Porta et al. (2008). 12 We note that adding an extensive set of country-level characteristics does not increase the explanatory power of the estimated model specification as much as including firm-level characteristics, a result that is similar to the conclusion by Pinkowitz et al. (2016). 13 Economics is currently experiencing a rapid increase in research related to country culture and its impact on behavior and decision-making by individuals as well as corporations. For recent reviews, we refer to Alesina and Giuliano (2015) and Guiso et al. (2015). 14 Religion effects have also been found within the U.S. (e.g., Hilary and Hui, 2009). 15 As an additional robustness check, we have also excluded all countries and regions which may be categorized as “Confucian” (i.e., China, Hong Kong, Japan, Korea, Singapore, Taiwan, and Vietnam), but we still find a statistically significant and economicallysizable effect.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

329

Table 4 Alternative measures of country culture.

Weak-FTR language Catholic Protestant Muslim Buddhist

(1)

(2)

(3)

(4)

Religion

Trust

Long-term persistence

Hofstede dimensions

0.028** (0.012) −0.066 (0.041) −0.025*** (0.008) −0.031*** (0.009) 0.024 (0.019)

0.027*** (0.006)

0.035*** (0.006)

0.029*** (0.010)

−0.010*** (0.002) −0.058* (0.031)

Corruption Trust

−0.027 (0.049) 0.077 (0.337)

Loss of war Civil war Uncertainty avoidance Long-term orientation Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls

0.185*** (0.030) 192,137 0.319 Yes Yes Yes Yes

0.233*** (0.045) 179,033 0.325 Yes Yes Yes Yes

0.295*** (0.051) 107,680 0.279 Yes Yes Yes Yes

0.017 (0.028) 0.106*** (0.037) 0.121** (0.054) 189,864 0.318 Yes Yes Yes Yes

The table reports results from alternative measures of country culture. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Country controls is the full set of country-level characteristics included in Table 3. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

4.4.3. Long-term persistence An emerging literature in economics report evidence of long-term and persistent culture effects on economic outcomes; see, e.g., Alesina and Giuliano (2015) and the references therein. We hypothesize that wars may be events that affect corporate savings behavior over a long time period. In column (3), we therefore include indicator variables for Loss of War and Civil War, but these measures do not explain the previously reported language effect. 4.4.4. Hofstede dimensions Culture research outside the area of economics sometimes relies on aggregate value-based measures of culture, such as those constructed by Hofstede, 1980. One problem with such measures is that they are contemporaneous to the studied economic outcomes and subject to endogeneity concerns (e.g., Ahern et al., 2015). In column (4), we include Uncertainty Avoidance and Long-Term Orientation (e.g., Hofstede, 1980). We find that the point estimate on Weak-FTR Language is still statistically significant at the 1%-level, and economicallysizable.16 4.5. Controlling for language evolution Languages may be related through cultural evolution (e.g., Dryer, 1989), which creates econometric challenges when estimating a relationship between cultural characteristics. That is, cultures are sets of characteristics, both linguistic and

16 We have also controlled for the risk aversion in a country using data from the World Value Survey (in particular question A195, on risk attitudes in a country). These data are only available after 2005. Again, we find that our conclusions are not affected.

330

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341 Table 5 Controlling for language evolution.

Weak-FTR language Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls Colony fixed effects Continent fixed effects Language family fixed effects

(1)

(2)

(3)

0.043*** (0.013) 0.187*** (0.038) 194,036 0.315 Yes Yes Yes Yes Yes No No

0.019*** (0.007) 0.125*** (0.033) 194,036 0.327 Yes Yes Yes Yes No Yes No

0.036*** (0.012) 0.237*** (0.046) 194,036 0.319 Yes Yes Yes Yes No No Yes

The table reports results from alternative measures of country culture. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Country controls is the full set of country-level characteristics included in Table 3. Colony fixed effects are based on data from Klerman et al. (2011). Language family fixed effect are based on data from the World Atlas of Language Structure and Roberts et al. (2015). Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

behavioral, which may be inherited as cultures split off or aggregated if cultures merge. This is a common concern when analyzing cross-cultural data. In other words, languages may not necessarily be considered independent observations (e.g., Roberts et al., 2015). In Table 5, we therefore control for language evolution, in particular the relatedness of languages, in several different ways. In column (1), we control for Colony Fixed Effects, based on Klerman et al. (2011), because colonization is a potentially important source of commonality of corporate behavior and is also naturally related to language evolution.17 In column (2), we control for geographical relatedness by including a set of Continent Fixed Effects. Within each continent, different languages may share common components. For example, the Chinese language has influenced the Japanese and Korean languages because of past regional wars. Finally, in column (3), we include a set of Language Family Fixed Effects based on data from the World Atlas of Language Structures. We find that the point estimate on Weak-FTR Language attenuates when we control for continent fixed effects, but remains statistically significant and economicallysizable. Importantly, the point estimate on Weak-FTR Language does not change significantly when we absorb heterogeneity based on geographical relatedness or historical language families. 4.6. Alternative estimation methods An econometric concern with the data set that we employ is that our regressions weigh countries dis-proportionately based on the number of firm-years in a country. This is an empirical challenge that our study has in common with contemporaneous research in international corporate finance and governance studies. If each firm-year is considered to be an independent observation, t-statistics may be overstated (e.g., Petersen, 2009). We address this concern by clustering all standard errors at the country-level in all relevant tables. In addition, Table 6 reports results using alternative estimation methods to confront concerns about sample inflation. In column (1), we regress average Cash/Assets on average firm characteristics, similar to some previous studies (e.g., Dittmar et al., 2003). We have also collapsed the data and re-estimated the model specifications. In column (2), we first estimate the regression without the Weak-FTR Language variable (but including all the other time-varying controls and fixed effects), then we collapse these residuals by country-years, and re-estimate the effect of language in this collapsed data set using weighted least squares (WLS) with the total number of firms per country and year as weights. In column (3), we first re-estimate the model specification without the Weak-FTR Language variable, but with country fixed effects (and all the other time-varying controls and fixed effects), and then we re-estimate the effect of language by regressing the country fixed effects on Weak-FTR Language using WLS.18 There is some variation with respect to the economic size of the point estimates, but the point estimates are all statistically significant at least at the 5%-level and economically sizable. Finally, in column (4), we match each weak-FTR language firm with one strong-FTR language firm using propensity score matching. Specifically, we construct the propensity score based on the size of the firm, the market-to-book ratio, and the

17

Economists have also found a relation between colonial origins and economic development (e.g., Acemo˘glu et al., 2001). We have checked that our results are unaffected if we use a bootstrap procedure. We compile a sample of the same size as the original sample by randomly drawing observations with replacement. We then estimate country fixed effects, and finally regress them on Weak-FTR Language using WLS. We repeat this procedure 1000 times. We find that zero is in the 1st percentile of the bootstrap distribution of the point estimate on Weak-FTR Language. 18

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

331

Table 6 Alternative estimation methods.

