al., 2004; and Girma, et. al., 2005). Therefore we only include data on UK-owned
firms in our subsequent analyses.
IV. Econometric Estimation Methods
To obtain estimate of total factor productivity (TFP), we firstly estimate an augmented
production function as follows:
yit =α0 +αEeit +αMmit +αKkit +αTt+γXit +εt (1)
where y , e , m , and k refer to the logarithms of real gross output, employment,
intermediate inputs and tangible assets in firm i in time t. We have also included a
vector of variables, X , that determine TFP; hence TFP growth in this instance is
defined as (dropping sub-scripts):
InTFp = α^t + γx ≡ y - α^Ee - α^Mm - α^Kk (2)
Since the problem under consideration is to understand the causes of TFP (e.g. the
role of exporting), the preferred approach to estimating TFP is to directly include the
determinants of output (and thus TFP) into the production function, as this avoids any
problems of statistical inefficiency and omitted variable bias associated with
estimating a two-stage model using a growth accounting approach20. Moreover, this
method also allows us to directly test whether such determinants are statistically
significant.
In estimating models to determine the linkages between exporting and productivity
using micro level data, the issues of primary concern are the problems of endogeneity
20 Refer to Harris (2005) for a detailed discussion of the issues concerning the measurement of TFP,
such as inefficient estimates (Newey and McFadden, 1999, Section 6), omitted variable problem (Wang
and Schmidt, 2002) when using a two-stage model and the endogeneity of inputs and outputs in a
production model.
15