Corporate Culture and Innovation: A Tale from an Emerging Market

We examine the influence of corporate culture on innovation using a unique set of data from Chinese listed companies over the period 2008-2017. Using the Competing Value Framework, we quantify corporate culture using textual analysis of financial statements. We find a positive and significant impact of a creation culture on innovation measured by both patent applications and citations as well as innovation efficiency. We address endogeneity concerns and conduct a battery of robustness tests, including alternative proxies for both corporate innovation and culture and conclude that variations in culture have a significant impact on firm-level innovation. We also show that a strong creation culture is more likely to spur innovation for firms in more competitive product markets and firms that are subject to higher managerial career concerns. We provide empirical evidence that corporate culture is an important driver in enhancing innovation. Our results have clear implications for directors and shareholders.


Introduction
Innovation has been considered a vital factor for organisations that aim to achieve and sustain a competitive advantage (Baer, 2012;Nelson and Winter, 1982;Porter, 1992).
Undoubtedly, companies with better innovative capabilities can enhance their profitability (Atalay, Anafarta and Sarvan, 2013;Tuan et al., 2016) and can lead to a country's economic development (Solow, 1957). Therefore, understanding the drivers of a firm's innovative ability is critical from both academic and industry perspectives.
However, innovative projects are often risky, unpredictable, lengthy and involve significant resources (Holmstrom, 1989). Therefore, stimulating innovation needs a strong organisational culture that encourages risk-taking, tolerance for failure in the short-term and rewards long-term success (Manso, 2011). Culture is among the primary levers at managers' disposal to maintain organisational viability in the wake of high uncertainty associated with innovation. Several studies, intuitively, argue that organisational culture is a key to innovation success and corporate performance (see Büschgens, Bausch and Balkin, 2013;and Sackmann, 2002 for review). The evidence on the nexus between culture and innovation is based on small sample surveys calling into question the generalisability and comparability of the observed findings. Large sample studies beyond developed markets  are even scant.
Corporate culture refers to the values and beliefs that provide norms for expected behaviour by employees (Schein, 1992) and organisations actively promote these values and principles (Guiso, Sapienza and Zingales, 2015). Culture is considered a fundamental asset to boost performance 1 and a key value driver by business managers (Graham et al., 2017). Anecdotal evidence from regulatory reviews and business leaders suggests that the culture of an organisation is set at the top management levels (CEO and Board) 2 and this view is supported by the academic literature (Kotter and 1 Global Human Capital Trends 2016, The new Organisation: Different by design, Deloitte University Press, Available at https://www2.deloitte.com/content/dam/Deloitte/global/Documents/HumanCapital/gx-dup-globalhuman-capital-trends-2016.pdf 2 For instance, a UK Financial Reporting Council (2016) report on "Corporate Culture and The Role of Boards" observes that " it is the board's role to determine the purpose of the company and ensure that the company's values, strategy and business model are aligned to it". The report further adds that the top leaders, particularly Chief Executive, "must embody the desired culture, embedding this at all levels and in every aspect of the business". Similarly, UK Corporate Governance Code in its preface (supporting principle A.1) states that "one of the key roles for the board includes establishing the culture, values and ethics of the company. It is important that the board sets the correct "tone from the top". The directors should lead by example and ensure that good standards of behaviour permeate throughout all levels of Heskett, 1992). Stakeholders, on the other hand, play a remarkable role in shaping and influencing corporate culture particularly in post Global Financial Crisis era. This resulted in an ever-increased emphasis on not only promoting the ethical and professional standards through governance reform but also on promoting green innovations for sustainable development. Therefore, understanding the drivers of a firm's innovative ability is critical for sustainable developments in an ever-challenging institutional environment.
The competitive conditions in the global business environment demand firms to pursue growth through innovation and creativity for long term sustainability. Such a strategy requires a consistent and continuous reliance on creative behaviour across all levels within the organisation (Bednall et al., 2018). Creative culture affects innovation as it shapes the patterns of novelty, individual initiatives and risk-taking behaviours (Kaasa and Vadi, 2010). Moreover, it promotes agility, adaptability and creativity.
Organisations with a rich creative culture are likely to be more innovative (Anderson, Potočnik and Zhou, 2014). 3 We test the culture-innovation relationship on a unique dataset of Chinese listed firms. Our firm-level measure of creation culture is based on the Competing Value Framework (Cameron et al., 2006). We quantify CVF culture dimensions (create, compete, control and collaborate) using text analysis of financial reports and measure the intensity of creation culture in an organisation following Andreou et al. (2019). The basic proposition of text analysis is that words and language used in the official documents reveal the culture of the organisation (Levinson, 2003). Our choice of Chinese market is motivated by two main reasons. First, China is the largest emerging market and second largest economy in the world. A large part of China's rapid the organisation. This will help prevent misconduct, unethical practices and support the delivery of longterm success" (FRC, 2016). Greg Medcraft, Chairman, Australian Securities and Investments Commission (ASIC), states that the role of board is absolutely critical in setting the culture and tone of the organisation and one of the ways in which board sets this tone is by selecting a CEO whose values are aligned with the a company's desired culture (Medcraft, 2016). Business leaders have similar views as above regarding the role of boards in shaping the corporate culture. For example, John Watson Chairman Bellway states that "The Board sets the values and culture (how people treat each other, how they operate within the supply chain and how they work with employees) and management is then responsible for implementing this" (FRC,2016). 3 Graham et al., (2016) report that more than 50 percent of CEOs and CFOs in their survey identified culture as an influential factor for creativity. economic growth could be attributed to the innovative capability of Chinese firms. 4 China provides an interesting setting to examine the role of organisational culture in fostering innovation outside developed markets. Our findings may have implications for other emerging economies competing for dominance in the global market (Fana, Weib and Xuc, 2011). Second, there are significant institutional and cultural differences between the Chinese and developed markets (Schneider, 1988). Our study provides a first-hand test of the generalisability, comparability and applicability of the CVF in an emerging market with significant differences in corporate governance regime compared to the developed markets.
We find overwhelming evidence that creation culture has a positive effect on firmlevel innovation for a large sample of Chinese listed firms. Our results are robust to a number of robustness tests including alternative measures of innovation output, creation culture and endogeneity. We also find that the effect of culture on innovation is more pronounced for firms that operate in more competitive industries and are subject to higher managerial career concerns. Our findings imply a strong cultural effect on innovation.
We make several contributions to the literature. First, we extend recent insights on the role of culture and innovation outside the developed world .