Weak-FTR language Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls

(1)

(2)

(3)

(4)

Regression of country means

Collapsing by firm-year and weighted least squares

Country fixed effects and weighted least squares

Propensity score matching

0.030*** (0.009) 0.078*** (0.027) 718 0.418 Yes Yes Yes Yes

0.019** (0.007) −0.009* (0.005) 691 0.074 Yes Yes Yes Yes

0.024*** (0.008) 0.180*** (0.005) 44 0.162 Yes Yes Yes Yes

0.034** (0.008) 0.131*** (0.041) 46,246 0.313 Yes Yes Yes Yes

The table reports results from alternative estimation methods. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Country controls is the full set of country-level characteristics included in Table 3. Column (1) reports regressions of country means. Column (2) reports estimates from the following estimation: We first estimate the model specification without the Weak-FTR language variable (but including all the other fixed effects and time-varying controls), then we collapse these residuals by country-years, and finally we re-estimate the effect of language structure in this collapsed data set using weighted least squares (WLS) with the total number of firms per country and year as weights. Column (3) reports estimates from the following estimation: We first re-estimate the model specification without the Weak-FTR language variable, but with country fixed effects (and all the other fixed effects and time-varying controls), then we collapse the data at the country level, and finally we re-estimate the effect of language by regressing the country fixed effects on Weak-FTR language using WLS. Column (4) shows the results based on propensity score matching. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

economic development of the county of the firm. We drop firm-years with caliper distances > 0.001. We find that the effect of weak-FTR language is still statistically significant at the 5%-level and economically sizable when analyzing matched data. Our results are similar if we match only to firms from a different language family. 4.7. Sources of additional cash holdings In Table 7, we examine the sources of the additional cash of weak-FTR language firms. First, in column (1), we analyze Investments, measured as capital expenditures divided by net property, plant and equipment at the beginning of the fiscal year. We find that weak-FTR language firms invest significantly less. As a result, one source of the additional cash is lower capital expenditures by weak-FTR language firms. This result also shows that weak-FTR language firms’ policies are affected more broadly than only cash holdings. Table 7 Sources of additional cash holdings. (1)

(2)

(3)

(4)

(5)

Depandent variable:

Investments

Interest payments

Debt issues

Equity issues

Dividends

Weak-FTR language

−0.048* (0.026) 0.178** (0.070) 224,282 0.175 Yes Yes Yes Yes Yes

−0.004*** (0.001) 0.038*** (0.005) 216,502 0.344 Yes Yes Yes Yes Yes

0.047** (0.021) 0.890*** (0.096) 187,348 0.032 Yes Yes Yes Yes Yes

−0.003 (0.004) 0.007 (0.012) 230,751 0.123 Yes Yes Yes Yes Yes

0.009*** (0.002) 0.011 (0.011) 236,279 0.158 Yes Yes Yes Yes Yes

Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls Continent fixed effects

The table reports regressions of corporate variables in which we control for industry, firm, and country heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. In Column (1), the dependent variable is Investments, which is capital expenditure divided by net property, plant and equipment at the beginning of the fiscal year. In Column (2), the dependent variable is Interest payments, which is total interest payments divided by total assets. In Column (3), the dependent variable is Debt issues, which is the change of debt from fiscal year t-1 to fiscal year t divided by the average debt. In Column (4), the dependent variable is Equity issues, which is the sales of common and preferred stock, net of purchases of common and preferred stock, divided by total assets. In Column (5), the dependent variable is Dividends, which is total dividends divided by total shareholder equity. WeakFTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Country controls is the full set of country-level characteristics included in Table 3. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

332

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341 Table 8 Alternative measures of corporate savings behavior. (1)

(2)

(3)

Dependent variable:

D Cash

Log (Cash/Sales)

Log (Cash/Net Assets)

Weak-FTR language

0.003*** (0.001) 0.030*** (0.008) −0.012** (0.005) 171,923 0.049 Yes Yes Yes Yes

0.506*** (0.121)

0.358*** (0.040)

−2.938*** (0.369) 194,036 0.316 Yes Yes Yes Yes

−2.472*** (0.159) 175,803 0.287 Yes Yes Yes Yes

Cash Flow × Weak-FTR language Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls

The table reports regressions of corporate variables in which we control for industry, firm, and country heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period. In Column (1), the dependent variable is DCash/Assets. In Column (2), the dependent variable is Log of (Cash/Sales) natural logarithm of cash and marketable securities divided by sales. In Column (3), dependent variable is Log of (Cash/Net Assets) natural logarithm of cash and marketable securities divided by net assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 2. Country controls is the full set of country-level characteristics included in Table 3. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

Another potential source of the additional cash is lower interest payments. We measure Interest Payments as total interest payment divided by total assets. In column (2), the estimated coefficient on the weak-FTR language indicator variable is negative and statistically significant. That is, lower interest payments is another partial explanation for the differences in corporate savings behavior. We measure Debt Issues as the change of debt from year t − 1 to year t divided by the average debt. In addition, we measure Equity Issues as sales of common and preferred stock, net of purchases of common and preferred stock, divided by total assets. In columns (3) and (4), we find evidence that weak-FTR language firms issue more debt, but no evidence that they issue more equity. That is, debt issues is another source of the additional cash among weak-FTR language firms. In addition, we measure Dividends as total dividends divided by total shareholder equity. In column (5), we find that weakFTR language firms pay out less in dividends. As a result, differences in dividends also partially explain the differences in corporate savings behavior. 4.8. Alternative measures of corporate savings behavior In Table 8, we study alternative measures of corporate savings behavior. First, in column (1), we examine the cash flow sensitivity of cash (e.g., Almeida et al., 2004 and Khurana et al., 2006). In this model specification, the dependent variable is change of cash holdings. We find that the point estimate on the interaction Cash Flow × Weak-FTR Language is positive (0.030) and statistically significant at the 1%-level. That is, we find that the propensity to save cash out of cash flows is significantly larger among weak-FTR language firms.19 Second, in column (2), we use the log of Cash/Sales, as the dependent variable because Sales may be less sensitive to country differences in accounting standards. We use log to address potential extreme values of this measure. The point estimate on Weak-FTR Language is positive (0.506) and statistically significant at the 1%-level. We find that the previous conclusion that weak-FTR language firms on average hold more cash is not affected by this alternative cash variable. Finally, in column (3), we use the log of Cash/Net Assets, as an alternative dependent variable (e.g., Dittmar et al., 2003 and Bates et al., 2009). The point estimate on Weak-FTR Language is positive (0.358) and statistically significant at the 1%-level. Again, the conclusion that weak-FTR language firms on average hold more cash is unaffected by this alternative measure of corporate savings behavior. 4.9. Robustness checks Table 9 reports several robustness checks.

