A second approach to dealing with self-selection bias is instrumental variable (IV)
estimation25. If a variable(s) can be found that affects whether a firm engages in
exporting but does not affect outcomes directly then such a variable(s) can be used to
instrument for the treatment and overcome the problem of self-selection. The main
issue in practice is finding an appropriate instrument(s).
In terms of the data available in FAME, the likely candidates as instruments are the
age of the firm and whether it possesses any intangible assets. Firm age is not usually
included in the production function, as the capital stock is presumed to provide an
adequate measure of the vintage of the assets used in production. As to intangible
assets (such as R&D and advertising), we follow the standard approach in the IO
literature and presume that most (sunk cost) investment in intangibles is to overcome
existing barriers to entry into new markets (see Carlton, 2005). Thus intangible assets
feature in Equation (3). Evidence in favour of this approach is based on estimating
industry-level production functions, where we find that these variables are always
statistically insignificant determinants of (real) gross output, having controlled for the
other covariates in the model, but they are usually highly significant in determining
whether the firm sells overseas. Consequently, we include the logarithms of age and a
dummy variable to indicate whether intangible assets are possessed, as part of the
instrument set when estimating the following dynamic panel-data model, allowing for
an autoregressive error term:
lnYit =β0+∑4 π1jxjit +∑4 π2jxji,t-1 +∑4 ∑4 π3j(Dlxjit)+π4lnYi,t-1
j=1 j=1 l=1 j=1
4 1 4 11 x
+∑∑γsDli,t-s+∑βlDl+∑δnREGn+∑τpINDp
l=2 s = -1 l=1 n=1p =1
+ηi +tt +(1-ρ)eit (5)
25 To our knowledge, there are few studies utilising instrumental variable estimation to examine the
causality between export and productivity, probably due to a lack of appropriate instruments.
18