Conceptualisation of Corporate Culture
There is no universally accepted definition of corporate culture and culture seems to be denoted by a variety of meanings and connotations in organisational behaviour, economics, and finance literature . Based on organisational behaviour literature, O'Reilly and Chatman (1996) define corporate culture as "a set of norms and values that are widely shared and strongly help throughout the organization". This view suggests corporate culture as a social control mechanism and is in line with Hofstede (1991) who labels corporate culture as "the collective programming of the mind" of people in an organisation. Culture acts as a coordinating mechanism to enable dealing with unforeseen contingencies and events (Kreps, 1996).
Every culture contains unspoken beliefs and norms that impact people's behaviour. Schein (2010) ascribes culture as a set of basic assumptions among people to evaluate situations, human relations and activities, and these assumptions form the shared norms and beliefs among a group of people. This concept of culture is also consistent with Martin (1992) view of integration perspective, which is based on homogeneity and harmony holding together a diverse group of people in an organisation. The literature in economics and finance has viewed culture as a mechanism of explicit and implicit contracts that governs behaviour and comprises of shared values, preferences and beliefs of individuals in an organisation (Bé nabou and Tirole, 2002;Guiso, Sapienza and Zingales, 2015;Van den Steen, 2010). Culture facilitates the complex interactions within the organisations as contracts for each relationship do not exist (Audi, Loughran and McDonald, 2016). Guiso, Sapienza and Zingales (2015) define culture as "principles and values that should inform the behavior of all the firm's employees". In 2013), stock liquidity (Fang, Tian and Tice, 2014) and anti-takeover provisions (Chemmanur and Tian, 2018) as the important determinants of firm-level innovation ability.
sum, culture represents rules based on common values and beliefs which determine the interactions within an organisation in order to accomplish certain goals.

The Competing Value Framework
For our definition of culture, we rely on CVF (Cameron et al., 2006), an organisational taxonomy widely used in literature (Hartnell, Ou and Kinicki, 2011 The literatuer has traditionally relied on survey-based instruments using small samples to test CVF propositions. More recently, Fiordelisi and Ricci (2014) operationalise and quantify the CVF culture types by employing a bag of words approach and textual analysis of financial reports (10-K reports). The underlying premise of the textual analysis is that the words used in corporate filings depict the cultural norms of a company. Financial reports are the most important documents which could be used by the managers to present major attributes and values of the company to the outside world (Audi, Loughran and McDonald, 2016). The CVF based culture measure is broadly supported by the growing literature within finance and accounting which provides evidence that textual analysis can be used to extract important aspects of corporate behaviour, tone and sentiment in publicly available official documents (Hanley and Hoberg, 2010;Hoberg and Phillips, 2010;Li, 2008;Loughran and McDonald, 2011). This, however, raises a question that whether text-based measures reliably capture corporate culture. The relevant literature suggests that textual analysis of financial statements provides valid and robust measures of industry competition (Li, Lundholm and Minnis, 2013), banking competition (Bushman, Hendricks and Williams, 2016) and financial constraints (Hoberg and Maksimovic, 2014) and these measures reliably predict firm financial outcomes. Recent studies also provide evidence on the validity of text-based corporate (Fiordelisi and Ricci, 2014;Nguyen, Nguyen and Sila, 2019) and creative  culture measures, firms' market orientation  and firms' thrust to compete (Andreou et al., 2019).
Overall, the evidence suggests that CVF based text measures reliably captures firms' corporate culture.