19 For consistency, we use the same model specification as in the rest of the paper, i.e., the same set of control variables, but we have checked that our results do not change if we use the model specification in Eq. (8) in Almeida et al. (2004) or Eq. (1) in Khurana et al. (2006).

Weak-FTR language Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Industry heterogeneity: 3-Digit SIC codes

Controlling for political orientation

Banking industry development

Excluding U.S.

Excluding largest weak-FTR and strong-FTR countries

Excluding Asian countries

Asian countries only

0.041*** (0.011) 0.175*** (0.034) 194,036 0.321 Yes Yes Yes Yes

0.035*** (0.013) 0.179*** (0.037) 184,220 0.307 Yes Yes Yes Yes

0.042*** (0.014) 0.187* (0.110) 72,434 0.306 Yes Yes Yes Yes

0.039*** (0.012) 0.186*** (0.044) 130,249 0.266 Yes Yes Yes Yes

0.048*** (0.009) 0.118*** (0.030) 92,844 0.252 Yes Yes Yes Yes

0.030** (0.012) −0.195*** (0.005) 107,698 0.348 Yes Yes Yes Yes

0.015** (0.007) −0.213*** (0.012) 86,338 0.287 Yes Yes Yes Yes

The table reports regressions of corporate cash holdings in which we control for industry, firm, and country heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2013 period, except in column (2), where data are not available for 2013. The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. This classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 3. Country controls is the full set of country-level characteristics included in Table 4. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

Table 9 Robustness checks.

333

334

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

4.9.1. Industry heterogeneity In the previously reported regressions we control for industry-level heterogeneity by including industry fixed effects based on 2-digit SIC codes. We found only a small effect of controlling for industry heterogeneity. In column (1), we therefore include fixed effects based on 3-digit SIC codes, but the results are very similar. 4.9.2. Political orientation In column (2), we use data from the World Bank to control for left-to-right political orientation because previous research has related conservatism to corporate policies (e.g., Hutton et al., 2014 and Dittmar and Duchin, 2016). Specifically, we use data from the “Database of Political Institutions” (e.g., Beck et al., 2001), which are available throughout 2012. Controlling for political orientation does not alter our conclusion. 4.9.3. Banking industry development In column (3), we control for the development of the banking industry in each country. In particular, we use data from the Financial Access Survey (FAS) on the reported number of depositors scaled by the adult population. We find that firms hold less cash when a country’s banking industry is relatively more developed, but controlling for banking industry development does not alter our conclusion related to weak-FTR language and cash holdings. We have also considered other measures of the development of a county’s banking industry, e.g., the Z-score for a country’s banking system stability and private credit by deposit money banks and other financial institutions to a country’s GDP, but our results are again very similar. 4.9.4. Multinational corporations Some companies in our data set may be characterized as multinationals. The presence of such companies in our data set, in which the English language may be spoken in the business domain, may be expected to reduce the estimated effect of weak-FTR language. For example, if all corporate decision-makers in weak-FTR language countries spoke only English, then we may not expect to find any significant difference between weak-FTR and strong-FTR language firms’ corporate behaviors.20 That is, the presence of multinational companies in our data set may bias the estimated language effect downwards. 4.9.5. Excluding U.S. Observations from the U.S. may be affected by the country’s status as a “melting pot” with significant variation in cultures and languages. This may potentially reduce the size of the estimated effect of Weak-FTR Language. In column (4), we exclude the U.S. and find that the estimated effect is indeed somewhat stronger than previously reported. In fact, the language effect remains significant at least at the 5%-level if excluding any one country in our data set. 4.9.6. Excluding largest weak-FTR and strong-FTR countries Observations from the largest weak-FTR and strong-FTR countries constitute a substantial proportion of the data set that we analyze. In column (5), we drop both the largest weak-FTR language country (Japan) in terms of number of firms and also the largest strong-FTR language country (U.S.). The point estimate is comparable to the one reported for the full data set. 4.9.7. Effect of Asian countries Because several Asian countries are weak-FTR language countries and also have high cash holdings, we have checked the robustness of our results by excluding Asian countries altogether in column (6) and examining Asian countries only in column (7). We conclude that Asian countries alone cannot explain our result, and that the estimated effect of Weak-FTR Language is large and statistically significant also within Asia. 5. Further evidence and extensions In this section, we report further evidence and extensions related to our linguistic hypothesis regarding a relationship between language and corporate financial decision-making, using several different research methodologies to rule out alternative explanations. 5.1. Difference-in-differences evidence An event in which a country exogenously changes from a strong-FTR to a weak-FTR language (or vice versa) would provide researchers with important additional evidence on the effect of language structure on corporate savings behavior. In reality, we have to rely on naturally occurring events, which are imperfect, but still informative. For example, the increasing importance of the Chinese language in the business domain in Hong Kong after the 1997 handover (i.e., the transfer of sovereignty over Hong

20 The market for CEOs of large public companies is segmented across countries and even within the U.S. (e.g., Yonker, 2017). That is, U.S. companies are managed by English-speaking CEOs, Chinese companies are managed by Chinese-speaking CEOs, and so on.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

335

Kong from the U.K. to China) provides an opportunity to examine whether Hong Kong firms’ cash holdings have changed as predicted by our linguistic hypothesis.21 Hong Kong was ceded to the British by the Qing Dynasty in the 19th century, and for 150 years English was “the principal medium for intra-governmental written communication and records [...], [and] the preferred language for written contracts and records in the commercial sector” (So, 1996, p. 11). Chinese was declared a co-language in 1974, but did not have an equal position with English until the 1990s. Post-1997, the official language status of Chinese is emphasized by Chapter 1, Article 9 of the Hong Kong Basic Law: “[I]n addition to the Chinese language, English may also be used as an official language by the executive authorities, legislature, and judiciary of the Hong Kong Special Administrative Region.” While Chinese was also spoken in the business domain prior to the 1997 handover, our identification only assumes an increase in the use of Chinese at Hong Kong firms post-1997, relative to the use of Chinese at the control group of firms. We employ a difference-in-differences approach to examine changes in corporate cash holdings during a period when Hong Kong companies experienced an increasing use of a weak-FTR language in the business domain. We estimate the following model specification: yijkt = b0 + b1 Hong Kongit × Post − 1997 − Handovert + b2 Xit + b3 Xkt + 0t + 0j + 4ijkt

