The present section starts with Hypothesis I that firm size results in foreign expansion when industry is controlled for. Foreign expansion is the network extensiveness defined as the total number of subsidiaries of a UJP in Europe. Network extensiveness (NetExt) should be regressed against firm size measured by its total fixed assets in order to test the theory (Penrose 1972). Firm size may be measured by the consolidated total assets, total fixed assets, and total revenues of the UJPs and since these indicators in the sample are highly correlated with r=0.96-0.98, the one with no missing data (total
18 Additionally, dependences among the MNE themselves could exist. Studies on
bunching behavior (Makino and Delios 2000) and suppliers following their clients are examples of this. However, these dependencies influence mainly the timing of the investment, not the final network itself. Given the assumption that a firm with lasting interest in Europe will become involved at some moment of time, the cross-sectional analysis is freed from such dependencies.
revenue) is selected19. Furthermore, the square root of this variable is used to reduce the range of the variable, and to represent the diminishing influence of this variable on the NetExt. From now on the variable FirmSize will signify the square root of total consolidated revenues of the UJP. Hypothesis I states that this variable is positively correlated with NetExt when industry is controlled for. Now Hypothesis I is split to three parts.
Hypothesis I.1: Firm size is positively related to Network Extensiveness
There is a need of other two variables, type of final market (FinMkt) and degree of involvement in Europe as compared to the rest of the world (EurFoc) in order to control for the effect of these circumstances on the FirmSize - NetExt relationship. The European focus is defined as the total number of subsidiaries in Europe divided to the total number of subsidiaries in the world, then this ratio is multiplied by 100. It shows what part of its resources an UJP pours into Europe. Only given the European focus will a test of hypothesis I.1 be valid. Alternately, given equal firm size, network extensiveness will increase with increase in European focus implying a positive relationship.
Hypothesis I.2: European focus is positively related to network extensiveness The type of final market (FinMkt) is a variable that places the UJP among the manufacturers of goods for final consumer market, or among the manufacturers of intermediate products, or both. Because the UJPs are often extremely diversified, of interest is only their production and sales abroad.
19 This correlation is exactly where Caves’ and Penrose’ variables meet, because it is
frequently found between technology (R&D) and assets (Padmanabhan and Cho 1996, Caves 1996).
Therefore, the average revenue from their commerce and manufacturing subsidiaries will show what is their place in the value chain. This proxy will arguably distinguish part suppliers from manufacturers for the final market but no specific hypothesis is set forth. Finally, the average age (AverAge) of all European subsidiaries measures the experience effect on NetExt, which is expected to be positive.
Hypothesis I.3: Experience and network extensiveness are positively related For summary of the variable definitions and descriptive statistics see Table 3. The FirmSize (i.e. total revenue or total fixed assets) is positively and significantly related to NetExt, contributing almost 50% of the explained variance (see Table 3, Panel A). EurFoc and FinMkt are important as well because they contribute additional 9% and 5 % of the adjusted R2. FinMkt is positively correlated with FirmSize probably because UJPs producing for other firms are often small firms, while the producers for a final market are big-sized, diversified firms. EurFoc shows negative correlation with FirmSize, because UJPs with a lot of assets are more likely to invest in Europe as well as in other regions than smaller UJPs, which will be present in few regions. Therefore those small UJPs that happened to be in Europe will likely have no other investment in the world. Finally, the small correlation between AverAge and EurFoc is not troublesome because the former variable turns out to be of marginal importance for explaining NetExt and its omission from the analysis does not affect the explanatory power of the model (R2 remains unchanged).
The results in Table 3 confirm Hypothesis I.1 and I.2 and reject Hypothesis I.3. The regression controls for industry by including dummies for the respective industries – construction (Constr), chemicals (CH), machines (MA), cars (CA),
precision (PR), resource-based (RB) and others (rest), mainly food, or unknown. The reference industry is electronics because its size is biggest of all (101 cases versus 82 for machines, 82 for chemicals, 64 for rest, 61 for cars, 29 for precision, 29 for resource-based, and 14 for construction)20 and because it turns out to be with second biggest effect on network extensiveness and this facilitates comparisons among industries. The results show that switching form electronics to any other industry except precision reduces the network extensiveness, and this is significant for almost all industries. When industry is controlled for, average age has no influence whatsoever on NetExt, while final market (FinMkt) has a negative influence (its coefficient is very small due to difference in the measurement units for NetExt and FinMkt).