Corporate Culture and Innovation
Culture is manifested in rituals, group norms, habits of thinking and espoused values (Schein, 1992). These attributes, in turn, help build a strong sense of ownership, teamwork and sense of mission among employees (Fey and Denison, 2003).
Organisational cultures influence employee attitudes, behaviours and commitment leading to organisational effectiveness (Gregory et al., 2009;O'Reilly, Chatman and Caldwell, 1991). Culture, as such, can be a powerful means to elicit desired corporate outcomes (Hogan and Coote, 2014). Thus, a large body of literature suggests a positive association between culture and performance (Denison and Mishra, 1995;Sackmann, 2002;Zuckerman, 2002). Innovation is one of the key drivers of business performance and value creation. Innovation confers a competitive advantage that helps successful organizations to stay ahead in business. Successful innovation determines a firm's future profitability and competitive edge to a large extent (Ettlie, 1998;Scherer, 1984).  (Cameron et al. (2006) Given the importance of innovation for an organisation's success and performance, previous literature posits that culture plays an important role in enhancing innovation.
The extant literature has identified a variety of cultural values related to innovation.
For instance, innovation-supportive cultures and transformational leadership are likely to enhance innovation (Chandler, Keller and Lyon, 2000;Gumusluoğlu and Ilsev, 2009). Organisations with supportive cultures, participative decision making, and higher tolerance-for-failure are found to be more productive at innovation (Abbey and Dickson, 1983;Danneels, 2008;Hurley and Hult, 1998). While it is widely accepted that culture fosters innovation, some aspects of culture could stifle innovation (Berson, Oreg and Dvir, 2008;Chatman and Flynn, 2001;Jaskyte, 2004). Sørensen (2002) argues that strong culture firms may find it difficult to engage in exploration learning and adapt to changes in the volatile environment. Dougherty and Heller (1994) suggest that organisational preference for stability may impede product innovation. Similarly, Staw and Nemeth (1989) argue that strong culture may stifle innovation as cohesion among organisational members leads to less deviation. Büschgens, Bausch and Balkin (2013), however, argue that the use of a multitude of culture values leads to a fragmented concept of culture and innovation. The authors show that CVF comprehensively describes corporate cultures. We use CVF and measure the strength of creation culture relative to other culture dimensions. The creation oriented (adhocracy) culture in Competing Value Framework (Cameron et al., 2006) suggests that such a culture would encourage idea-sharing, entrepreneurial thinking and vision building among the employees. This type of culture promotes agility, adaptability and creativity, which are key drivers of innovation. The flexibility orientation of create dimension encourages acceptance of deviation from norms and allows tolerance-for-failure and greater risk-taking (Büschgens, Bausch and Balkin, 2013;Danneels, 2008). By allowing freedom of thought and action among employees, companies rich in creative culture aim to develop radical new processes and product technologies, innovate in logistics and redefine entire industries (Fiordelisi and Ricci, 2014;Fiordelisi et al., 2019). Our primary hypothesis is, therefore; H1: Firms with a stronger creation-oriented culture are more likely to engage in innovation.

Sample
Our sample includes all Chinese listed companies on both the Shanghai and Shenzhen Stock Exchanges (SHSE, SZSE) between 2008 and 2017. 6 Data on firm-level innovation and accounting variables are collected from the China Stock Market & Accounting Research (CSMAR) database. We also match innovation variables (patents and citations) in CSMAR with the China National Intellectual Property Administration (CNIPA) 7 to ensure that our innovation variables are consistent and accurate. In case of conflict between CSMAR and CNIPA, we use the patent data from CNIPA. We use annual reports of Chinese listed firms to construct culture related variables using textual analysis. We exclude financial firms and firms with missing values from our sample following Yuan and Wen (2018). All variables are winsorized at the 1 st and 99 th percentiles to mitigate the influence of outliers.