(3)

where y is cash holdings, i indexes firms, j industries, k countries, and t years. Hong Kong is an indicator variable that is one for Hong Kong companies, i.e., firms that are headquartered in Hong Kong, and zero otherwise. Mainland China firms that crosslist on the Hong Kong Stock Exchange are not included. Post-1997-Handover is an indicator variable that is one after the 1997 handover, and zero otherwise. X are vectors of time-varying firm and country control variables, 0 are sets of fixed effects, and 4 is an error term. Table 10 shows the results. In column (1), we find that the point estimate on the interaction effect Hong Kong × Post-1997-Handover is positive and statistically significant at the 1%-level. That is, Hong Kong firms on average increased their cash holdings in the post-1997 handover period compared to all non-Hong Kong firms in our data set. It is important to emphasize that this result cannot be explained by industry-, firm-, or country-characteristics for which we control. In addition, because of China’s “One Country, Two Systems” approach, there was no significant change in the economic systems or the financial market development that may have been expected to affect corporate cash holdings. We also note that the credit ratings of Hong Kong did not change significantly around the handover event. An Agence France-Presse article from October 31, 1997, confirms that “U.S. ratings agency Standard and Poor’s affirmed its stable sovereign outlook for Hong Kong.” That is, Hong Kong maintained a long-term local currency rating of AA- after the handover. This suggests that there was no dramatic increase in country risk that can easily explain the post-1997 increase in cash holdings among Hong Kong based firms. While this difference-in-differences approach addresses several empirical challenges, one concern is that the control group comprises a very heterogeneous set of firms. In column (2), we therefore use a smaller set of firms in South East Asian countries also affected by the 1997 financial crisis as our control group. We find that our estimated effect is somewhat reduced, but remains statistically significant and economically sizable. In column (3), we use only Singapore firms as the control group. Both Hong Kong and Singapore are British ex-colonies, have Chinese and English as two of their official languages, have common law origin and similar legal rights indexes, and were both affected by similar and potentially confounding events around the 1997 handover (e.g., the South East Asian financial crisis). We find that Hong Kong firms increased their cash holdings by about 2.6 percentage points (or 14.0% compared to the mean cash holdings of Hong Kong companies) in comparison with Singapore firms in the same industry and with similar firm characteristics. In column (4), we use only Malaysian firms as an alternative control group, but find an effect comparable to the estimate when using all firms affected by the South East Asian financial crisis. The difference-in-differences evidence involving Hong Kong companies complements the previously reported panel data evidence: An increase in the importance of a weak-FTR language in the business domain is associated with larger corporate cash holdings. Finally, we want to emphasize that if Hong Kong companies expected more uncertainty as a result of the transfer of sovereignty from the U.K. to China, they would increase their cash holdings already prior to the handover. In contrast, we find that firms in Hong Kong increased their corporate savings rate after the handover. 5.2. Within-country evidence One of the challenges of the previously reported panel data evidence is to appropriately control for measures of country culture. As a result, in this section, we report within-country evidence. 5.2.1. Weak- and strong-FTR languages within one country Switzerland has several official languages: German, French, Italian and Romansh. Three of them are classified as strong-FTR languages: French, Italian, and Romansh. Switzerland therefore provides an interesting opportunity to examine the effect of

21 There exist other, but even more challenging, events. For example, after the fall of the Wall in Europe in 1989, the significance of the Russian language diminished in the business domain in former Eastern European countries. There are several problems with an analysis of this event. First, it is difficult to obtain pre-1989 data on corporate financial decision-making for Eastern European firms. Second, this event coincides with dramatic changes of the economic systems in the affected countries. Finally, the Russian language was replaced by another strong-FTR language in several of the more market-oriented former Eastern European countries (e.g., Poland).

336

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

Table 10 Difference-in-differences evidence. (1)

(2)

(3)

(4)

Control group:

All non-Hong Kong firms

Firms affected by the 1997 Asian financial crisis

Singapore firms

Malaysia firms

Hong Kong × Post-1997-handover

0.047*** (0.005) 0.017*** (0.005) 0.162*** (0.008) 117,406 0.321 Yes Yes Yes Yes

0.037*** (0.006) −0.001 (0.006) 0.051* (0.031) 14,472 0.248 Yes Yes Yes Yes

0.026** (0.011) −0.009 (0.009) −0.067 (0.433) 4410 0.281 Yes Yes Yes No

0.038*** (0.010) 0.160*** (0.022) 0.028** (0.013) 6274 0.328 Yes Yes Yes No

Hong Kong Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls Country controls

The table reports regressions of corporate cash holdings in which we control for industry, firm, and country heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the 1989–2005 period. The dependent variable is Cash/Assets. Hong Kong is an indicator variable that is one for Hong Kong companies, i.e., firms that are headquartered in Hong Kong, and zero otherwise. Post-1997-Handover is an indicator variable that is one after the 1997 handover, and zero otherwise. The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 3. For columns (1) and (2), country controls is the full set of country-level characteristics included in Table 4. For columns (3) and (4), we do not include non-time-varying country controls. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

language within one country, holding country-specific institutional characteristics fixed. Specifically, we estimate the following model specification: yijt = b0 + b1 Weak − FTR Languageit + b2 Xit + 0t + 0j + 4ijt

(4)

where y is cash holdings, i indexes firms, j industries, and t years. X is a vector of time-varying firm control variables, 0 are sets of fixed effects, and 4 is an error term. We use the dominant language spoken in the region (canton) of a Swiss firm’s headquarters to classify a company as a weakFTR or strong-FTR language firm. As a result, in this within-country analysis, Weak-FTR Language is an indicator variable that is one if a firm is headquartered in a region where the dominant language is a weak-FTR language, and zero otherwise. While this measure is an imperfect approximation, we believe that it is more likely to be noisy than systematically biased in favor of our linguistic hypothesis. Table 11 reports the results. Column (1) shows that Swiss companies headquartered in regions that are dominated by weak-FTR languages hold significantly more cash compared to strong-FTR language companies in the same industry and with similar firm characteristics. The size of the point estimate (0.037) is comparable to what we found previously in our cross-country analysis. This analysis constitutes additional evidence that language explains heterogeneity in corporate financial decision-making. Table 11 Within-country evidence.

