4.1 Data sources
To test our hypotheses we combine information from several sources. The sample of startup firms is taken from a comprehensive dataset of semiconductor new ventures that entered the industry between 1997 and 2007. This dataset was provided by
Semiconductor Times, a magazine published monthly by
Pinestream Communications, a private consultancy company specialized in the semiconductor industry. The magazine records new startups in the industry each month and provides a profile of each company, including a description of their product offerings and activities. We consider this source, which has been used in prior research, both exhaustive and reliable (Adams et al.,
2016).
The fate of each startup in terms of survival or exit was tracked until November 2014. In cases of acquisition, we were able to identify the buyer firm through Lexis-Nexis. We then collected information on the acquirers, including their product strategies, through IC Insights. Information on the characteristics of the acquisition deal such as deal type, purpose, size, and overall value (when available) was gathered from Factset. The main SIC and/or NAICS codes necessary for classifying the buyer firm were retrieved through Orbis. Our final dataset of acquired firms includes 395 (or 42.2%) of the 936 firms in the original dataset.
4.2 Variables
The first step in our empirical analysis is the identification of the knowledge heritage and product entry strategy of acquired (i.e. the targets) firms, as well as the identification of industry buyers and their product strategy.
Consistent with the arguments developed in Adams et al. (
2016) we have identified the knowledge heritage of the targets by their industry of origin and then identified their product strategy. Their product strategy was identified by their product line at entry under the assumption that, given their young age and relative lack of resources, new entrants will initially focus on a single product line.
Depending on their knowledge heritage, targets are defined as:
focal spinouts (186 or 47% of the sample) if they were founded by entrepreneurs whose last employment was in a semiconductor firm;
user-industry spinouts (108 or 27.4% of the sample) if they were founded by entrepreneurs who were previously employed in industries that use semiconductors as components in their final products;
other de-novo (101 or 25.6% of the sample). Depending on the product market they enter targets can operate in:
generic semiconductors (124 or 31.3% of the sample) if they produce devices used in a wide range of systems and designed without a specific application in mind,
specific semiconductors (199 or 50% of the sample) if they work on solutions tailored to specific users in specific submarkets (i.e. computing, communication, storage),
other semiconductors (72 or 18.7% of the sample) if they are active in other semiconductor areas such as customized ASIC design services and electronic design automation (EDA) tools. Table
1 reports the breakdown of start-ups by knowledge heritage and type of product.
Table 1
Breakdown of the sample by knowledge heritage and type of product
Focal spinout | 70 | 84 | 32 | 186 |
User-industry spinout | 19 | 80 | 9 | 108 |
Other de-novo | 35 | 35 | 31 | 101 |
Total | 124 | 199 | 72 | 395 |
A relatively higher share of user-industry spinouts (80/108 or 74% of the total in this category) enter specific semiconductors. Focal spinouts also enter into specific product categories (84/186 or 45% of the entrants in this category), although they tend to enter into generic semiconductors more than user-industry spinouts (70/186 or 37.6% vs. 19/108 or 17% for the latter). Indeed, focal spinouts account for the majority of entrants into generic products (70/124 or 56%). It is interesting to note that ‘other de-novo’ entrants distribute almost equally across the three categories of products (35% in generic and specific, 30% in other semiconductors) a finding which is consistent with the idea that knowledge heritage does not guide their product entry strategy.
Buyer firms were first coded on the basis of their industry of activity as identified by their primary SIC or NAICS code and then assigned to one of the following three macro-categories using the logic explained in Adams et al. (
2013):
semiconductor buyer (211 or 53% of buyers) if they belong to SIC code 3674 (until year 1999) and NAICS code 33441 for the subsequent years;
user buyer (120 or 30.3% of buyers) if they belong to one of the following downstream industries: industrial and commercial machinery (also including electric, gas, and sanitary services); computer and office equipment; electronic and other electrical equipment and components; telecommunications (also including communication services); automotive (also including transportation equipment); instrumentation (also including medical instruments); aerospace and defense
5;
other buyer (64 or 16.7% of buyers).