Innovation measures
We construct innovation variables using data from CSMAR and CNIPA which contain information on patent application date, application identification, grant date and application institution. The Chinese patents are classified into three categories: invention patents, utility model patents and design patents. Invention patents are granted for a new solution to a product or technical process. Utility model patents are for a new technical solution or improvement with lower degree of inventiveness relating to certain features of a product such as shape or structural physical features. Design patents are granted for innovations in external features of a product such as shape, pattern and/or colour, which make the product attractive and fit for industrial application. Given the limited technological innovation involved in design patents, we use only invention and utility model patents to construct our innovation measures following relevant literature (Fang, Lerner and Wu, 2017;Jiang and Yuan, 2018;Tan et al., 2015). Consistent with innovation literature, we use application year rather than grant year for our patent counts because applicant year corresponds more closely to the 6 We chose 2008 as our starting year since the Chinese government adopted a new accounting standard in 2007 that required disclosure of R&D information. 7 See http://english.sipo.gov.cn/ for more details actual time of the innovation (Fang, Tian and Tice, 2014;He and Tian, 2013;Jiang and Yuan, 2018).
Our firm-level innovation output is measured in four ways. First, Inven, is the raw count of invention patents filed and eventually granted to a firm in a given year. Second, Inven+Utility, is the raw count of invention and utility patents filed and eventually granted to a firm in a given year. Third, Innovation Efficiency, is a measure of how efficiently a firm transforms its innovation input (R& D expenditure) into innovation output (patents). We follow Cao, Cumming and Zhou (2020) and Hirshleifer, Hsu and Li (2013) to construct innovation efficiency, which is measured as the number of patents of a firm i in year t scaled by firm i's cumulative R&D investment in year t-2 through year t. 8 Fourth, Citations, is the number of forward citations received by a firm's invention patents. While patent counts measure the raw output of a firm's innovative activities, citations capture the technological and economic importance of patents granted to a firm (Hall, Jaffe and Trajtenberg, 2005). We use natural logarithm of one plus patent counts (Inven, Inven+Utility) and Citations in our analysis to address the skewness concerns in patent-related variables (Fang, Lerner and Wu, 2017;Jiang and Yuan, 2018).

Creation culture measure
A firm's culture is manifested in its symbols, rituals and values, and that "symbols are words, gestures, pictures or objects that carry a particular meaning within a culture" (Hofstede et al., 1990). Thus, words can be used to convey the values and attributes that are part of a firm's culture. The management's disclosures in financial documents can provide insights into the value system of the firm as the distinctive features of any firm are mirrored in its written documents (Fiordelisi and Ricci, 2014). Text analysis is a systematic and objective means to examine the specific features of a text and is widely supported from research in sociology (Atkinson, 1990), communication theory (Mumby and Stohl, 1991), cultural anthropology (Clifford and Marcus, 1986) and lately from finance and management literature Antweiler and Frank, 2004;Hanley and Hoberg, 2010;Hoberg and Phillips, 2010;Li, 2008;8 Innovation efficiency is measured following Cao, Cumming and Zhou (2020). The equation for the Innovation Efficiency as follow: Loughran and McDonald, 2011;Nguyen, Nguyen and Sila, 2019). Fiordelisi and Ricci (2014) apply text analysis on 10-K filings in US to measure CVF culture dimensions using a two-step procedure. First, they identify words related to each culture dimension in CVF using Cameron et al. (2006). Second, they use Harvard IV-4 Psychosocial Dictionary to identify synonyms for words selected in first step to create four bags of words for each culture dimension listed in panel A of Table A.2 (Appendix).
We operationalise CVF culture dimensions by applying text analysis to annual reports of Chinese listed firms following bag of words developed by Fiordelisi and Ricci (2014). Specifically, we translate authors' bag of words into Chinese using The Oxford English-Chinese Dictionary and The Oxford English Dictionary following Jiang et al. (2017). 9 We further validate our translated keywords for their appropriateness in the Chinese context by using The Contemporary Chinese Dictionary and only choose the first explanation of each word in the dictionary to avoid double-counting. To further authenticate our approach, we use external verification of our approach from an expert Chinese linguist. 10 We quantify each CVF culture dimension, using Chinese bag of words, by counting the occurrences of keywords (synonyms) associated with that culture in each annual report. We obtain four values for each of the culture dimensions for our firm-years. Using raw counts of each CVF culture dimension, we construct our Creation Culture variable as follows 11 ; We scale the frequency of creation culture words by the total number of words in all culture dimension to capture the relative emphasis or intensity of the creation culture compared to other cultures. Our measure reflects the relative importance that firms place on their creative values leading to a creation culture. We also use an alternative approach where the creation culture keyword count is scaled by the total number of words in the annual report (Creation Culture-Alternative) consistent with prior 9 The Chinese "Bag of words" can be found in panel B of Table A.2 (Appendix). 10 We thank and appreciate Professor Zhiyang Sun, a linguist from Jiangnan University, for his help in authenticating our approach. 11 Andreou et al.(2019) use similar approach for constructing their thrust to compete measure. literature (Fiordelisi and Ricci, 2014;Nguyen, Nguyen and Sila, 2019). The higher ratio indicates a higher emphasis on the creation culture in a firm. 12 Although, our measurement of creation culture follows the extant literature (Fiordelisi and Ricci, 2014;Andreou et al., 2019), there might be concerns regarding the applicability of CVF in the Chinese context. The CVF has been developed and applied in western countries (predominantly USA) and might not be applicable to Chinese firms due to institutional differences between China and the developed countries (You, Zhang and Zhang, 2017). The relevant literature, however, suggests the validity and applicability of the CVF model across different countries. For instance, Mujen and Turnipseed (1999) apply CVF model to construct an international instrument for measuring organisational culture across 12 European countries. Other studies provide validity of CVF for Hong Kong (Lau and Ngo, 2004;Kwan and Walker, 2004), South Korea (Dastmalchian, Lee and Ng, 2000), Australia (Lamond, 2003), China and other Asian countries (Deshpandé and Farley, 2004;Lau, David and Zhou, 2002;Ralston et al., 2006). More recently, Zhang, Li and Wei (2008) use CVF to investigate the relationship between corporate culture and performance of Chinese firms and Yu and Wu (2009) argue that CVF provides better validity and reliability in the Chinese context. Overall, CVF is a widely applicable model for corporate culture across different countries and institutional contexts including China, and the Western origin of the CVF is not a weakness of the model (Büschgens, Bausch and Balkin, 2013).

