**5. Research results**

*Current Issues in Knowledge Management*

quantitative data (in our case, regressors).

a data set as well as the probability that the *i*

respects, probit and logit analyses are similar.

and nine companies in the Moscow Region).

Characteristics of the selection of firms in the sector (%)

Average headcount characteristics of companies (%)

Foreign proprietary ownership characteristics of companies (%)

interested in the behavior of the likelihood function.

if a certain condition is met [32].

**Table 3**). To eliminate the endogeneity problems "associated with the different direction of the cause-and-effect relationships between the size indicators and property parameters, the values of these predictors in the model are taken for the previous period" [27]. An attempt to use a linear regression to predict innovation activity of enterprises

after entry into a foreign market does not make sense, as the linear form values are on a continuous quantitative scale, while the variable is measured discreetly [44]. Therefore, it is recommended that special regression models be constructed to investigate dependencies between binary variables (innovation indicators) and

There are two approaches that allow to construct such models. The first one involves building a linear probability model (using robust standard errors), which will not be used by us, while the second one involves building nonlinear models (logit and probit) [37]. These models capture dependencies between a variable and

The probit model differs from the logit model only in that the normal distribution density function is used instead of derivative logistic curve. In the other

Their idea is that the likelihood function is maximized—there is a probability that what is present in our sample will be obtained randomly. In practice this means that we no longer pay attention to the sums of squares of the residuals and are

We performed the required analysis of the collected data for 252 Russian compa-

High-technology industries 4.6 28.9 25.4 Middle-technology industries 45.7 34.9 44.9 Low-technology industries 49.7 36.2 29.7 Total 100 100 100

100–199 5.4 2.7 3.0 200–499 7.9 6.2 7.1 500–999 7.6 13.4 9.7 1000–4999 52.4 47.9 51.7 5000–9999 16.3 15.5 16.1 10,000 and more 10.4 14.3 12.4 Total 100 100 100

Share of exporting companies with foreign ownership 34.2 49.8 54.2 Share of non-exporting companies with foreign ownership 7.1 22.4 16.5

*Descriptive statistics of inspected firms in the analyzed timeframe of 2003–2017, % of respondents.*

nies, different in terms of affiliation with a variable, to construct a model. In our sample, 55% of the respondents are located in the capital and in the Moscow Region (128 companies in the two capitals, Moscow and St. Petersburg,

th value of a binary variable is equal to 1

**2003 2017 Panel**

**10**

**Table 4.**

**Table 5** presents the results of the calculation of the relationship between the innovation behavior indicators and the export status of industrial companies.

The hypotheses put forward by us on the selectivity of enterprises ("selfselection" for foreign markets), the existence of learning-by-exporting effects, and the influence of the duration of exports on the enhancement of learning spillover effects were confirmed (the first hypothesis—partially).

Thus, "new" exporting companies, unlike "permanent" exporters, do not have a visible relationship between implementation of new products, technologies, and the start of exports (the significance of the coefficients was not confirmed, Ɓ < p, and Ha is not rejected, where Ɓ is the level of significance, Ha is the hypothesis on the absence of dependencies, or Ɓ\_*i* = 0). The coefficients themselves and the probabilities of the innovation behavior under study being exhibited are much lower than for similar traditional exporters. This can be explained by the fact that R&D investments which might have been initiated after or at the time of entry into foreign markets have not yet yielded results. That said, the status of "traditional" exporters increases the likelihood of investments in advanced research and development by 38%. We believe that this statement is also true vice versa.

For all innovation behavior indicators out of the five indicators considered for a group of traditional exporters, the sign in the models estimating regressor dependencies for a past period (2015) considered by us is positive, and the statistical significance (at the level of 1, 5 and 10%) was proven, indicating that stable export activities serve as an incentive for industrial companies to apply new technological, process, and marketing innovations, which previously were not included in the firm's plans, much more often compared to non-exporting firms.

Our research shows that the impact of external knowledge effects on the productivity of industrial companies depends on the geographical destination of exports: (a) markets in CIS countries plus Russia itself and (b) markets in non-CIS countries. In the case of exports abroad (primarily to West Europe and America), knowledge effects have a significant positive impact on Russian industrial companies, which begin to develop state-of-the-art technologies and increase R&D and marketing expenditures to boost sales of products and services and increase the


*Source: constructed by the authors.*

*Note: Standard errors were calculated from the Hessian.*

*\*\*\*Significance at the level of 1%.*

*\*\*Significance at the level of 5%.*

*\* Significance at the level of 10%.*

#### **Table 5.**

*Results of the regression analysis of seven models measuring the relationship between the innovation behavior indicators and various criteria of the export status of industrial companies.*

share of the international market. The dependence of spillover effects and innovation activity, efficiency across the high-tech industry, is quite high. It should be emphasized that learning requires special efforts, the ability to assimilate knowledge, and time, and therefore learning effects do not manifest themselves immediately, and they become visible only with a certain time lag.