To test hypotheses H4a and H4b semiconductor buyers have also been assigned to one of the following categories according to their own product strategy:
semiconductor generic-buyers (77 or 36.6% of semiconductor buyers);
semiconductor specific-buyers (109 or 51.6% of semiconductor buyers);
semiconductor other-buyers (25 or 11.8% of semiconductor buyers).
6 Table
2 reports the breakdown of the start-ups in our sample by type of buyer. This table allows us to make some preliminary observations concerning the relationship between the knowledge heritage of the target firms and the type of buyer.
Table 2
Breakdown of the sample by type of entrant, type of product, and type of buyer
Type of firm |
Focal spinouts |
Generic | 13 | 19 | 13 | 20 | 5 | 70 |
Specific | 19 | 28 | 1 | 30 | 6 | 84 |
Other | 3 | 2 | 2 | 7 | 18 | 32 |
User-industry spinouts |
Generic | 8 | 4 | 0 | 5 | 2 | 19 |
Specific | 12 | 31 | 3 | 25 | 9 | 80 |
Other | 1 | 1 | 0 | 3 | 2 | 9 |
Other de-novo |
Generic | 7 | 6 | 3 | 13 | 6 | 35 |
Specific | 8 | 13 | 1 | 10 | 3 | 35 |
Other | 5 | 4 | 2 | 7 | 13 | 31 |
Total | 76 | 108 | 25 | 120 | 64 | 395 |
The first observation concerns the distribution of the potential targets across type of buyers. The data shows that focal spinouts distribute more evenly than user-industry spinouts. Indeed, 65/99 (or 66%) of user-industry spinouts are acquired by only two types of buyers (semiconductor firms producing market specific product or user buyers). This number is lower for focal spinouts (97/154 or 63%) suggesting that focal spinouts appealing to more types of buyers. The second observation refers to the ‘destination industry’ of user-industry spinouts active in specific products. Such spinouts seem relatively more likely to be acquired by two types of buyers: semiconductor firms active in specific product categories and user-industry firms. Together these two types of buyers account for 56/80 (or 70%) of the acquisitions of user-industry spinouts active in specific semiconductors.
Our list of controls includes two sets of variables. The first set accounts for the characteristics of the target firm and its founder. We control for the
size of the target firm as measured by the (logarithm of) total number of employees at the time of entry. Prior studies have found that there are benefits from firm size as relatively larger startups generally experience higher survival rates. However, prior studies have also found a positive correlation between size and exit by acquisition for those startups that do not survive (Arora & Nandkumar,
2011). Thus, we consider firm size a proxy for the quality of the human capital of the target firm. We also account for potential diminishing return to size of the target by including a squared-term in our specifications.
7
An additional measure of the quality of a venture is the innovative activity of the target firm. Prior studies have highlighted that innovative firms (if they do not survive) have a higher probability of exiting by acquisition (Arora & Nandkumar,
2011; Cockburn & MacGarvie,
2011; Hsu & Ziedonis,
2013). In our study, we measure innovativeness in terms of patenting. Specifically, the variable
patent is equal to one if, at entry, the startup had filed at least one patent with the US Patent and Trademark Office (USPTO), and zero otherwise. Finally, we control for whether or not the target semiconductor firm is a fabless. The dummy variable
fabless is equal to one if the firm is only engaged in design and zero otherwise.
Our controls for the characteristics of the founder(s) include:
serial entrepreneur which is equal to one if the founder, or a member of the founding team, has previously founded another firm and zero otherwise;
founding team which is equal to one if the target firm is founded by a team of employees, and zero otherwise
8;
PhD which is equal to one if at least one member of the founding team possesses a doctoral degree. Indeed, previous research (Roberts,
1991) suggests that diminishing returns to firm’s performance are triggered by an excessive amount of higher education.
