The positive influence of the innovation performance on a firm's financial performance is extensively studied (CHENG; HUIZINGH, 2014; FAEMS; VAN LOOY; DEBACKERE, 2005). In addition to the impact on financial performance, innovation is essential for the firm's survival in today's uncertain environment (TEECE, 2007). Concurrently, with the approach of the relationship between innovation and firm's performance, research has emerged examining internal and external sources of innovation (FRENZ; IETTO-GILLIES, 2009). Internal sources come mainly from R&D inside the boundaries of the organization. On the other hand, external sources can be innovations acquired from other firms, merges, acquisitions and collaboration with other players in the industry (CHESBROUGH, 2003; FAEMS, 2005).

We find that the relationship between internal R&D and innovation performance is positive in many studies (BELUSSI; SAMMARRA; SEDITAB, 2010; FAEMS, 2005), although it is dependent on the firm's structure (ARORA; BELEZON; RIOS, 2014) and on the type of knowledge developed (PÉREZ-LUNO; MEDINA; LAVADO; RODRIGUEZ, 2011). The influence of external sources on innovation performance presents several contingencies such as the type of partner (CHATTERJI; FABRIZIO, 2014; SOH; SUBRAMANIAN, 2014), the type of knowledge pursued (PÉREZ-LUNO et al. 2011) and the absorptive capacity of the firm (COHEN; LEVINTHAL, 1990). As absorptive capacity rises with higher levels of internal R&D (COHEN; LEVINTHAL, 1990; TSAI, 2009), internal and external R&D may be positively related. In contrast, some papers found a relationship of substitutability between them (HAGEDOORN; WANG, 2012).

Another relevant factor for innovation success is the environment, beginning with the country. The innovation process and the accumulation of technological capabilities for innovation development occurs in a different manner in developing countries compared to developed economies (CHOUNG; HWANG; SONG, 2014; FIGUEIREDO, 2016; KIM, 1997). According to Nelson (1993), the National System of Innovation (NSI) determines the innovative performance of national firms. Brazil occupies the 69th position in the 2016 ranking of the most innovative countries according to the Global Innovation Index (DUTTA; LAVIN; WUNSCH-VINCENT, 2016). In contrast, European countries dominate the top positions of the list, with four of the top five countries of the ranking, and 15 countries among the 25 most innovative. In addition to Brazil's unenthusiastic position, when it comes to transforming the innovative capacity into financial results, Brazilian firms fail, even in comparison with neighbours of South America (INFOMONEY, 2015). This reality requires Brazilian firms to make enormous efforts to achieve positive results. In order to try to find lessons on how Brazilian manufacturing firms may improve its innovation and financial performance, we propose the following research question: How do internal R&D and external R&D (from strategic alliances) influence innovation performance and how can innovation have a positive impact on a firm's performance in the context of the Brazilian manufacturing firms?

To answer the proposed question and building on the literature mentioned above, we propose a theoretical model which investigates the relationship among internal R&D, external R&D (strategic alliances), innovation performance, and financial performance of firms. The model is operationalized for Brazilian manufacturing firms using PINTEC 2011 (Innovation Survey) and PIA 2009 to 2013 (Annual Industrial Survey) databases, conducted by IBGE (Brazilian Institute of Geography and Statistics). Other research projects are also testing this model, with some adaptations, for other regions and industries, using different databases.

The remainder of the paper consists of the literature review, where we formulate the hypotheses and present the proposed model. Henceforth, we present the methodology, which includes the description of the data and the explanation of the sample selection, a description of the variables and the statistical method. Then, we describe the results and discuss them. Subsequently, we present the conclusion, containing the implications for academics and practitioners, the limitations and some suggestions for future research.

The research on innovation in organizations shows that several dimensions influence innovation performance (IP). Ahuja, Lampert and Tandon (2008) identified four dimensions that influence innovation efforts and its outputs: industry structure, firm's characteristics, interorganizational attributes, and institutional influence. The R&D strategy influences some elements of the firm's characteristics as far as it defines the sources from where the firm will acquire or develop knowledge to generate innovation. Therefore, the choice of R&D sources, both internal and external, is central for the firm's IP, as well as IP is central for the firm's financial performance.