Weak-FTR language

(1)

(2)

Switzerland: Weak- and strong-FTR languages within one country

Canada: Several strong-FTR languages within one country

0.037** (0.013)

Placebo weak-FTR language Constant N Adjusted R-squared Year fixed effects Industry fixed effects Firm controls

0.185*** (0.058) 373 0.304 Yes Yes Yes

0.006 (0.004) 0.115*** (0.016) 8,217 0.340 Yes Yes Yes

The table reports regressions of corporate cash holdings in which we control for industry, firm, and country heterogeneity. Only Switzerland-based companies are included in the analysis in column (1). Only Canada-based companies are included in the analysis in column (2). The dependent variable is Cash/Assets. Weak-FTR language is an indicator variable that is one if a firm is headquartered in a region (Swiss canton) where the main language is a weak-FTR language, and zero otherwise. Placebo Weak-FTR Language is an indicator variable that is one if a firm is headquartered in a region (Canadian province) where English is the dominant language, and zero otherwise. The language classification is based on Chen (2013). The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in column (7) of Table 3. Standard errors are reported within parentheses and are heteroskedasticity-robust and clustered by region. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

337

5.2.2. Several strong-FTR languages within one country Canada has two official languages: English and French. Both of them are classified as strong-FTR languages, but in terms of the continuous language structure measures, English is somewhat weaker compared to French. Canada therefore provides an interesting opportunity to obtain placebo estimates of weak-FTR language. We therefore estimate the following model specification: yijt = b0 + b1 Placebo Weak − FTR Languageit + b2 Xit + 0t + 0j + 4ijt

(5)

where y is cash holdings, i indexes firms, j industries, and t years. X is a vector of time-varying firm control variables, 0 are sets of fixed effects, and 4 is an error term. We use the dominant language spoken in the region (province) of a Canadian firm’s headquarters to classify a company as a placebo weak-FTR language firm. As a result, in this within-country analysis, Placebo Weak-FTR Language is an indicator variable that is one if a firm is headquartered in a region where the slightly weaker among the strong-FTR languages is the dominant language, and zero otherwise. Column (2) reports that the size of the placebo weak-FTR language effect is only about 16.2% (= 0.006/0.037) of the size of the actual within-country weak-FTR language effect in the previous column. This placebo analysis constitutes additional support for our linguistic hypothesis. 5.3. Pre- versus post-financial crisis evidence Using different data sets and empirical approaches, we have so far reported evidence that supports our linguistic hypothesis, i.e., language explains heterogeneity in corporate savings behavior. This naturally raises the question of what may be the mechanism behind such an effect. In Section 2, we argued that language structure may systematically affect the formation of beliefs about the timing of an event (e.g., credit market freeze) that adversely affects a firm’s ability to access external finance. For example, language may affect time-distance beliefs. Speaking about future events in the present tense may result in weak-FTR language speakers perceiving adverse credit events are more imminent. A belief that an adverse credit event is more immediate may result in relatively larger precautionary cash holdings. The 2008 financial crisis increased the salience of an event that adversely affects credit markets. We examine whether firms in weak-FTR versus strong-FTR language countries responded differentially to the increase in the salience of a credit market freeze. We compare pre-crisis (2004–2006) and post-crisis (2010–2012) corporate savings behavior. We therefore estimate the following model specification: yijkt = b0 + b1 Weak − FTR Languageit + b2 Post − Crisist + b3 Weak − FTR Languageit × Post − Crisist + b4 Xit + b5 Xkt + 0j + 4ijkt

(6)

where y is cash holdings, i indexes firms, j industries, k countries, and t years. Post-Crisis is an indicator variable that is one for the 2010–2012 period, and zero otherwise. X are vectors of time-varying firm and country control variables, 0 are sets of fixed effects, and 4 is an error term. Table 12 reports our results. Column (1) re-estimates the previously reported model specification to verify that the relationship between language structure and corporate financial decision-making is found also for the analyzed sub-period. Column (2) shows that firms’ cash holdings on average are higher in the post-crisis subperiod. This finding is consistent with preexisting evidence (e.g., Song and Lee, 2012 and Pinkowitz et al., 2016). Importantly, column (3) shows that the post-crisis effect is significantly stronger among weak-FTR country companies. In other words, weak-FTR language firms responded more significantly to the financial crisis compared to strong-FTR language firms. The point estimate on the interaction effect Weak-FTR Language × Post-Crisis is positive and statistically significant at the 1%-level. We find that strong-FTR language firms increased their cash by 1.9 percentage points in the post-crisis period, while weak-FTR language firms increased their cash by an additional 2.5 percentage points. We control for companies’ operating cash flows; thus, our evidence is not simply driven by weak-FTR language firms being less adversely affected by the crisis. Finally, because the interaction effect, and not Weak-FTR Language per se, is the variable of main interest in this analysis, column (4) estimates an alternative model specification with country fixed effects, but the result does not change. The pre- versus post-financial crisis evidence is consistent with firms’ corporate cash holdings responding to a salient financial crisis event, and the response function being differential for weak- compared to strong-FTR language companies because future expected crises events appear relatively more vivid, immediate, and less distant in weak- compared to strong-FTR countries. This evidence suggests that language-induced heterogeneity in belief formation within firms is one mechanism which explains the reported relationship between language and corporate savings behavior. 6. Conclusion and discussion The idea that languages may affect decision-making has a long history in interdisciplinary research (e.g., von Humboldt, 1820 and Campbell, 2003). Indeed, in his classic On the Origin of Species, Darwin (1859) described language as a form of institutional “memory” that stores information about historical characteristics of a country’s culture in a “genome-like mode.” More

338

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341 Table 12 Pre- versus post-financial crisis evidence.

Weak-FTR language

(1)

(2)

(3)

(4)

0.041*** (0.014)

0.041*** (0.014) 0.020* (0.011)

0.041*** (0.015) 0.018** (0.007) 0.025*** (0.005) 0.057 (0.051) 60,844 0.303 Yes No Yes Yes

0.028*** (0.005) 0.026*** (0.003) 0.284* (0.153) 60,844 0.323 Yes Yes Yes Yes

Post-crisis Weak-FTR language × Post-crisis Constant N Adjusted R-squared Industry fixed effects Country fixed effects Firm controls Country controls

0.189*** (0.055) 60,844 0.300 Yes No Yes Yes

0.189*** (0.055) 60,844 0.300 Yes No Yes Yes

The table reports regressions of corporate cash holdings in which we control for industry, firm, and country heterogeneity. The data are public firms from Standard & Poor’s Compustat Global database and cover the pre-crisis (2004–2006) and post-crisis (2010–2012) periods. The dependent variable is Cash/Assets. Weak-FTR Language is an indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. The language classification is based on Chen (2013). Post-Crisis is an indicator variable which is one for the 2010–2012 period, and zero otherwise. The Appendix contains the definitions of all the variables. Firm controls is the full set of firm-level characteristics included in Table 3. Country controls is the full set of country-level characteristics included in Table 4 or the subset that is time-varying when country fixed effects are included. Standard errors are reported within parentheses and are White (1980) heteroskedasticity-robust and clustered by country. ***, **, * means that the point estimate is significantly different from zero at the 1%, 5%, and 10% levels, respectively.