The second set of controls includes the characteristics of the acquisition deal. In line with prior research on acquisitions (Puranam et al.,
2003) we control for the size of the acquisition deal.
Full acquisition is equal to one if the deal involved 100% of the target firm (370 cases or 93.6% of the total) and zero otherwise.
Finally, in each regression we control for fixed effects linked to time of entry and exit, location of the target, and characteristics of the buyer not related to its industry and/or product strategy.
The complete list of our variables and their summary statistics are reported in Table
3Table 3
Variables list and summary statistics
Semi buyer | Dummy | 395 | 0.534 | 0.499 | 0 | 1 |
User buyer | Dummy | 395 | 0.304 | 0.460 | 0 | 1 |
Other buyer | Dummy | 395 | 0.162 | 0.369 | 0 | 1 |
Semi generic buyer | Dummy | 395 | 0.195 | 0.397 | 0 | 1 |
Semi specific buyer | Dummy | 395 | 0.276 | 0.448 | 0 | 1 |
Semi other buyer | Dummy | 395 | 0.063 | 0.244 | 0 | 1 |
Focal spinout | Dummy | 395 | 0.471 | 0.500 | 0 | 1 |
User-industry spinouts | Dummy | 395 | 0.273 | 0.446 | 0 | 1 |
Other de novo | Dummy | 395 | 0.256 | 0.437 | 0 | 1 |
Generic semi | Dummy | 395 | 0.314 | 0.465 | 0 | 1 |
Specific semi | Dummy | 395 | 0.504 | 0.501 | 0 | 1 |
Other semi | Dummy | 395 | 0.182 | 0.387 | 0 | 1 |
Full acquisition | Dummy | 395 | 0.937 | 0.244 | 0 | 1 |
Fabless | Dummy | 395 | 0.476 | 0.500 | 0 | 1 |
PhD | Dummy | 395 | 0.387 | 0.488 | 0 | 1 |
Serial entrepreneur | Dummy | 395 | 0.253 | 0.435 | 0 | 1 |
Founding team | Dummy | 395 | 0.587 | 0.493 | 0 | 1 |
Patent | Dummy | 395 | 0.372 | 0.484 | 0 | 1 |
Size (Ln) | Continuous | 395 | 3.227 | 1.000 | 0 | 4.564 |
Size sq (Ln) | Continuous | 395 | 11.413 | 4.413 | 0 | 20.833 |
Exit year | Continuous | 395 | 2006.37 | 4.156 | 1997 | 2014 |
Entry year | Continuous | 395 | 2000.304 | 2.395 | 1997 | 2007 |
Most of the firms in our sample (58.7%) have been founded by a team of entrepreneurs. In one fourth of the cases, at least one of the founders had prior experience as a founder (i.e. a ‘serial entrepreneur’) and in 38% of the cases one of the founders held a PhD. The average size of the firms in our sample is 25 employees in year 1 activity, while the average age is 6 years by the time of acquisition. The largest firm has 95 employees at the time of entry, which satisfies our threshold for being considered a startup (i.e. less than 100 employees). Forty-eight percent of the firms in the sample are fabless and only a minority (37%) have filed for a patent at the USPTO at the time of entry. The latter evidence is consistent with the idea that firms in our sample possess few assets, other than their knowledge heritage, to offer buyers.
4.4 Results
We present the results from the first estimate in Table
5.