The interpretation of the above findings has to be very careful, as is always the case with observational studies. First, theory claims that resources will bring about expansion abroad, and our finding of significant firm size confirms this theoretical claim. In order to confirm this effect, for the 462 UJPs present in Europe, additional analysis finds that only their firm size (square root of total revenue) accounts for 62% of the network extensiveness variance in the whole world. After exclusion of 4 outliers – JT, Toyota, Honda, and Matsushita – the relationship is almost perfectly linear one with no heteroscedasticity and the coefficient (slope) is 0.4, implying that every 25 million $ in revenue bring about two more foreign subsidiaries. In this case no other control variable seems necessary. Yet it is clear that we have – in the world and in Europe – only association, and no causal relationship. Clearly,
20 From all 462 UJPs with available data, 138 have no FinMkt data (3 from these 138
having subsidiaries abroad will bring rise in revenues, or in total assets. Correction of this problem may come by measuring UJP size only by its assets at home, although this will distort the reality of the UJP ownership advantages. Another solution would be to include time and time-series analysis for the relation between created subsidiaries at time t and assets at time t-1. At present, the result may be interpreted only as association.
In addition, even if we accept that the association is persuasive enough to show that assets possess the “internal inducement” property (Penrose 1972) for expansion both in Europe and the world, the EurFoc variable does not in fact answer the question why MNEs have different degrees of European involvement, it only reflects the already existing degrees. The question refers directly to the motivation for FDI, and since most of the UJP have market-seeking orientation in Europe, but not so in some of the other parts of the world, the answer lies in the specific external production - external market ratio set in the strategy of each UJP. The FinMkt variable cannot distinguish the influence of the type of final market from that of size in the case of MNE from Japan (explained below). It is better to omit it in the present study or to omit FirmSize in order to solve the problem of causality and replace it with narrower in scope variables like FinMkt, which reflect various sides of the reality for which firm size is a summary proxy. Finally, the insignificance of the average age (AverAge) should not mean that experience is irrelevant for expansion21; it rather means that another better proxy based on subsidiary age is necessary. All of the above interpretations were tested with the F-test for
21 The older is the European presence the more likely the present network is extended.
However, most of these extensions could be recent or “young” in age, and therefore average age may “siphon out” experience instead of measuring it.
incremental contribution (restricted least squares). Industry has a significant incremental contribution with F7,312 = 5.39 (p<0.01). So does FinMkt with F1,312=4.23 (p=0.04). So does EurFoc, with F1,312=80.2 (p=0.000). However, AverAge does not contribute anything with F1,312=0.48 (not significant).
When the regressions exclude EurFoc or the industry dummies, FinMkt remains with stable negative sign but is less significant. Otherwise, all variables’ coefficients and levels of significance remain stable when AverAge is excluded and relatively stable when EurFoc or FinMkt or dummies are excluded (individually, not together).
When FirmSize is excluded (its contribution is huge), EurFoc coefficient drops somewhat in value, probably due to the zero-order negative correlation between it and FirmSize. AverAge and FinMkt become significant and positive. It is not clear why this happens to AverAge (zero zero-order correlation with FirmSize), but the effect on the FinMkt variable is important - FirmSize inclusion makes it shift signs being in both cases significant. FinMkt is affects negatively the dependent variable (NetExt) in the presence of firm size and this is due to the high zero-order correlation between FirmSize and FinMkt. In fact FirmSize seems to absorb all the influence that FinMkt could have on NetExt, because of the shift in signs of FinMkt. The reasons to assume that the population distributions of those two variables have independent influence on NetExt evidently do not hold for the MNE from Japan, probably because of the industrial organization features there.
With respect to the industry dummies, only if firm size is omitted does precision (PR) lose its significance and machines (MA) become significant, both signs being the same. A test for structural stability showed that all but “Rest”
industries are concurrent with zero intercept and slopes (of the main FirmSize variable) decreasing in the order (i) precision; (ii) electronics; (iii) cars, machines and chemicals, and (iv) construction, resource-based and rest, with differences in the slopes being significant between these four groups (see Table 3, Panel D). The results show that network extensiveness of precision and electronics respond faster to changes in FirmSize than other industries, and those of construction, resource-based and rest (mainly food) respond slower. These results seem logical and realistic. Table 3, Panel B shows that multi-collinearity, except for FinMkt, is not present in the sample.
Additional extensions and verifications of the above analysis were carried out by taking the log of the dependent variable in order to smooth the initial distribution. The results are almost identical, as well as the conclusions concerning signs and significances. Only the variable FinMkt loses its significance. Further modifications included weighting the manufacturing subsidiaries in the definition of NetExt, but differences in results were minimal22. Additional verification is possible by replacing the regression model with a model based on a Poisson distribution.
The question of whether to give equal weights to different types of subsidiaries in the network arises naturally. The justification of any weighting procedure, however, is quite difficult. Therefore, it is better to analyze separately the peculiarity of each type, and the following section undertakes this task for the subsidiaries in sales and manufacturing.
22 Since manufacturing subsidiaries require more assets and involvement, they were
weighted twice as much as commerce ones in calculating the EurFoc. Results did not change much. It is not suitable, however, to perform such a weighting procedure across the board because the manufacturing and commerce subsidiaries may differ with respect to the value-added-intensity of the products they supply.