From the sample of 2,810 Brazilian manufacturing firms that devoted themselves to some innovation activity between 2009 and 2010, 1,495 successfully introduced at least one product or process innovation, which represents 53.20% of the firms. On the other hand, 1,315, or 46.80% of the firms, did not introduced any innovation during this three-year period. Table 2 shows the mean and standard deviations of the constructs' proxies for the whole sample. From the table, we can verify that 38.50% of the firms introduced product innovation and 26.30% of the firms introduced process innovation. Another interesting observation we make from Table 2, is that the average performance of the firms in 2010 and 2011 overcomes the future performance of 2013. Turnover growth was on average 36.20% in 2010 and 2011 and declined to 13.6% in 2013. Employees' growth was 9.70% on average in 2010 and 2011 and only 3.00% in 2013, and the value added improved 23.80% in one year on the average of 2010 and 2011 compared to a 21.80% growth in 2013. Although growth declined, all the performance indicators based on growth are positive, which may indicate that the manufacturing firms are still recovering from the 2008 global crisis, that hit the Brazilian industry strongly, of which the recovery started in 2010 (CUNHA; LELIS; FLIGENSPAN, 2013).

Table 2 (Cont.)

The first step of the analysis was to execute the EFA to reduce the number of variables that would reflect the constructs of SEM. We applied EFA separately for each construct presented in Table 1, as described in the method section, and could successfully reduce only two constructs: Innovation Performance (down to four factors) and External R&D - Strategic Alliances (down to one factor). We present the resulting factors of these two constructs, as a result of the EFA, after the varimax rotation in Table 3. The only variables we use, that weighted 0.7 or higher in each factor to calculate the final variable and are shown in Table 3.

For Innovation Performance, we formed factor 1 by the average of IN_IMP_8, IN_IMP_9 and IN_IMP_11, which are variables that measure the impact of the innovation on costs (in the case of these variables, production costs, labour costs and energy costs). For that reason, we termed factor 1 as Innovations' Cost-Reduction Performance (CR_PERF). We formed factor 2 by IN_IMP_3 and IN_IMP_4, which represent the impact on keeping and increasing the firm's participation in the market. Therefore, we termed factor 2 Innovations' Market Performance (IN_PERF). We formed factor 3 from the introduction or not of product innovation (INOVPROD) and from the innovativeness degree of the product innovation (IPROD_DEGREE) and, for that reason, we termed it Product Innovation Introduction (INTRO_PROD). Factor 4, which we formed from the introduction or not of process innovation (INOVPROC), was termed Process Innovation Introduction (INTRO_PROC). In the case of the construct External R&D - Strategic Alliances, the EFA resulted in only one factor, which we formed from AL_CLI, AL_SUP, AL_CONSUL, AL_UNIV, AL_CENTERS, and AL_TEST_INST. All these forming variables measure the importance of some type of partner for innovation. Therefore, we named this factor as Strategic Alliances (STR_ALL). We do not consider in the SEM all the other variables of the constructs that do not form one of the factors. For the other constructs (Internal R&D - Absorptive Capacity, Current Financial Performance and Future Financial Performance), we used all the variables separately in the SEM.

A CFA is the next step of the analysis, considering the four constructs of the model that contains more than one proxy (as we reduce the construct External R&D - Strategic Alliance down to a unique variable, it does not participate of the CFA). We conduct the CFA analysis using Bayesian estimation. After several runs and the analysis of the modification indexes, we removed the variable CR_PERF from the construct innovation performance and fixed the error variance of IN_PERF as 0. We make no other changes to the measurement model. The results of the measurement model are in Table 4.

Although all the standardized regression weights of the measurement model should be all greater than 0.7 to guarantee the reliability of the individual index, all the regression weights are significant, with the p-value of 0.05, except for TRAIN_EXP in the construct Internal R&D - Absorptive Capacity. However, we decided to keep it to a minimum of three proxies for each construct. The only constructs that presented the minimum level of average variance extracted - AVE, that should be 0.5, and of composite reliability - CR, that should be 0.7, is Innovation Performance. All the others are below these levels. However, some authors consider that the AVE is a very conservative criterion for convergent validity and the researcher may decide that the construct has convergent validity even with more than 50% of the variance explained by error (MALHOTRA; DASH, 2011). Considering this, Future Financial Performance is almost acceptable and we decided not to change the other constructs to keep three variables in each one.

