Global Production Chains and Export Survival

Abstract

In this article, we empirically investigate the role of global production chains (GPCs) in export survival by using highly disaggregated machinery exports data from Turkey for the period 1998–2013. We contribute to the literature by offering alternative indicators adopted from intra-industry trade to represent more effectively the extent of GPC in a survival analysis. Based on the estimation of a discrete-time logit model with random effects, we find that GPCs have a strong positive influence on the survival probability of Turkey’s machinery exports. This impact is even stronger if trade involves vertically differentiated trade and parts and components in particular.

Keywords:

• export duration
• fragmentation of production
• global production chains
• parts and components trade
• survival analysis

JEL Classification:

• F10
• F14
• C41

Notes

1. The prevalence of longer relations in P&C trade has been also reported (Corcoles, Diaz-Mora, and Gandoy 2014, 2015; Diaz-Mora, Corcoles, and Gandoy 2015 ; Shao, Xu, and Qui 2012).

2. The proposed indicator was first used in Türkcan and Saygılı (2018) to investigate the impacts of economic integration agreements on the survival rates of Turkey’s P&C exports. While the analysis in this article is built on the empirical methodology of Türkcan and Saygılı (2018), it differs in two substantial ways: (1) the focus is on the role of GPCs on export survival; (2) we use a more comprehensive set of explanatory variables to make better assessments of the determinants of export survival.

3. We would like to thank the anonymous referee in contributing to this issue.

4. According to the Exporter Dynamics Database of the World Bank, the number of exporting firms increased from 30,000 to 48,000, while the number of exporters per export destination increased from 500 to 1,000 between 2002 and 2010. Meanwhile, the share of the top ten markets in Turkey’s total exports decreased from 62% in 2000 to 48% in 2010.

5. See, for instance, Albornoz, Fanelli, and Hallak (2016); Alvarez and Lopez (2008); Cadot et al. (2013); Diaz-Mora, Corcoles, and Gandoy (2015); Esteve-Perez, Requena-Silvente, and Pallardo-Lopez (2013); Esteve-Perez et al. (2007), Görg, Kneller, and Muraközy (2012); Fugazza and McLaren (2014); Gullstrand and Persson (2015); Ilmakunnas and Nurmi (2010); Stirbat, Record, and Nghardsaysone (2015); Volpe-Martincus and Carballo (2009).

6. See Gaulier and Zignago (2010), for a detailed description of this database.

7. The BACI database does not include flows below US$1,000.

8. Kaminski and Ng (2006) report that the share of medium- and high-tech products in Turkish exports to the EU-25 increased from 13.3% in 1995 to 37.3% in 2004, whereas the share of low-tech laborintensive products dropped from 69.6% to 46.5% in the same period. However, Gros and Selçuki (2013) report that low-tech products continue to play a major role in Turkish exports.

9. We omit these results here for brevity.

10. From Table 1, it can be seen that left-censoring data spells represent around 11% of all spells, whereas right-censoring spells account for about 30% of all spells.

11. Bilateral nominal exchange rates (i.e., Turkish Lira per importer currency) were computed by using bilateral nominal exchange rates (i.e., foreign currency per U.S. dollar) taken from the World Bank. Note that an increase reflects a depreciation of the Turkish Lira against the importer currency. In the second step, bilateral real exchange rates between Turkey and its trading partners were constructed by deflating nominal rates by using national consumer price indices (CPI) available from the World Bank. Bilateral real exchange rates were normalized by the average real exchange rates before computing annual percentage changes in relative real exchange rates (in logarithmic terms).

12. Ideally, intrafirm trade statistics would be used to assess the role of GPCs on export survival. Unfortunately, such data are also not available at the required level of detail.

13. Unit values at the 6-digit product level of the HS are constructed as the value of imports and exports of the product divided by the corresponding quantities.

14. The unit values ratio is zero when Turkey does not export or import the same HS item. It is also possible that when Turkey has either an export or import relationship, total trade will be defined as inter-industry trade (i.e., IIT equal to zero). IIT consists of two parts: vertical and horizontal. When the VD dummy is zero, it either means there is no IIT trade or there is some IIT that does not fit our criteria of VD (possibly HIIT).

15. We also explored the robustness of the results by using a 15% threshold. The results were consistent with those in the main text. For brevity, we do not report these results, but they are available from the authors upon request.

16. Of 257,791 bilateral observations for P&C from 1998 to 2013, a total of 169,720 observations were classified as HD export flows, whereas 88,062 observations were VD export flows (about 34% of the total observations).

17. For a given covariate, when the odds ratio is greater than 1, an increase in an extra unit of this covariate increases the risk of export failure by (odds ratio-1) percent; when the odds ratio is less than 1, an increase in an extra unit of covariate decreases the risk of export failure by (1-odds ratio) percent.

18. We tested the symmetry of the estimated coefficients for cases that “Log abs. difference in PCGDP” is greater than that of the partner country or less than that of the partner country. The logit model was estimated separately by using two alternative definitions of the “Log abs. difference in PCGDP.” The regression results showed that the direction and significance of both variables were quite similar, indicating a symmetrical pattern. In addition, we applied the “suest” procedure in STATA to test the equality of coefficients across these two models. Because the test results further proved that coefficients of both variables do not significantly differ at the 1% level [χ² (1) = 5.00, p-value > 0.01], we decided to retain our original empirical model.

19. This result is consistent with the recent findings of Frede and Yetkiner (2017), who report a negative relationship between the EU-Turkey Customs Union and Turkey’s exports.