According to the performed calculations, investments of industrial companies in R&D, marketing, and release of new products are more characteristic for metropolitan companies (at a significance level of 1%). The relationship between the availability of an international office and introduction of innovations, on the contrary, was not proven. The companies' size (based on the logarithm of the number of employees) only had an impact on the production of new technologies: if a company

**13**

*Knowledge Spillover Effects: Impact of Export Learning Effects on Companies' Innovative…*

Age 1.561 Size 1.293 Foreign 1.274 Region 1.149 Exp\_period *i* 6 < *xi* < 7 Exp\_status *i* 1.5 < *xi* < 3 Sector *i* 1 < *xi* < 2.5

belongs to medium-sized enterprises (101–250 people) or is larger, the probability of

*independent variables. As all values of the coefficients are <10, the models do not exhibit a strong correlation between* 

*), where R(j) is the multiple correlation coefficient between variable j and other* 

It can also be concluded that the impact of learning spillover effects of knowledge is manifested in industrial companies as a result of a change in their innovation behavior: the longer a company operates in foreign markets, i.e., the longer the learning process, the flow of knowledge, the more pronounced the transformation of the firm's innovation behavior (changes in business processes, renewal of company staff, increase in the creativity and skills of employees, changes in the business

The study has shown that the duration and destination of exports significantly

It should be noted that we also attempted to build linear probability models. We considered a large number of variations of factors that could influence innovation behavior. However, the same variables proved to be significant as in the probit model analysis. We also considered variants with logarithms of multiple status variables, the period of exports, and specialization, which changed the situation slightly. The number of correctly predicted cases was about 196–209 (77.6–82.9%). The R-squared in all models fluctuated around 0.20, which is not high enough to

When constructing models, we also tested variables for multicollinearity by the

The study carried out by us was aimed at exploring the impact of knowledge spillover effects on the innovative activity of industrial companies in Russia. Special attention was paid to which characteristics of a company contributed to knowledge

The obtained results allow drawing conclusions about the positive impact of knowledge spillover effects stemming from the companies' export activities. "New" exporting companies, unlike "permanent" exporters, do not have visible links between implementation of new products, technologies, and the start of exports. The coefficients themselves and the probabilities of the innovation behavior under

influence organizations' innovative activities, but innovations do now always

encourage managers of industrial companies to start exporting.

accumulation and stimulated an increase in innovation activity.

inventing innovations is increased by 22% (at a significance level of 1%).

*DOI: http://dx.doi.org/10.5772/intechopen.86255*

**Values > 10.0 may indicate multicollinearity**

**Minimum possible value = 1.0**

*Note: VIF(j) = 1/(1−R(j)<sup>2</sup>*

*the explanatory variables.*

**Table 6.**

model and other indicators).

*Analysis of the multicollinearity of indicators.*

confirm the hypotheses put forward by us.

inflation factor method (**Table 6**).

**6. Conclusions**

*Knowledge Spillover Effects: Impact of Export Learning Effects on Companies' Innovative… DOI: http://dx.doi.org/10.5772/intechopen.86255*


*Note: VIF(j) = 1/(1−R(j)<sup>2</sup> ), where R(j) is the multiple correlation coefficient between variable j and other independent variables. As all values of the coefficients are <10, the models do not exhibit a strong correlation between the explanatory variables.*

#### **Table 6.**

*Current Issues in Knowledge Management*

Exp\_period2 \*

Region \*

Ind1 \*

Ind5 0.102 (0.045) \*

Ind10 \*\*0.193 (0.095) \*

(0.005)

*Note: Standard errors were calculated from the Hessian.*

Ind8 −0.081

*Source: constructed by the authors.*

*\*\*\*Significance at the level of 1%. \*\*Significance at the level of 5%.*

*Significance at the level of 10%.*

McFadden R-squared

*\**

**Table 5.**

Previous 0.264 (0.119) \*\*0.269 (0.147) \*

Exp\_period1 \*\*\*0.381 (0.305) \*\*0.182 (0.049) \*

Size 0.252 (0.227) 0.338 (0.211) \*

0.109 (0.081) \*

**12**

share of the international market. The dependence of spillover effects and innovation activity, efficiency across the high-tech industry, is quite high. It should be emphasized that learning requires special efforts, the ability to assimilate knowledge, and time, and therefore learning effects do not manifest themselves immedi-