Firm controls
Following innovation literature, we control for an array of well-known firm-level characteristics that may affect firms' innovation. We include firm size following Hall and Ziedonis (2001), who argue a positive relation between firm size and the number of patents and citations. Firm age is included to control for differences in the stages of development across firms (Amore and Failla, 2018). We also include leverage and return on assets (ROA) to control firms' debt and profitability (He and Tian, 2013;Mazouz and Zhao, 2019). We also include book-to-market ratio to account for growth opportunities (Lev and Sougiannis, 1999) and state ownership (SOE) following Cao, Cumming and Zhou (2020). A detailed definition of these control variables is provided 12 The CVF argues that although all culture dimensions are typically present in an organisation, one or two typically are dominant.
in Table A.1 (Appendix). In robustness tests, we also include a number of board and CEO characteristics as additional controls.

Empirical Model
We use following model to estimate the relationship between corporate culture and innovation: where i indexes firms and t indexes time. The innovation variables are the natural logarithm of one plus the number of invention patents filed (and subsequently granted), the natural logarithm of one plus the number of invention and utility patents, innovation efficiency and the natural logarithm of one plus the number of citations per invention patent. Our innovation variables account for both the quantity (patent numbers) and quality (citations) of the innovation. The key explanatory variable is our creation culture measured as in equation (1). Firm Controls, include firm size, firm age, leverage ratio, ROA, book to market ratio and SOE. Finally, we include industry and province fixed effects to control industry and province specific time-invariant heterogeneity and year fixed effects to control overall trends in innovation. Robust standard errors are clustered at firm level.

Results and Discussion
Summary statistics Consistent with our prediction, these industries also rank higher in terms of innovation outputs. Although there is a substantial variation in creation culture across industries, the variation across sample years is small (Panel B). This is consistent with the general notion that culture is persistent over time and changes slowly .
Panel C shows that our sample firms apply, on average, 8.03 invention patents each year which subsequently receive 5.57 citations. In Table 2, we present correlations between variables. All our innovation-related variables are significantly positively correlated with creation culture. This indicates that a higher creative culture increases the innovation output consistent with our expectations.  Table 3 presents estimation results of our baseline regression in equation (2) with all four proxies of innovation in columns 1 to 4, respectively. Creation Culture is positively and significantly associated with all innovation proxies across all the models.

Multivariate results
These results suggest that firms with a stronger creation culture generate more patents which subsequently receive higher citations and are more efficient in terms of innovation activity. In terms of economic significance, a one standard deviation increase in creation culture increases invention patents (invention plus utility) by 6.44% (11.15%) and improves citations by 4.67%. Given that our creation culture measure is based on individual word counts, the observed effect is economically significant. 13 The results for control variables are also consistent with prior literature. Firm Size is positive and statistically significant in models 1,2 and 4 suggesting that larger firms produce more patents and are associated with higher patent citations consistent with Mazouz and Zhao (2019). However, the coefficient of Firm Size becomes negative when we use Innovation Efficiency (Model 3) as the dependent variable. This indicates that larger firms' innovation efficiencies decrease, even if they are producing more innovation (Janz, Lööf and Peters, 2003) due to economies of scale consistent with Cao et al. (2016). Firm Age is negatively related to innovation in all models showing that younger firms generate more patents than mature firms (Coad, Segarra and Teruel, 2013). Both Leverage and ROA indicate a positive and statistically significant relationship with innovation, while Book to Market negatively affects innovation. These results are largely consistent with the previous studies (David and O'Brien, 2006).
Overall, the above results strongly support our hypothesis that creation culture is positively associated with firms' innovation output. 14 13 We also re-estimate equation (2) using one period ahead innovation measures ( , +1 ) and our results are consistent with those reported in Table 3. Results are not reported but available upon request. 14 In order to further strengthen our results, we use simultaneous quantile regressions as the robustness test to check whether the relationship between corporate culture and innovation is persistent. Specifically, we bootstrap the regression 500 times at 50%, 75% and 90% quantile points. The unreported results are consistent with our baseline regression results and support our hypothesis. Unreported results are available from authors upon request.  (1), Ln (Inven+Utility+1), as the natural logarithm of the number of invention and utility patent applications plus one in column (2), Innovation Efficiency, as the number of invention patents scaled by firm's cumulative R&D investment from year t-2 to year t in column (3) and Ln (Citations+1), as the natural logarithm of the number of invention patent citations plus one in column (4). The full sample comprises of 17,295 firm-year observations with 2,583 unique firms between 2008 and 2017. All variables are defined in Appendix 1. All models include year, industry and province fixed effects. Robust standard errors are clustered at firm level. t Values are in parentheses. *, **, and *** represent significance levels at 10%, 5% and 1%, respectively.