The discriminant validity check is successful in discriminating the constructs between each other, as the construct's AVE are much higher than all the squared estimated correlations between the constructs. For nomological validity, we expected to have significant but low covariances between constructs. Covariances are significant for p<0.05 and low between Internal R&D - Absorptive Capacity and Current Financial Performance (0.045), between Internal R&D - Absorptive Capacity and Innovation Performance (0.134) and between Innovation Performance and Current Financial Performance (0.047). The covariances between Future Financial Performance and the other three constructs are not significant. This may indicate a specific problem with the construct Future Financial Performance's nomological validity. However, the CFA validates the construct, which presents acceptable discriminant and convergent validity. In addition, the results of SEM with this construct present a significant relationship between this construct and the construct Innovation Performance, as shown below. For these reasons, we decided to move forward without additional changes in the measurement model.

The following step of the analysis is the test of the causal model with SEM using Bayesian estimation. We summarize the results of the analysis in Table 5. Figure 2 shows the complete model, the standardized coefficients and the error terms. The model presents a fourth construct, Ext. R&D x AC, representing the moderation relationship of absorptive capacity, as previously explained. The model's posterior predictive is 0.56, which we can consider a good fit, as it is close to 0.5 (GELMAN, 2013).

We examined the standard regression weights of the relationships in Table 5 to analyse the hypotheses. The first hypothesis suggested that the higher the level of external R&D of a Brazilian manufacturing firm, mainly accessed through strategic alliances, the higher its innovation performance. This hypothesis (H1) is supported, with a positive and significant (p < 0.05) path coefficient of 0.105. The open innovation that occurs mainly through strategic alliances in manufacturing Brazilian firms is corroborated by the fact that the construct External R&D is reduced to a variable formed only by the proxies representing the importance of alliances with several types of partners (clients, suppliers, consultants, universities, research centres and test institutes). We exclude the proxies that represent purely acquisition of external knowledge or R&D from the model by the EFA. All the types of partners have a similar weight in the factor that represent External R&D, which indicates that the firms that use strategic alliances for open innovation use all those types of partners and consider them similarly important.

Hypothesis 2 is rejected in the analysis. The path coefficient between the constructs Internal R&D - Absorptive Capacity and Innovation Performance is positive (0.004), but not significant for p < 0.05. In our model, internal R&D intensity, training expenses and personnel education represent this construct. However, training expenses do not reflect the construct significantly. High investments in internal R&D are costly for the firms and are more intensely made by firms in high-tech industries, which are closer to the technological frontier. In the case of Brazil, manufacturing firms are mainly concentrated in the lower-technology based industries and the ones in more high-tech industries are not close enough to the technological frontier so that their internal R&D and knowledge may generate high-performance innovation. These firms are still in the process of technological capability accumulation, in which it is important to learn from more technologically developed partners. However, higher levels of internal R&D, or absorptive capacity, may be important to potentiate the effects of the strategic alliances on the innovation performance, as proposed by hypothesis 3. Our analysis corroborated this fact as H3 is confirmed for p < 0.05 with a path coefficient of 0.993. It is interesting to observe that the only factor of absorptive capacity that had a significant moderation effect with the strategic alliances in the model is level of education of the employees. This means that, for Brazilian manufacturing firms, it is important to have a well-educated team to work on open innovation activities, as they are more qualified to evaluate opportunities, to identify and absorb external knowledge to generate innovation. The level of internal R&D intensity still does not allow enough accumulation of capabilities to improve the absorptive capacity of these firms.

We expected to find a positive relationship between innovation performance and future financial performance, as indicated by hypothesis 4; however, we found a significant negative relationship (-0.897 with p < 0.05), rejecting H4. Although we consistently find the relationship between innovation performance and financial performance in the academia, an amount of time is necessary between the introduction of innovation and the verification of economical results from it. In this study, we considered the innovation introduced between 2009 and 2011 and the indicators of financial performance measured were from 2013. This two-year time lag was not enough for the innovations to translate into performance improvement considering indicators such as turnover growth, employees' growth or value-added growth. However, it is enough to reflect the increase in costs and the decrease in revenues caused by the redirection of resources from marketing and sales to innovation activities (such as internal R&D), and by the management of collaboration (FAEMS et al., 2010).

We did not verify the effect of current financial performance in the innovation performance, which would indicate that successful firms innovate more (path coefficient of 0.001, not significant with p < 0.05, indicating the rejection of H5a). It suggests that Brazilian manufacturing firms that are not yet financially successful may still succeed in innovation if they accumulate enough absorptive capacity and balance their strategic alliances. The greater importance of personnel's education in comparison with internal R&D intensity as a dimension of absorptive capacity helps the achievement of this equality of chances as the former demands less money than the latter to be improved. Neither did we find a significant effect between current financial performance and future financial performance (path coefficient of 0.259, not significant with p < 0.05, indicating the rejection of H5b). It indicates that firms are alternating in the industry leadership in time, with new entrants overcoming incumbents frequently. This may be an effect of innovations introduced previously to 2009 and reinforces the time-lag between the introduction of innovation and financial performance improvements.

