

臺大管理論叢
第
27
卷第
3
期
117
Hypothesis 4: Exploitation mediates the relationship between prior exploration and
firm performance.
3. Method
3.1 Data Sample
We chose firms in the semiconductor and related device industry in the United States
(SIC 3674) for our sample. We chose this industry for two reasons. First, it exhibits high
levels of both exploration and exploitation. For example, we learned from our field
interviews with executives in the semiconductor industry that the reduction they achieved in
the unit cost of their integrated circuits (ICs) is the main source of their competitive
advantage. This was especially true for foundry and Dynamic Random Access Memory
(DRAM) firms. Reducing the unit size of their wafers and the distance between circuit lines
were two approaches they used to reduce the unit cost of their ICs. The investment in these
two activities was huge
2
and irreversible. Therefore, most of the companies in this industry
phase-in their investments, a ROR application that minimizes their investment risk early on.
Second, the semiconductor industry experienced tremendous cyclicality during our sample
period. This circumstance provides fertile ground for testing ROR because the uncertainty
should increase the value of the options.
All of the financial data were taken from Standard & Poor’s Compustat North America
database. We initially collected 2,633 observations from 255 firms. We then eliminated firms
with annual sales less than 10 million to mitigate small-firm bias. Because we wanted to
study the effects over a relatively long period of time, we also eliminated firms with less than
eight consecutive years of R&D expense data. To add uncertainty to our models, we
collected monthly stock price data for each firm from the Center for Research in Security
Price (CRSP) database. After this pruning, the final sample consisted of 526 observations
from 63 firms.
Data on the ages of the firms (how long they had been in existence) were obtained by
searching the Factiva news database and the firms’ websites. Finally, we obtained shipment
data for the worldwide semiconductor industry from 1976 to the present from the global
Semiconductor Industry Association (SIA) billing report.
2 The capital investment of Motorola Semiconductor Product Sector, for example, was $2.4 billion in 2000
(Motorola Inc., 2001).