*Results of the regression analysis of seven models measuring the relationship between the innovation behavior* 

**Y1 (R&D) Y2(New\_Tech) Y3(New\_Prod) Y4(Exp) Y5(Marketing)**

0.105 (0.046) \*\*0.214 (0.184) \*

0.081 (0.051) \*\*0.241 (0.231) \*

0.226 (0.173) −0.006

−0.876 (0.782) 0.008 (0.002) 0.019 (0.025) 0.137 (0.066)

0.522 (0.524) 0.134 (0.086) 0.177 (0.151) −0.132 (0.069)

0.282 (0.169) 0.174 (0.134) 0.057 (0.098) \*

(0.003)

0.851 (0.771)

0.085 (0.071)

−0.113 (0.093)

0.163 (0.134)

0.028 (0.005)

0.239 (0.194)

\*

0.032 (0.018) −0.344 (0.289)

Const 0.416 (1106) 0.392 (0.209) 0.254 (0.022) 0.169 (0.138) 0.675 (0.563)

0.361 (0.302) 0.159 (0.123) 0.172 (0.125) 0.012 (0.004) \*

Exp\_period3 0.124 (0.001) −0.331 (0.210) −0.319 (0.238) Dropped −0.378 (0.267) Exp\_status1 0.016 (0.004) \*\*−0.302 (0.193) −0.351 (0.268) 0.016 (0.007) −0.461 (0.386) Exp\_status2 0.081 (0.017) −0.041 (0.019) −0.134 (0.089) 0.029 (0.019) 0.018 (0.009) Exp\_status3 0.256 (0.119) 0.087 (0.052) Dropped 0.068 (0.033) 0.225 (0.193)

Age −0.206 (0.102) 0.356 (0.245) Dropped −0.059 (0.031) 0.118 (0.109)

Foreign 0.015 (0.006) −0.289 (0.192) 0.073 (0.019) 0.134 (0.042) −0.153 (0.097)

Ind2 −0.379 (0.302) 0.082 (0.061) 0.014 (0.007) 0.178 (0.160) 0.128 (0.106) Ind3 Dropped Dropped 0.005 (0.000) Dropped −0.167 (0.143) Ind4 −0.289 (0.141) −1.441 (0.046) −0.018 (0.012) 0.153 (0.127) 0.007 (0.001)

Ind6 Dropped Dropped Dropped Dropped Dropped Ind7 −0.488 (0.279) −0.656 (0.739) −0.497 (0.362) −0.041 (0.022) −0.443 (0.368)

−0.089 (0.495) −0.021

Ind9 −0.479 (0.056) 0.121 (0.797) 0.015 (0.004) 0.051 (0.022) −1.884 (0.974)

(0.007)

0.221 0.229 0.189 0.271 0.261

0.561 (0.368) \*\*0.374 (0.371) 0.269 (0.156) 0.247 (0.237) \*

According to the performed calculations, investments of industrial companies in R&D, marketing, and release of new products are more characteristic for metropolitan companies (at a significance level of 1%). The relationship between the availability of an international office and introduction of innovations, on the contrary, was not proven. The companies' size (based on the logarithm of the number of employees) only had an impact on the production of new technologies: if a company

ately, and they become visible only with a certain time lag.

*indicators and various criteria of the export status of industrial companies.*

*Analysis of the multicollinearity of indicators.*

belongs to medium-sized enterprises (101–250 people) or is larger, the probability of inventing innovations is increased by 22% (at a significance level of 1%).

It can also be concluded that the impact of learning spillover effects of knowledge is manifested in industrial companies as a result of a change in their innovation behavior: the longer a company operates in foreign markets, i.e., the longer the learning process, the flow of knowledge, the more pronounced the transformation of the firm's innovation behavior (changes in business processes, renewal of company staff, increase in the creativity and skills of employees, changes in the business model and other indicators).

The study has shown that the duration and destination of exports significantly influence organizations' innovative activities, but innovations do now always encourage managers of industrial companies to start exporting.

It should be noted that we also attempted to build linear probability models. We considered a large number of variations of factors that could influence innovation behavior. However, the same variables proved to be significant as in the probit model analysis. We also considered variants with logarithms of multiple status variables, the period of exports, and specialization, which changed the situation slightly. The number of correctly predicted cases was about 196–209 (77.6–82.9%). The R-squared in all models fluctuated around 0.20, which is not high enough to confirm the hypotheses put forward by us.

When constructing models, we also tested variables for multicollinearity by the inflation factor method (**Table 6**).