specifically, the grammatical structure of a language may shape speakers’ representations of reality (e.g., Whorf et al., 1956 and Boroditsky, 2003). In recent years, research by economists has contributed new evidence regarding the effects of language on human decision-making (e.g., Chen, 2013 and Sutter et al., 2015). An important objective of this study has been to expand this pre-existing research by introducing a linguistic hypothesis into corporate finance research, and examine if language heterogeneity can explain differences in corporate decision-making. In this study, we have focused on one aspect of corporate financial decision-making, specifically the propensity of firms to hold cash, because it is the closest analogy to individual savings behavior. We first show that there is a strong positive correlation between individual and corporate savings propensities for some of the largest countries in the world. That is, corporations on average hold significantly more cash in countries where individuals save a larger proportion of their incomes. This may be attributable to differences in savings cultures across countries, with language heterogeneity partly being responsible for variation in savings behavior of individuals as well as corporations. We document empirically that language affects not only economic behaviors at the level of an individual, but such language effects carry over to behaviors at the corporate level. While studies of cultural effects on human behavior are currently a vibrant research direction in economics (e.g., Alesina and Giuliano, 2015 and Guiso et al., 2016), such analysis comes with a set of empirical challenges. We confront these challenges by reporting a collage of empirical evidence. We first report evidence from a variety of standard regression model specifications where we carefully address concerns about: i) inflation of sample observations by analyzing country averages and by re-estimating the models on collapsed data sets, ii) omitted variables bias by controlling for differences in industry-, firm-, and country-level heterogeneity, and iii) language evolution by controlling for relatedness caused by colonization, continents, or language families. In addition to this large-sample evidence we also report other evidence, in particular results from changes of the relative importance of different languages in the business domain in a country, evidence from within one country with several distinct languages, and results related to changes following a severe financial crisis. Studies in international corporate finance and governance are commonly subject to concerns about reverse causality. The idea that countries with specific savings cultures (because of some exogenous factor not caused by language) change language structures does not seem likely. For culture to affect language structure, it has to be changing slower than the resulting language changes. Empirically, future time reference in languages is very stable over time, indeed within the top 6% of the most stable language characteristics. In other words, a language’s FTR structure is a stable characteristic inherited from a very distant past that changes very slowly (e.g., Christiansen and Kirby, 2003), reducing endogeneity concerns. We want to emphasize several directions for future research related to language and corporate financial decision-making. First, we apply our linguistic hypothesis to corporate cash holdings, because it is the closest analogy to individual savings behavior. Researchers may also examine other corporate behaviors related to the future, e.g., investment and R&D policies (e.g., Liang et al., 2014). Second, there are several other important dimensions of language heterogeneity of relevance in international corporate finance research. For example, it has been well-established in linguistics research that languages have different ways of grammatically referencing gender. Heterogeneity in gender referencing may affect the proportion of females among corporate executives and on boards of directors (e.g., Santacreu-Vasut et al., 2014). Finally, because language structure measures may be expected to be exogenous to current economic outcomes, they may also be natural candidates to serve as instrumental variables in future research.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

339

Appendix A. Variable definitions Category/variable Cash holdings Cash/Assets Log (Cash/Sales) Language structure measures Weak-FTR language Firm characteristics Firm size Market-to-book ratio Cash Flow/Assets Cash flow volatility Capex/Assets Net Working Capital/Assets R&D/Sales Acquisitions/Assets Leverage Dividend indicator Investments Interest payments Debt issues Equity issues Dividends Profit Growth Tangibility R&D missing Industry Tax Liquidity RE TE ROA SGR Country characteristics GDP/Capita GDP growth Real interest rate French legal origin German legal origin Scandinavian legal origin Creditor rights index Shareholder rights index Legal rights index Catholic Protestant Muslim Buddhist Trust Corruption Loss of war Civil war Uncertainty avoidance Long-term orientation Political orientation Banking industry development Inflation Cinfo

Description

Cash and marketable securities, divided by total assets. Natural logarithm of cash and marketable securities, divided by total sales. An indicator variable which is one if a country’s or region’s dominant language is identified as a weak-FTR (“future time reference”) language, and zero otherwise. Data from . Log of book value of assets. Book value of assets minus book value of equity plus market value of equity, divided by book value of assets. Earnings after interest, dividends, and taxes but before depreciation, divided by total assets. Standard deviation of industry cash flow to total assets. Industries are classified using 2-digit SIC code. For each firm-year, we calculate the standard deviation of cash flow to assets over the past 10 years. Capital expenditures divided by total assets. Net working capital minus cash divided by total assets. R&D expenses divided by sales. The variable is set to zero if the value of R&D expenses is missing. Cash flows associated with acquisition divided by total assets. Long-term debt plus debt in current liabilities divided by total assets. An indicator variable that is one if a firms pays dividends, and zero otherwise. Capital expenditure divided by net property, plant and equipment at the beginning of the fiscal year. Total interest payments divided by total assets. Change of debt from fiscal year t-1 to fiscal year t divided by the average debt. Sales of common and preferred stock, net of purchases of common and preferred stock, divided by total assets. Total dividends divided by total shareholder equity. Earnings before interest and taxes divided by total assets Market value of equity divided by book value of equity. The ratio of net property, plant, and equipment to total assets. A dummy variable equals one if research and development expenses are unreported, zero otherwise. The median leverage ratio of the firm’s industry (two-digit SIC code). Total income taxes divided by pre-tax income. Current assets divided by current liabilities. Retained earnings delfated by total assets. Common equity divided by total assets. Net income divided by total assets. The logarithm of sales growth ratio. Gross domestic product (GDP) per capita. Data from the World Bank. Growth in GDP/Capita. Data from the World Bank. Government interest rate adjusted for inflation. Data from the World Bank, International Monetary Fund, and International Financial Statistics. Indicator variable that is one if a firm is from a country or a region with French legal origins. Data from Djankov et al. (2008). Indicator variable that is one if a firm is from a country or region with German legal origins. Data from Djankov et al. (2008). Indicator variable that is one if a firm is from a country or region with Scandinavian legal origins. Data from Djankov et al. (2008). Creditor rights index. Data from Djankov et al. (2007). Shareholder rights index. Data from Djankov, Djankov et al. (2008). Index to measure creditor and lender protection. Data from the World Bank. Indicator variable that is one if >50% of the inhabitants in a country or region are Catholics. Data from The World Factbook. Indicator variable that is one if>50% of the inhabitants in a country or region are Protestants. Data from The World Factbook. Indicator variable that is one if >50% of the inhabitants in a country or region are Muslims. Data from The World Factbook. Indicator variable that is one if >50% of the inhabitants in a country or region are Buddhists. Data from The World Factbook. Trust index. Data are from La Porta et al. (1997a,b). Corruption perception index. A higher value implies less corruption. Data from Transparency International. Number of times a country or region lost a war from 1900 to a sample year (divided by 100). Data from Correlate of War Project, Polynational War Memorial, and Uppsala Conflict Data Program. Number of years a country or region was in a civil war from 1900 to a sample year (divided by 100). Data from Correlate of War Project, Polynational War Memorial, and Uppsala Conflict Data Program. Index to measure the uncomfortability level a society may have towards uncertainty and ambiguity. Data from Hofstede’s website. Index to measure whether a society encourages and makes efforts in preparing for the future. Data from Hofstede’s website. A right government has a value of 1, a center government a value of 2, and a left government a value of 3. Data from the World Bank and Beck et al. (2001). Number of depositors scaled by adult population (multiplied by 100,000). Data from Financial Access Survey (FAS). Inflation rate per year. Data from the World Bank. A dummy variable equals one if a public credit registry operates in the country, 0 otherwise. A public registry is defined as a database owned by public authorities (usually the Central Bank or Banking Supervisory Authority) that collects information on the standing of borrowers in the financial system and provides it to financial institutions.