Robustness tests
In this section, we check the robustness of our key conclusion that a creation culture enhances innovation by conducting additional tests.

Additional controls
Although we control for a comprehensive set of firm characteristics that may affect firm-level innovation, literature suggests that board and CEO characteristics are also important determinants of innovation (Balsmeier, Fleming and Manso, 2017;Bertrand and Schoar, 2003;Huang and Kisgen, 2013;Islam and Zein, 2020;Simsek, 2007). We control for CEO characteristics including gender, age, education, tenure and whether a CEO is politically connected. In addition, we also control for board size, board independence and whether the CEO is also a chairman of the board 15 . The results presented in Table 4 show that coefficient for creation culture is positive and statistically significant across all innovation variables, and therefore our baseline results are robust to these additional controls. 16 15 Gender is a dummy equal to one if CEO is female, and zero otherwise. Age is the natural logarithm of CEO actual age. Education dummy equals one when a CEO has master's degree or above, and zero otherwise. Tenure is CEO's contract period in natural logarithm. Political Connection is a dummy equal to one if CEO is connected to the central government, and zero otherwise. Board Size is measured as the total number of board members. Board Independence is measured as the percentage of independent director on board. Duality is a dummy equal to one if CEO is also the Chairman of the board, and zero otherwise. 16 In addition, to alleviate the concerns that other CVF culture dimensions (compete, control, collaborate) may impact innovation or weaken the effect of creation culture, we control for these culture dimensions in our baseline regression model. We find a consistent and statistically significant impact of creation culture on innovation even after controlling for the other CVF culture dimensions. Results are not reported but available from authors upon request. Overall, the presence of competing CVF culture dimensions do not attenuate the observed relation between creation culture and innovation.  (1), Ln (Inven+Utility+1), as the natural logarithm of the number of invention and utility patent applications plus one in column (2), Innovation Efficiency, as the number of invention patents scaled by firm's cumulative R&D investment from year t-2 to year t in column (3) and Ln (Citations+1), as the natural logarithm of the number of invention patent citations plus one in column (4). The full sample comprises of 17,295 firm-level observations with 2,583 unique firms between 2008 and 2017. All variables are defined in Appendix 1. All models include year, industry and province fixed effects. Robust standard errors are clustered at firm level. t Values are in parentheses. *, **, and *** represent significance levels at 10%, 5% and 1%, respectively.