This paper investigates the role of internal and external R&D on the innovation performance of Brazilian manufacturing firms, as well as the effect of the innovation performance on the future financial performance of these companies. The analysis of a sample of 2,810 Brazilian manufacturing firms, which conducted innovation activities (successfully or not) from 2009 to 2011, according to the PINTEC 2011 survey, allowed us to find some interesting conclusions.

The main goal of the study is twofold. First, it intended to understand how the different sources of R&D, namely internal and external, influence innovation performance, with innovation performance being the successful introduction of innovation by firms and its impacts on the firms' processes and markets. We achieved our objective. We could evidence that, in Brazilian manufacturing firms, external R&D, mainly from strategic alliances, had a positive influence on the innovation performance, as the theory predicted. Internal R&D investments, such as R&D intensity and level of education of employees, on the other hand, did not affect the innovation performance directly. However, it increased the positive effect of the strategic alliances on the innovation performance, mainly the proxy level of education of employees. The support of this hypothesis is in line with the absorptive capacity theory. However, in the Brazilian manufacture industry case, the proxy of absorptive capacity that is currently more effective, is the one that needs less monetary investments among the three proxies (level of education of employees), which is an important dimension of absorptive capacity but alone could not guarantee an increase of innovation performance. Therefore, we can conclude that continuous investment in all dimensions of internal R&D should be done as the level of technological capabilities accumulated by the Brazilian firms is still low. If they catch-up to the technological frontier, this accumulation of internal R&D will tend to influence the effect of the strategic alliances on the innovation outcomes more (by an increase of the absorptive capacity), as well as, affect innovation directly.

The second objective is to verify the effect of innovation performance on the firms' future financial performance. We could not successfully verify a positive impact as predicted. On the contrary, the effect is negative. The time-lag between the innovation performance indicators and the future financial performance indicators is only of two years (2011 to 2013) and is not enough to reflect the effect of new products or services on the financial performance. However, it is interesting to find that the financial performance in the short term has a decrease provoked by innovation, introduced because of the management costs of the strategic alliances and by the redirection of marketing and sales resources to innovation activities, such as internal R&D, causing a momentary decrease in the financial performance.

One interesting finding is that product innovation had a higher incidence than process innovation (38.50% vs. 26.30%) and new products are responsible for on average 10% of the turnover of the firms in our sample. This higher incidence of product innovation may have emerged from a view by Brazilian firms and the government that product innovation is nobler and more effective to improve financial performance than process innovation, which may have caused a higher level of public financing for the former type of innovation. This may have motivated manufacturing firms to make less effort in generating process innovation, which is usually responsible for improving the production process, which lower costs and may reflect faster in the financial performance than product innovation does, which depends on marketing efforts to introduce the new product in the market and improve the firm's market-share and revenues. This may be one of the reasons we could not identify a positive relationship between innovation performance and future financial performance, with the short time-lag available.

We can point several limitations in this study. The first limitation is that we had to make concessions on the internal validity of the constructs because of the low level of some validity's indicators. Another limitation is that we worked with Brazilian manufacturing firms, which can make the results country and industry-specific. The last limitation we can identify is that, the PINTEC database has some qualitative questions based on feelings and experiences of the respondents, which may bias the analysis.

Regardless these limitations, this paper presents a valuable contribution to understand the relationship among strategic alliances, absorptive capacity, innovation performance and financial performance of the Brazilian manufacturing firms. As implications for practitioners, we identified in this study that Brazilian manufacturing industries should invest more on internal R&D, together with strategic alliances, to improve its absorptive capacity; and focus more on process innovation than we do today. These two strategies should be the focus of governmental programs that are now focusing more on product innovation. As future studies, we suggest conducting the analysis of manufacturing firms in more developed countries, which are in different innovative stages compared to Brazil. We can compare Brazilian manufacturing firms to the ones from Europe or from the USA, for instance, in order to try identifying lessons that Brazilian organizations can learn. Another possible analysis is to test the model with service firms, which are less frequently studied in the literature and have a different innovation process.