The Appendix contains the definitions of all the variables analyzed in this study. We obtain these data from Standard & Poor’s Compustat Global and CapitalIQ databases unless otherwise stated.

340

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

References Acemo˘glu, D., Johnson, S., Robinson, J.A., 2001. The colonial origins of comparative development: an empirical investigation. Am. Econ. Rev. 91 (5), 1369–1401. Ahern, K.R., Daminelli, D., Fracassi, C., 2015. Lost in translation? The effect of cultural values on mergers around the world. J. Financ. Econ. 117, 165–189. Alesina, A., Giuliano, P., 2015. Culture and institutions. J. Econ. Lit. 53 (4), 898–944. Almeida, H., Campello, M., Cunha, I., Weisbach, M.S., 2014. Corporate liquidity management: a conceptual framework and survey. Annu. Rev. Financ. Econ. 6 (1), 135–162. Almeida, H., Campello, M., Weisbach, M.S., 2004. The cash flow sensitivity of cash. J. Financ. 59 (4), 1777–1804. Aronoff, M., Rees-Miller, J., 2003. The Handbook of Linguistics. Blackwell Publishers, Oxford, U.K.. Bates, T.W., Kahle, K.M., Stulz, R.M., 2009. Why do U.S. firms hold so much more cash than they used to? J. Financ. 64 (5), 1985–2021. Beck, T., Clarke, G., Groff, A., Keefer, P., Walsh, P., 2001. New tools in comparative political economy: the database of political institutions. World Bank Econ. Rev. 15 (1), 165–176. Berger, P.G., Ofek, E., Yermack, D.L., 1997. Managerial entrenchment and capital structure decisions. J. Financ. 52 (4), 1411–1438. Boroditsky, L., 2001. Does language shape thought? Mandarin and English speakers’ conceptions of time. Cogn. Psychol. 43 (1), 1–22. Boroditsky, L., 2003. Linguistic relativity. In: Nadel, L. (Ed.), Encyclopedia of Cognitive Science. MacMillan Press, London, U.K.. Brannen, M.Y., Piekkari, R., Tietze, S., 2014. The multifaceted role of language in international business: unpacking the forms, functions and features of a critical challenge to MNC theory and performance. J. Int. Bus. Stud. 45 (5), 495–507. Campbell, L., 2003. The history of linguistics. In: Aronoff, M., Rees-Miller, J. (Eds.), The Handbook of Linguistics. Blackwell Publishers, Oxford, U.K.. Carroll, C.D., Rhee, B.-K., Rhee, C., 1994. Are there cultural effects on saving? Some cross-sectional evidence. Q. J. Econ. 109 (3), 685–699. Carroll, C.D., Rhee, B.-K., Rhee, C., 1999. Does cultural origin affect saving behavior? Evidence from immigrants. Econ. Dev. Cult. Chang. 48 (1), 33–50. Cavalli-Sforza, L.L., 2001. Genes, Peoples, and Languages. University of California Press, Berkeley, CA. Chang, Y.-C., Hong, H.G., Tiedens, L., Wang, N., Zhao, B., 2015. Does diversity lead to diverse opinions? Evidence from languages and stock markets. Working paper. Stanford Graduate School of Business., Chen, M.K., 2013. The effect of language on economic behavior: evidence from savings rates, health behaviors, and retirement assets. Am. Econ. Rev. 103 (2), 690–731. Chomsky, N., 1957. Syntactic Structures. Mouton, Paris, France. Christiansen, M.H., Kirby, S., 2003. Language Evolution. Oxford University Press, Oxford, U.K.. Dahl, Ö., 1985. Tense and Aspect Systems. Blackwell Publishers, Oxford, U.K.. Dahl, Ö., 2000. The grammar of future time reference in European languages. Tense and Aspect in the Languages of Europe, (6). , pp. 309–328. Darwin, C.R., 1859. On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life. John Murray Publishers, London, U.K. Dittmar, A., Duchin, R., 2016. Looking in the rearview mirror: the effect of managers’ professional experience on corporate financial policy. Rev. Financ. Stud. 29 (3), 565–602. Dittmar, A., Mahrt-Smith, J., 2007. Corporate governance and the value of cash holdings. J. Financ. Econ. 83 (3), 599–634. Dittmar, A., Mahrt-Smith, J., Servaes, H., 2003. International corporate governance and corporate cash holdings. J. Financ. Quant. Anal. 38 (1), 111–133. Djankov, S., La Porta, R., Lopez-de Silanes, F., Shleifer, A., 2008. The law and economics of self-dealing. J. Financ. Econ. 88 (3), 430–465. Djankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 countries. J. Financ. Econ. 84 (2), 299–329. Dryer, M.S., 1989. Large linguistic areas and language sampling. Stud. Lang. 13 (2), 257–292. Evans, N., Levinson, S.C., 2009. The myth of language universals: language diversity and its importance for cognitive science. Behav. Brain Sci. 32 (5), 429–448. Gao, H., Harford, J., Li, K., 2013. Determinants of corporate cash policy: insights from private firms. J. Financ. Econ. 109 (3), 623–639. Ginsburgh, V.A., Weber, S., 2014. Culture, linguistic diversity, and economics. In: Ginsburgh, V.A., Throsby, D. (Eds.), Handbook of the Economics of Art and Culture. Elsevier, The Netherlands. Graham, J.R., Harvey, C.R., Puri, M., 2013. Managerial attitudes and corporate actions. J. Financ. Econ. 109 (1), 103–121. Guiso, L., Sapienza, P., Zingales, L., 2004. The role of social capital in financial development. Am. Econ. Rev. 94 (3), 526–556. Guiso, L., Sapienza, P., Zingales, L., 2006. Does culture affect economic outcomes? J. Econ. Perspect. 20 (2), 23–48. Guiso, L., Sapienza, P., Zingales, L., 2015. Corporate culture, societal culture, and institutions. Am. Econ. Rev. 105 (5), 336–339. Guiso, L., Sapienza, P., Zingales, L., 2016. Long-term persistence. J. Eur. Econ. Assoc. 14 (6), 1401–1436. Han, S., Qiu, J., 2007. Corporate precautionary cash holdings. J. Corp. Finan. 13 (1), 43–57. Harford, J., Mansi, S.A., Maxwell, W.F., 2008. Corporate governance and firm cash holdings in the U.S.. J. Financ. Econ. 87 (3), 535–555. Hilary, G., Hui, K.W., 2009. Does religion matter in corporate decision making in America? J. Financ. Econ. 93 (3), 455–473. Hofstede, G., 1980. Culture’s Consequences: International Differences in Work-Related Values. Sage Publications, New York. Hubbard, R., Skinner, J., Zeldes, S., 1995. Precautionary saving and social insurance. J. Polit. Econ. 360–399. Hutton, I., Jiang, D., Kumar, A., 2014. Corporate policies of Republican managers. J. Financ. Quant. Anal. 49 (5–6), 1279–1310. Jakobson, R., Halle, M., 1956. Fundamentals of Language. Mouton Publishers, Netherlands. Johansson, S., 2005. Origins of Language: Constraints on Hypotheses. vol. 5. John Benjamins Publishing. Kalcheva, I., Lins, K.V., 2007. International evidence on cash holdings and expected managerial agency problems. Rev. Financ. Stud. 20 (4), 1087–1112. Keynes, J.M., 1934. The Applied Theory of Money. Macmillan Publishers, London, U.K.. Khurana, I.K., Martin, X., Pereira, R., 2006. Financial development and the cash flow sensitivity of cash. J. Financ. Quant. Anal. 41 (4), 787–808. Kim, J., Kim, Y., Zhou, J., 2017. Languages and earnings management. J. Account. Econ. 63 (2-3), 288–306. Kimball, M.S., 1990. Precautionary saving in the small and in the large. Econometrica 58 (1), 53–73. Klerman, D.M., Mahoney, P.G., Spamann, H., Weinstein, M.I., 2011. Legal origin or colonial history? J. Leg. Anal. 3 (2), 379–409. Kramsch, C., 1998. Language and Culture. vol. 3. Oxford University Press, Oxford, U.K.. La Porta, R., Lopez-de Silanes, F., Shleifer, A., 2008. The economic consequences of legal origins. J. Econ. Lit. 46 (2), 285–332. La Porta, R., Lopez-de Silanes, F., Shleifer, A., Vishny, R.W., 1997a. Legal determinants of external finance. J. Financ. 52 (3), 1131–1150. La Porta, R., Lopez-de Silanes, F., Shleifer, A., Vishny, R.W., 1997b. Trust in large organizations. Am. Econ. Rev. 87 (2), 333–338. La Porta, R., Lopez-de Silanes, F., Shleifer, A., Vishny, R.W., 1998. Law and finance. J. Polit. Econ. 106 (6), 1113–1155. Lewis, M.P., 2009. Ethnologue: Languages of the World. SIL International, Dallas. Liang, H., Marquis, C., Renneboog, L., Sun, S.L., 2014. Speaking of corporate social responsibility. Working paper. Harvard Business School., Lins, K.V., Servaes, H., Tufano, P., 2010. What drives corporate liquidity? An international survey of cash holdings and lines of credit. J. Financ. Econ. 98 (1), 160–176. Lusardi, A., 1998. On the importance of the precautionary saving motive. Am. Econ. Rev. 88 (2), 449–453. Modigliani, F., Miller, M.H., 1958. The cost of capital, corporation finance and the theory of investment. Am. Econ. Rev. 48 (3), 261–297. Opler, T.C., Pinkowitz, L., Stulz, R.M., Williamson, R., 1999. The determinants and implications of corporate cash holdings. J. Financ. Econ. 52 (1), 3–46. Petersen, M.A., 2009. Estimating standard errors in finance panel data sets: comparing approaches. Rev. Financ. Stud. 22 (1), 435–480. Pinker, S., 1994. The Language Instinct: How the Mind Creates Language. Harper, New York, NY. Pinkowitz, L., Stulz, R.M., Williamson, R., 2006. Does the contribution of corporate cash holdings and dividends to firm value depend on governance? A cross-country analysis. J. Financ. 61 (6), 2725–2751. Pinkowitz, L., Stulz, R.M., Williamson, R., 2016. Do U.S. firms hold more cash than foreign firms do? Rev. Financ. Stud. 29 (2), 309–348. Roberts, S.G., Winters, J., Chen, M.K., 2015. Future tense and economic decisions: controlling for cultural evolution. PLoS ONE 10 (7), 1–46.