Endogeneity
Although the results of our baseline models suggest that creation culture has a positive effect on firm innovation, the endogeneity issues between the culture and innovation cannot be ignored, since the association between corporate culture and innovation may reflect causalities other than those hypothesised by us (Hartnell, Ou and Kinicki, 2011). Therefore, our results might be subject to two types of endogeneity concerns: omitted variables and reverse causality. The omitted variables may bias our coefficient estimates if the unobservable firm characteristics are correlated with both creation culture and innovation output. Further, our observed relation between creation culture and innovation may be spurious if the unobservable firm characteristics affect culture and innovation jointly. The reverse causality is another endogeneity concern which could result in biased and inconsistent OLS regression estimates. We observe that a more creative culture leads to higher innovation output and efficiency, but some of this observed relation could be attributed to the possibility that the level of expected innovation predicts a firm's creative culture. Although, it is difficult to alleviate endogeneity concerns completely, we use a set of robustness tests to mitigate the potential bias.
The omitted variable problem could arise from both time-varying and timeinvariant unobservable. Our earlier test including a comprehensive set of additional controls for CEO and board characteristics, to a large extent, mitigates the effect of time-varying omitted variables which might be correlated with both creation culture and innovation. To mitigate omitted variable concerns caused by time-invariant firm characteristics, we substitute year, industry and province fixed effects in our main regression (Table 3)  In order to address reverse causality, we employ instrumental variable analysis using two-stage least-squares (2SLS) approach. It is important, however, to note that finding suitable instruments for culture and specifically corporate culture is practically very difficult (Nash and Patel, 2019). Nonetheless, we rely on extant literature in formulating our instrument for the creation culture (Cumming et al., 2019;Mazouz and Zhao, 2019). Following this literature, we use industry-year average of the creation culture as our main instrumental variable in 2SLS regressions. 19 We believe that our instrument is valid because the industry culture may strongly influence the culture of any given firm in the industry. Firms operating in an industry with a creative culture are more likely to regard the industry culture as a useful economic attribute and try to build a creative culture. On the other hand, it is unlikely that these industry-level instrumental variables directly predict firm-level innovation that cannot be explained by the firm-level culture. Furthermore, industry-level variable is less likely to be affected by an individual firm's policy and satisfies both exclusion and relevance 17 Assuming = 1. 18 We also use Year×Industry and Industry×Province fixed effects and the unreported results are consistent. Moreover, our unreported results remain qualitatively same after including firm fixed effects as well as including lagged values (2 and 3) of creation culture. 19 We also use Industry Median as an instrument for creation culture in our 2SLS regression and find that the results remain unchanged. Moreover, use of province average as an instrument yields qualitatively similar result. The unreported results are available from authors upon request.
condition. Our approach is consistent with Jiang et al. (2017) and Li et al. (2019), who use industry average for instrumenting disclosure tone and integrity culture. Table 5 reports results for the instrumental variable regressions. First stage results are presented in column (1) Table 3.  (2), Ln (Inven+Utility+1), as the natural logarithm of the number of invention and utility patent applications plus one in column (3), Innovation Efficiency, as the number of invention patents scaled by firm's cumulative R&D investment from year t-2 to year t in column (4) and Ln (Citations+1), as the natural logarithm of the number of invention patent citations plus one in column (5).The full sample comprises of 17,295 firm-year observations with 2,583 unique firms between 2008 and 2017. The Wald F Statistic represents Kleibergen-Paap rk Wald F Statistic in our regression. All variables are defined in Appendix 1. All models include year, industry and province fixed effects. Robust standard errors are clustered at firm level. t Values are in parentheses. *, **, and *** represent significance levels at 10%, 5% and 1%, respectively.

Alternative measures of creation culture
It is possible that our findings are sensitive to how we measure creation culture. We employ three approaches to minimise the potential measurement issues with our creation culture measure. First, we scale creation words by the total number of words in firms' annual reports following Fiordelisi and Ricci (2014). Different to our main measure, this measure shows the relative importance of creation words in the whole annual report. The relevant results presented in Table A.4 (Appendix) show that our key findings persist.
Second, it might be argued that not all words in our creation bag equally characterise a creation culture 20 . In order to mitigate the noise and measurement problems in our creation culture variable, we use a reduced bag of words consisting of most releveant words. We reconstruct our creation culture variable using only words such as idea*, innovate*, learn*, new*, start*, vision*. The relevant results reported in Table A.5 (Appendix) show a consistent and significant effect of creation culture across all measures of innovation.
Finally, we use a more robust CVF bag of words measure for cultural dimensions following . The authors develope this new bag of words based on the keywords used within the Organizational Culture Assessment Instrument (OCAI),a survey developed by Cameron and Quinn (2011) to measure firms' culture through the responses of firms' employees 21 .The OCAI based bag of words captures diverse dimensions considered important by the firms in measuring their internal culture.  demonstrate that OCAI based approach provides relatively better and robust measure of corporate culture using content, external, dimensionality and predictive validity test. We reconstruct CVF culture dimensions using the OCAI based bag of words and recalculate our creation culture measure using equation (1). The results of our baseline regressions using the OCAI based creation culture are presented 20 Words stems such as dream*, envis*, freedom*, risk*, thought* etc. might not be considered relevant to a creative culture. 21 Cameron and Quinn (2011) base OCAI along six dimensions: dominant characteristics, organizational leadership, management of employees, organization glue, strategic emphases and criteria of success.
in Table A.6 (Appendix) and show our key findings to be unchanged. 22 Overall, results from tests using alternative measures of culture show that our key findings are not sensitive to the measurement of creation culture.

Alternative estimation method
Since our patent and citation variables are count based with many observations having zero patents and citations, we use Poisson and Tobit regression models following Jiang and Yuan (2018) and Mazouz and Zhao (2019). The results reported in Table A

