Notes

^1.

See Wagner (2007) for a survey of the literature. Based on a review of 54 studies for 34 countries published between 1995 and 2006, he concluded "exporters are found to be more productive than non-exporters, and the more productive firms self-select into export markets, while exporting does not necessarily improve productivity." For other recent surveys, see López (2005) and Greenaway and Kneller (2007).

^2.

Studies that have looked at the productivity performances of firms entering export markets as well as exiting include Baldwin and Gu (2003) for Canada, Clerides et al. (1998) for Colombia, Bernard and Wagner (1997) for Germany, and Girma, Greenaway and Kneller (2003) for the United Kingdom. For a complete list, please refer to Table A1 in Wagner (2007).

^3.

As Dixit (1989) explained, drawing on the financial market literature on options pricing, given uncertainty about future tariffs and real exchange rates, the decision of firms to enter export markets is equivalent to a select group of plants with superior chances of succeeding in export markets choosing to exercise the option to experiment in these markets.

^4.

The survey data are derived from long-form questionnaires (often given to larger plants) and short-form questionnaires (often given to smaller plants). The long-form questionnaires contain much more detailed information than the short-form questionnaires.

^5.

For the post-2000 period, plants used in the analysis consist of those that fill in the long form and those whose data are from tax records. The former are typically larger plants, while the latter smaller ones.

^6.

According to a 1974 survey that collected export data for all plants, only 0.4% of plants that filled in the short-form questionnaires reported exports (Baldwin and Gu, 2003).

^7.

More specifically, they are excluded except in section 4.1, where we calculate the total entry and exit rate, and the total export participation rate.

^8.

Real value-added is calculated using corresponding industry deflators.

^9.

We are grateful to Alla Lileeva for providing us with the tariff data. For details on the sources and construction of the tariff data, see the Appendix in Trefler (2004).

^10.

Other studies have used an alternative industry-specific real exchange rate, generated by calculating the weighted average of exchange rates between Canada and its trading partners, with weights being countries' trade shares for each industry (Baggs et al., 2009). There are two problems with this approach. First, for Canada, trade-weighted industry-specific real exchange rates show little variability across industries since the U.S. trade weight dominates across manufacturing industries. Secondly, this approach assumes the same price adjustments to nominal exchange-rate movements across industries. However, Baldwin and Yan (2007, 2008) find a high degree of heterogeneity in industries' responses. The price-adjusted real exchange rate is a better indicator of an industry's international competitiveness. It measures the price spread between an industry's product price and the landed price charged by industries in other countries.

^11.

To prevent possible endogeneity, we measure industry-specific real gross output as the sum of real shipments at the 4-digit SIC level minus the real shipment of the plant itself.

^12.

This is because the statistical software packages, such as STATA's mfx and dprobit commands, do not know that a variable is an interaction term and thus do not take the full derivative. As a result, when a variable is interacted with another (or has higher order terms) in a nonlinear model, mfx and dprobit will give the wrong marginal effect of the interaction term. Instead, the marginal effect of the interaction term requires computing the cross derivative or cross difference as defined in equation (2).

^13.

Baldwin and Yan (2010) find a tariff reduction increases the probability that plants will close down completely, in particular for exporters. Here we further show that tariff reduction does not impact on the decision of an exporter to become a non-exporter among continuing plants.

^14.

The "treatment" terminology derives from medical experiments assessing the effects of new drugs or medical procedures using randomly assigned treated and control groups to allow accurate identification of the effect of the drug or procedure being tested. In the present application, given the absence of a randomly assigned control group, propensity scoring is used to construct such a control group.

^15.

We use one-to-one nearest neighbour matching without replacement and with common support (i.e., there are both treated and non-treated plants for each characteristic which we want to compare. If the common support is not satisfied in the treatment group, then these plants are dropped from the sample).