S. Chen et al. / Journal of Corporate Finance 46 (2017) 320–341

341

Santacreu-Vasut, E., Shenkar, O., Shoham, A., 2014. Linguistic gender marking and its international business ramifications. J. Int. Bus. Stud. 45, 1170–1178. Schiffrin, D., 1981. Tense variation in narrative. Language 57 (1), 45–62. Slobin, D.I., 1987. Thinking for speaking. Proceedings of the Berkeley Linguistics Society. vol. 3. pp. 435–445. So, D.W.C., 1996. Hong Kong. In: Dickson, P., Cumming, A. (Eds.), Profiles of Language Education in 25 Countries. National Foundation for Educational Research, U.K.. Song, K.R., Lee, Y., 2012. Long-term effects of a financial crisis: evidence from cash holdings of east Asian firms. J. Financ. Quant. Anal. 47 (3), 617–641. Stulz, R.M., Williamson, R., 2003. Culture, openness, and finance. J. Financ. Econ. 70 (3), 313–349. Sutter, M., Angerer, S., Glätzle-Rützler, D., Lergetporer, P., 2015. The effect of language on economic behavior: experimental evidence from children’s intertemporal choices. IZA Discussion Paper. Tabellini, G., 2008. Presidential address to the European Economic Association: Institutions and culture. J. Eur. Econ. Assoc. 6 (2–3), 255–294. Thieroff, R., 2000. On the areal distribution of tense-aspect categories in Europe. Empir. Approaches Lang. Typol. (6), 265–308. von Humboldt, W.F., 1820. Über das vergleichende Sprachstudium in Beziehung auf die verschiedenen Epochen der Sprachentwicklung. White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48, 817–838. Whorf, B.L., Carroll, J.B., Chase, S., 1956. Language, Thought and Reality: Selected Writings of Benjamin Lee Whorf. MIT Press, Cambridge. Yonker, S.E., 2017. Geography and the market for CEOs. Manag. Sci. 63, 609–630.