Further analysis
In this section, we use cross-sectional variation in firms' product market competition and managers' career concerns to further explore the effect of creation culture on innovation. Firms in a competitive product market strive to achieve and maintain a competitive advantage which is necessary for survival in the market. Investment in R&D and innovation are vital sources for firms to attain a sustainable competitive advantage. If a creation-oriented culture leads to higher innovation, we expect to have a stronger effect of creation culture in a highly competitive product market. In order to test this conjecture, we use median value of Herfindahl Hirschman Index (HHI) to split our sample in a high competitive product market (HHI below sample median) and a low competitive product market (HHI above sample median) following Chemmanur and Tian (2018) and Mazouz and Zhao (2019). Columns (1) to (4) of Table 6 report results using high and low product market competition. 23 The results show that the magnitude of coefficient estimates on creation culture is much larger and highly significant for firms in more competitive environment compared to firms in low market competition. The chi-square test values from columns (1) to (4) indicate that the difference in culture coefficients in low and high product market competition subsamples is statistically significant. These results suggest that creation culture exerts 22 We also test the robustness of our results using OCAI based measure in two ways: (1) using other culture dimensions as controls in the baseline regression model and (2) scaling the OCAI creation bag of words by the total number of words in the annual report. The unreported results remain unchanged. 23 We use our baseline regression specification (eq.2) augmented with additional controls for board and CEO characteristics. For brevity, we only report results using invention patents (Inven) and Citations as dependent variables. Unreported results are mainly consistent if we use other innovation proxies. a more pronounced effect on innovation in firms operating in a highly competitive product market and are consistent with our prediction.
A creation oriented culture focuses on creating or generating new resources through innovation by fostering independent thinking, autonomy, flexibility and risk-taking (Hartnell, Ou and Kinicki, 2011). However, experimentation and creativity in new products and services is associated with failures and costs. Organisations that instil a creatrive culture provide a flexible structure with adaptability where failure is accepted, tolerated and considered as a part of the cost of doing novel work (Hutchison-Krupat and Chao, 2014). Moreover, given the uncertainty and long-term nature of innovative projects, creative cultures shield managers from short-term performance evaluations by providing higher tolerance-for-failure. A strong creative culture is likely to reduce managers' career concerns thereby facilitating innovation, and we expect that the effect of creation culture on innovation varies with the extent of manager's career concerns within a cross-section of firms. We follow Kim, Park and Song (2019) and use firmlevel profitability growth as a proxy of managers' career concerns. Firms are split into high and low career concerns using the change in ROA growth rate during a given year compared to the previous year. Firms with a ROA growth that changes from top to bottom tertile are defined as firms with high career concern. Sample firms that do not belong to high career concern subsample are defined as firms with low career concerns.
The results of our analysis using subsamples based on career concerns are reported in columns (5) to (8) of Table 6. While significant in both subsamples, the coefficient estimates of creation culture are much larger in high career concern firms compared to the low career concern ones for both invention patents and citations. Moreover, the difference in culture coefficients between low and high career concern subsamples is statistically significant. Consistent with our expectations, creation culture has a larger effect on innovation in firms with high managerial career concerns.  (1) and (3) presents the results for low competitive product market subsample, while columns (2) and (4) show the results from subsample of firms in high competitive product market. Columns (5) and (7) presents the results for the subsample of firms with low career concerns, while Columns (6) and (8) show the results from subsample of firms with high career concern. The dependent variables are: Ln (Inven+1), as the natural logarithm of the number of invention patent applications plus one and Ln (Citations+1), as the natural logarithm of the number of invention patent citations plus one. All variables are defined in Appendix 1. Firm controls and characteristics for CEO and board are included. All models include year, industry and province fixed effects. Robust standard errors are clustered at firm level. t Values are in parentheses. *, **, and *** represent significance levels at 10%, 5% and 1%, respectively.

Conclusion
In this study, we investigate whether, and to what extent, corporate creation culture influences firm innovation in the Chinese market. We use textual analysis of annual reports following (Fiordelisi and Ricci, 2014;Andreou et al., 2019) to quantify a creation culture. Using all listed firms on both the Shanghai and Shenzhen Stock Exchanges between 2008 and 2017, we find strong and consistent evidence that a creation culture positively affects innovation. Our results are robust to using alternative proxies of innovation and creation culture as well to the alternative model specifications including tests related to endogeneity issues. We also find that firms in more competitive markets and subject to higher managerial career concerns benefit most from the cultural effect.
We contribute to the growing stream of literature on corporate culture as well as the determinants of innovation. Recent studies have documented corporate culture as an important factor in determining firm-level economic outcomes such as risk-taking and CEO turnover (Fiordelisi and Ricci, 2014;Nguyen, Nguyen and Sila, 2019). We add to this stream of literature by showing that corporate culture is also an important factor for the innovative performance of the firms. Corporate innovation is the key to firms' long-term competitiveness and sustainability and finding factors that determine 26 innovation is equally important for academic experts as well as practitioners. Our findings suggest that a creation culture is essential to spur innovation in firms and these results have important practical implications for managers, boards, shareholders and policymakers alike.
Our study is not without limitations. First, although we have carried out a battery of robustness tests and provide evidence suggesting a causal interpretation, we cannot rule out that unobserved characteristics could bias our findings. Second, we rely on patent application data, and while patent-based proxies have several advantages over the other innovation-related metrics such as R&D, they are imperfect measures of innovation and do not capture innovations that are not patented by the firms.