Table 5
Multinomial Logit regression – acquisition by categories of buyers
Focal spinout | Generic semi | 2.311 | [0.782]*** | 2.057 | [0.796]** |
| Specific semi | 1.789 | [0.633]*** | 1.956 | [0.668]*** |
| Other semi | − 1.051 | [0.725] | − 0.689 | [0.676] |
User-industry spinout | Generic semi | 1.478 | [0.954] | 1.206 | [1.073] |
| Specific semi | 1.336 | [0.666]*** | 1.318 | [0.711]* |
| Other semi | 0.730 | [1.080] | 1.166 | [1.069] |
Other de novo | Generic semi | 0.732 | [0.748] | 1.122 | [0.780] |
| Specific semi | 2.325 | [0.794]*** | 2.048 | [0.842]** |
| Other semi | Ref | Ref | Ref | Ref |
Fully acquired | | 1.342 | [0.620] | 0.994 | [0.640] |
Fabless | | 0.509 | [0.385] | 0.331 | [0.414] |
PhD | | − 0.526 | [0.356] | − 0.389 | [0.392] |
Serial entrepreneur | | 0.046 | [0.418] | 0.024 | [0.435] |
Founding team | | − 0.199 | [0.413] | − 0.562 | [0.425] |
Patent | | 0.290 | [0.343] | 0.118 | [0.363] |
Size (Ln) | | − 2.049 | [0.803]** | − 1.407 | [0.811]* |
Size sq (Ln) | | 0.460 | [0.145]*** | 0.281 | [0.149]* |
Constant | | 1.009 | [1.530] | 0.879 | [1.564] |
# of observations | | 395 |
Log-pseudolikelihood | | − 338.516 |
Chisq | | 146.507*** |
Pseudo Rsq | | 0.1359 |
Columns (1) and (2) report the regression coefficients for those startups that have been acquired by Semiconductor buyers and User industry buyers respectively. The reference category is ‘Other buyers’.
In column (1) the coefficient estimate of focal spinouts producing generic products is positive and significant indicating that these firms are likely to be acquired by semiconductor buyers. The coefficient is also positive and significant in column (2), indicating that focal spinouts are also likely to be acquired by user buyers. These results are in line with H1.
The coefficient estimates of user-industry spinouts producing specific products is positive and significant in columns (1) and (2). These results are consistent with H2 as they indicate that spinouts from a focal industry that produce market-specific products are likely to be acquired by buyer firms in both the focal industry and in user-industries. The coefficient estimates of user-industry spinouts producing generic products is not significant, suggesting that these firms are not more (or less) likely than other de-novo firms to be acquired by either firms in the focal industry or by firms in user industries. This result is in line with H3.
Concerning our control variables, we find a non-linear relationship between firm size and the probability of being acquired by both semiconductor and user buyers. None of the remaining control variables is significant. Although this result might seem surprising, it should not be taken as an indication that the characteristics of the founders (as measured by serial entrepreneur) or firm quality (as measured by patents) do not affect acquisition. Both of them are important drivers of acquisition, but they do not seem relevant to discriminate between types of buyers.
To better understand the implications of our estimations we need to bear in mind that the estimated coefficients in Table
5, when exponentiated, represent the relative probabilities for a specific category of start-up to be acquired by a certain type of buyer
with respect to the reference category (i.e. ‘other buyers’). We can assess and compare the overall probability, instead of the relative probability, by computing the marginal effects. These results are summarized in Table
6.
9Table 6
Marginal effects for acquisition by categories of buyers
Type of startup | Interaction | M. Effect | M. Effect |
Focal spinout | Generic semi | 0.1819715 | 0.0415166 |
| Specific semi | 0.0897535 | 0.100831 |
User-industry spinout | Generic semi | 0.1353809 | 0.0033491 |
| Specific semi | 0.0871673 | 0.0480197 |
The probability to be acquired by semiconductors firms is about 18% points higher for focal spinouts producing generic products (than for other de-novo startups producing ‘other semiconductors’); 9% points higher for focal spinouts producing market-specific products; 13.5% points higher for user-industry spinouts producing generic products, and 8.7% points higher for user-industry spinouts producing market-specific products. The probability to be acquired by a user buyer is about 4% points higher for focal spinouts producing generic products, 10% points higher for focal spinouts producing market-specific products, 0.3% points higher for user-industry spinouts producing generic products (albeit the estimated coefficient in this case is not statistically significant from zero), and 4.8% points higher for user-industry spinouts producing market-specific products.
