classifications used for the panel mergers analysis are specified in table 5b.14 The
following model specification was used to test the effect of lagged NPIs on mergers.
Mergersit = α + β NPIi,t-k+ δ X + εt
where the subscript i denotes the aggregate food industry for the aggregate analysis and
food industry segment i for the panel analysis.
Mergersit : Number of mergers in the USA food processing industry i in the year t
NPIi,t-k : Number of new product introductions in the USA food processing industry i in
the year t-k.
X: contains different combinations of the following variables
- total number of mergers in the US economy15 in the year t,
- number of firms in the food industry i in the year t-k.
- sales growth in the food industry i in the year t-k.
We tried sixteen variations of the model using different combinations of the above
variables. We tried alternate values of k ranging between 1 and 7.
Results: The results of the aggregate analysis and the grouped analysis are
presented in Table 6A and 6B respectively. The tests for autocorrelation do not reject the
null of no serial correlation. The results correspond to OLS estimates for the aggregate
models. In case of the panel analysis, the number of mergers is very small in several
cases. Count data models can better handle the discrete nature of the dependent variable
with small values. Hence we used Poisson estimation for the panel analysis of mergers.
14 This classification excludes the diversified mergers category. Hence it does not incorporate all the
mergers that are included in the aggregate analysis. We attempted to allocate the diversified category of
mergers to the categories corresponding to the NPIs based on the sales information of diversified (SIC
2000) firms. The adhoc allocation method did not yield any significant results. Since we do not have very
strong faith in the allocation criteria we have refrained from presenting those results here.
15 Data Source: Mergerstat Review
13