To test the remaining hypotheses we need to account explicitly for the product strategy of the buyer firms. We do this in Table
7 where we distinguish four types of buyers: Semiconductor buyers producing generic products (column 3), Semiconductor buyers producing market specific products (column 4), Semiconductor buyers producing other products (column 5), and User buyers (column 6). As in the previous case, the reference category is ‘Other buyers’.
Table 7
Multinomial Logit regression–acquisition by categories of buyers and product strategy
Focal spinouts | Generic semi | 1.735 | [0.892]* | 2.444 | [0.957]** | 3.127 | [1.072]*** | 1.996 | [0.792]** |
| Specific semi | 1.685 | [0.755]** | 2.291 | [0.845]*** | 0.171 | [1.388] | 1.966 | [0.664]*** |
| Other semi | − 1.178 | [0.947] | − 1.241 | [1.152] | − 0.575 | [1.060] | − 0.713 | [0.680] |
User-industry spinouts | Generic semi | 1.738 | [1.061] | 1.493 | [1.161] | − 11.112 | [1.169]*** | 1.180 | [1.065] |
| Specific semi | 0.758 | [0.807] | 2.035 | [0.876]** | 0.614 | [1.010] | 1.330 | [0.710]* |
| Other semi | 0.895 | [1.260] | 1.029 | [1.276] | − 11.152 | [1.115]*** | 1.159 | [1.059] |
Other de novo | Generic semi | 0.617 | [0.911] | 0.856 | [0.995] | 1.031 | [1.100] | 1.106 | [0.780] |
| Specific semi | 2.026 | [0.930]** | 3.073 | [1.024]*** | 0.808 | [1.501] | 2.068 | [0.852]** |
| Other semi | Ref | Ref | Ref | Ref | Ref | Ref | Ref | Ref |
Fully acquired | | 1.966 | [0.866]** | 0.928 | [0.720] | 1.581 | [1.027] | 0.996 | [0.644] |
Fabless | | 0.478 | [0.437] | 0.625 | [0.416] | 0.041 | [0.621] | 0.333 | [0.416] |
PhD | | − 0.443 | [0.405] | − 0.774 | [0.399]* | − 0.114 | [0.531] | − 0.393 | [0.391] |
Serial entrepreneur | | 1.357 | [0.492] | 0.281 | [0.455] | − 0.691 | [0.631] | 0.045 | [0.440] |
Founding team | | − 0.147 | [0.462] | − 0.352 | [0.463] | 0.138 | [0.542] | − 0.595 | [0.430] |
Patent | | 0.440 | [0.405] | 0.201 | [0.387] | 0.292 | [0.552] | 0.020 | [0.368] |
Size (Ln) | | − 1.852 | [0.938]** | − 2.445 | [0.872]*** | − 0.375 | [1.118] | − 1.479 | [0.830]* |
Size sq (Ln) | | 0.416 | [0.170]** | 0.538 | [0.164]*** | 0.188 | [0.237] | 0.293 | [0.151]* |
Constant | | − 0.341 | [1.848] | 0.885 | [1.730] | − 5.654 | [2.116]*** | 1.027 | [1.606] |
# of observations | | 395 |
Log-pseudolikelihood | | − 499.454 |
Chisq | | 3868.88*** |
Pseudo Rsq | | 0.1601 |
Accounting for the product strategy of the buyer allows us to highlight similarities and differences with respect to the prior analysis. On the one hand, the coefficient estimates of focal spinouts producing generic and specific products is still positive and significant in columns (3), (4) and (6), indicating that these firms are likely to be acquired by all types of ‘Semiconductor buyers’ as well as by ‘User buyers’. By contrast, the coefficient estimates for user-industry spinouts producing specific products are not significant in column (3), but positive and significant in columns (4) and (6). This finding suggests that spinouts from user industries that produce market-specific products are not likely to be acquired by buyer firms in the focal industry that produce generic products. This is consistent with H4a. It also indicates that spinouts from user industries that produce market-specific products are likely to be acquired by buyer firms in the focal industry that produce market specific products and by buyer firms in user industries. Both results provide support for H4b. In particular, the computed marginal effects for these two cases indicate that the probability to be acquired by a buyer in the focal industry that produces market specific products is about 20% points higher for a user-industry spinout producing market-specific products (than for other startups producing other semiconductors). For the same category of firms, the probability to be acquired by a user-industry buyer is 3.7% points higher.
Overall, our empirical analysis provides two sets of results. First, we find evidence that the knowledge heritage of spinouts influences their probability to be acquired by different types of buyers. Second, we find evidence that product strategy plays a role particularly for the acquisition of spinouts that enter the focal industry from downstream industries. Indeed, only user-industry spinouts that produce market-specific products are likely to be acquired, and their buyers are limited to firms that are active in market-specific product categories.
Additional analysis is conducted to check for possible biases in our results. Our analysis considers the case of a firm (the target) that must choose among a set of potential buyers. It might be argued, however, that the choice of the buyer is indeed consequential to the choice of being acquired versus remaining an independent entity. If this is the case, then failing to account explicitly for this possibility would introduce a sample selection bias in our results. To overcome this potential problem we re-estimate our model using a two-step procedure similar to Heckman (
1979). Specifically, we first estimate a binary Logit model on the overall sample of 936 semiconductor firms. The dependent variable in this case is equal to one if the firm has been acquired and zero otherwise.
10 We then use the residuals from the first stage regression to compute inverse Mills’ ratios that are added as covariates in the multinomial regression in the second stage. The results of the second stage estimations are reported in Tables
8 and
9.
Table 8
Robustness check—multinomial Logit regression–acquisition by categories of buyers (second stage Heckman selection model)
Focal spinout | Generic semi | 2.391 | [0.789]*** | 2.001 | [0.799]** |
| Specific semi | 1.848 | [0.648]*** | 1.925 | [0.674]*** |
| Other semi | − 1.037 | [0.731] | − 0.718 | [0.683] |
User-industry spinout | Generic semi | 1.475 | [0.954] | 1.180 | [1.071] |
| Specific semi | 1.360 | [0.670]*** | 1.304 | [0.710]* |
| Other semi | 0.853 | [1.108] | 1.122 | [1.081] |
Other de novo | Generic semi | 0.774 | [0.748] | 1.055 | [0.779] |
| Specific semi | 2.367 | [0.792]*** | 2.000 | [0.835]** |
| Other semi | Ref | Ref | Ref | Ref |
Fully acquired | | 1.387 | [0.627]** | 0.963 | [0.640] |
Fabless | | 1.024 | [0.482]** | 0.242 | [0.511] |
PhD | | − 0.749 | [0.371]** | − 0.339 | [0.399] |
Serial entrepreneur | | 0.184 | [0.427] | 0.019 | [0.444] |
Founding team | | 0.626 | [0.622] | − 0.710 | [0.649] |
Patent | | 0.744 | [0.414]* | − 0.053 | [0.426] |
Size (Ln) | | − 2.932 | [1.044]*** | − 1.230 | [1.055] |
Size sq (Ln) | | 0.697 | [0.208]*** | 0.236 | [0.209] |
Inverse Mills' ratio | | 5.206 | [2.541]** | − 0.758 | [2.699] |
Constant | | − 2.922 | [2.339] | 1.479 | [2.429] |
# of observations | | 395 |
Log-pseudolikelihood | | − 335.709 |
Chisq | | 333.05*** |
Pseudo Rsq | | 0.1431 |
Table 9
Robustness check—multinomial Logit regression–acquisition by categories of buyers and product strategy (second stage Heckman selection model)
Focal spinouts | Generic semi | 1.553 | [0.807]* | 2.027 | [0.847]** | 2.872 | [1.007]*** | 1.791 | [0.735]** |
| Specific semi | 1.696 | [0.722]** | 2.212 | [0.781]*** | 0.063 | [1.332] | 1.984 | [0.657]*** |
| Other semi | − 0.981 | [0.856] | − 1.254 | [1.077] | − 0.503 | [1.090] | − 0.649 | [0.685] |
User-industry spinouts | Generic semi | 1.984 | [1.043]* | 1.466 | [1.093] | − 11.094 | [1.116]*** | 1.337 | [1.058] |
| Specific semi | 0.760 | [0.732] | 1.865 | [0.769]** | 0.574 | [0.969] | 1.302 | [0.668]* |
| Other semi | 0.916 | [1.175] | 1.041 | [1.177] | − 11.321 | [1.251]*** | 1.174 | [1.047] |
Other de novo | Generic semi | 0.703 | [0.869] | 0.736 | [0.883] | 0.885 | [1.062] | 1.156 | [0.716] |
| Specific semi | 1.999 | [0.887]** | 2.777 | [0.968]*** | 0.867 | [1.354] | 1.994 | [0.851]** |
| Other semi | Ref | Ref | Ref | Ref | Ref | Ref | Ref | Ref |
Fully acquired | | 1.746 | [0.885]** | 0.871 | [0.770] | 1.450 | [1.104] | 0.829 | [0.678] |
Fabless | | 0.550 | [0.433] | 0.732 | [0.413]* | 0.287 | [0.647] | 0.429 | [0.410] |
PhD | | − 0.405 | [0.395] | − 0.703 | [0.382]* | 0.073 | [0.516] | − 0.368 | [0.373] |
Serial entrepreneur | | 0.118 | [0.460] | 0.221 | [0.423] | − 0.921 | [0.670] | − 0.007 | [0.412] |
Founding team | | − 0.016 | [0.455] | − 0.147 | [0.445] | 0.214 | [0.575] | − 0.434 | [0.415] |
Patent | | 0.337 | [0.377] | 0.345 | [0.365] | 0.311 | [0.559] | 0.014 | [0.346] |
Size (Ln) | | − 1.707 | [0.921]* | − 2.283 | [0.864]*** | − 0.836 | [1.160] | − 1.445 | [0.837]* |
Size sq (Ln) | | 0.372 | [0.177]** | 0.482 | [0.168]*** | 0.253 | [0.270] | 0.274 | [0.160]* |
Inverse Mills' ratio | | 0.072 | [0.844] | 0.637 | [0.831] | 0.829 | [1.224] | 0.431 | [0.706] |
Constant | | − 0.803 | [1.971] | 0.068 | [1.761] | − 5.583 | [2.267]*** | 0.599 | [1.602] |
# of observations | | 395 |
Log-pseudolikelihood | | − 521.491 |
Chisq | | 1766.38*** |
Pseudo Rsq | | 0.1231 |
A comparison between these results and those of our primary analysis indicates that, in the case of acquisitions by categories of buyers (compare with Table
5), the sign and the statistical significance of the coefficients capturing the interaction between firm origin and product strategy is maintained for both semi buyer and user buyer. The magnitude of the estimated coefficients is slightly higher than in our baseline regressions. Also, when we account for the product strategy of buyers (compare with Table
7) the sign and the statistical significance of the coefficients is maintained. However, in this case, the coefficients are lower than in the baseline regression. It may be noted that the estimated coefficient of the inverse Mills' ratio is positive and statistically significant only in model (7) in Table
8, indicating that sample selection bias is a potential problem only for semiconductor buyers. To check whether this result might be the consequence of unadjusted standard errors due to the fact that our estimation procedure deviates from the traditional (i.e. linear) two-stage model, we reran the estimation presented in Table
8 via bootstrapping. Results of these further regressions produced non-significant coefficients for the inverse Mills' ratio that suggest that selection bias is not a concern in our data.
11