Trade and Infrastructure in the Belt and Road Initiative: A Gravity Analysis Based on Revealed Trade Preferences

Round 1

Reviewer 1 Report

Please see file attached

Comments for author File: Comments.pdf

Author Response

We are very grateful for all your comments, which we believe have allowed us to improve the quality and clarity of our paper.

We submit you our detailed answers to each of your comments below.

 

1. First of all, the choice to use as a dependent a measure of “revealed trade preference” (RTP) is not sufficiently motivated in your case.

1.A) My first concern is that, by using a RTP measure, which is a standardized measure, you will fail to measure the effect of infrastructures on the intensity of trade (as you assert in the manuscript), because RTP does not measure the intensity of trade, but the relative preference. I have the idea that RBI is intended to increase the volume of trade, not just change the relative preference of a certain country to import from another certain country.

1.B) Moreover, I think that, using RTP, you are going to disregard the separation among intensive and extensive margins of trade. Correctly modelling intensive and extensive margins in the gravity of trade is possible with the Zero inflated methods by Silva and Tenreyro (Silva, J. S., & Tenreyro, S. (2006). The log of gravity. The Review of Economics and statistics, 88(4), 641-658.), of using Heckman 2-part modelling strategy.

 

Response 1: We have tried to better explain our choice of using RTP indices as the dependent variable of our gravity equation. In our view, RTP indices represent a simple way to control for multilateral trade resistance and, at the same time, to prevent the problem created by the negative relationship between country size and trade openness, which affects traditional gravity equations using GDP as a measure of country size.

1.A) By ‘trade intensity’ we do not mean the value of bilateral trade, but its relative importance with respect to the geographic neutrality threshold. We agree that BRI infrastructure investment is intended to increase the volume of total trade more than its geographic direction. Yet, we are interested in better understanding to what extent it could also affect bilateral trade preferences.

1.B) Distinguishing between intensive and extensive margins of trade goes beyond the scope of our paper. We do not consider country pairs that do not report reciprocal trade, and therefore we are not exposed to the econometric problems discussed by Silva and Tenreyro (2006) for zero-value observations. Furthermore, by relying on the RTP index (instead of the value of bilateral trade) as the dependent variable of our regression, we do not need to resort to its logarithmic form.

 

 

 

2. I’m a bit skeptical about splitting the regression exercise in two part (i.e. using the residuals of the first step regression as dependent in the second step). The choice is not sufficiently

motivated. I also think that references are not up-to-date. Based on Bassanini and Scarpetta (2003) or other works, are you able to discuss possible “cons” of using estimated residuals as a dependent in a second step regression (e.g. endogeneity, violation of classical regression assumption)?

In general, I do not think a two-step strategy is needed here. Could you please better motivate your choice?

 

Response 2: We have tried to better motivate our two-step strategy in the following way. The first step is based on a gravity equation in which revealed bilateral trade preferences are explained using only "dyadic" explanatory variables (specific to the country pair) as independent variables, which may be related to bilateral trade costs.

The resulting estimated residuals can be considered as a ‘gravity-adjusted’ measure of revealed trade preferences, that is a measure of the untapped trade potential, which is not explained by dyadic trade cost variables.
Our second equation is aimed at testing to what extent gravity-adjusted RTPs are influenced by the quality of infrastructure, which is a non-dyadic (country-specific) variable, as well as by any other country-specific fixed effect.

Regarding the references, we have removed Bassanini and Scarpetta (2001). Although the idea of a two-step regression was originally based on their work on openness indicators, we understand that the link with our work is too weak to justify the reference.
We are not aware of any recent paper using a similar approach for trade intensity indices.

 

3. Despite I recognize the validity of using your RTP measure to overcome the issue of estimating multilateral trade resistance (MTR), many other alternative techniques have been proposed in recent literature about estimating MTR in the gravity of trade. So, I recommend author(s) to include a more recent literature on MTR (now, you just have Yotov et al. 2016) and to compare the results from your model with those from using the following gravity models (In terms of model fitting and in terms of the estimated beta coefficients associated to infrastructure variables)
4. a) Baier and Bergstrand (2009) first-order Taylor series approximations of the MTR terms. (Baier SL, Bergstrand JH (2009) Bonus vetus OLS: a simple method for approximating international trade-cost effects using the gravity equation. J Int Econ 77(1):77–85)
5. b) Fixed effects model used by Feenstra RC (2004) Advanced international trade: theory and evidence. Princeton University Press, Princeton
6. c) Eigenvector Spatial Filters proposed by Patuelli, R., Linders, G. J. M., Metulini, R., & Griffith, D. A. (2016). The space of gravity: Spatially filtered estimation of a gravity model for bilateral trade. In Spatial econometric interaction modelling (pp. 145-169). Springer, Cham and by Metulini, R., Patuelli, R., & Griffith, D. A. (2018). A spatial-filtering zero-inflated approach to the estimation of the gravity model of trade. Econometrics, 6(1), 9.

 

Response 3: We have mentioned the above papers in our short survey of alternative methods to control for MTR in gravity equation. However, a direct comparison between the results obtained in these articles and ours would be very difficult and perhaps misleading, given the differences in the research questions they address, and the nature of the dependent variable used in our paper.

We have also added a reference to Novy D. (2013) Gravity Redux: Measuring International Trade Costs with Panel Data. Economic Inquiry, 51(1), 101-121, whose approach to the measurement of trade costs bears some degree of similarity to our motivations for using RTP indices.

 

 

4. By using a two-step regression strategy and without including mass variable, you have obtained a small R^2 (less than 0.5) while, traditionally, gravity equation explains about 0.8 of the total variance in the dependent variable. Moreover, the second step model reports a very small R^2 (0.017). This is another reason to prefer (my opinion) a 1-step regression with all variables together (mass variables as well).

 

Response 4: In our view, the main reason for which the R^2 of traditional gravity equation is so high is precisely the use of GDP or population among the regressors. By doing so, the gravity equation captures the rather trivial fact that large countries tend to record large trade values only because of their size. On the contrary, we focus our attention on the role of bilateral vs. multilateral trade costs and claim that our approach may give a precise measure of their effect on RTPs, regardless of country size.

We have already argued above why we prefer a two-step approach. However, we have tried to improve the specification of our first equation by replacing imports with total trade (exports plus imports) in the computation of RTP indices, which is more consistent with the choice of working only with dyadic trade cost variables. Also, the second equation now includes only country-specific fixed effects, and its R^2 has significantly increased.

 

Minor comments:

1. Pag. 3, lines 97-98: It is better to say that mass and distance are used in the general gravity model. Economic size and trade cost when talking about the gravity of trade

 

Response 1. We have replaced “mass” with “size” wherever appropriate. We have made extensive references to trade costs, mentioning distance only as one of the variables affecting trade costs.

 

 

2. Pag. 4, lines 165: please, define M

 

Response 2. In this revised version of the article, we have replaced imports (M) with total trade (T) defined as exports plus imports (see p. 5, line 198).

 

 

3. Section 4: having 3763 valid observations means that your dataset is unbalance (i.e. is not said that you both have country pair i --> j and country pair j -->i). How do you manage with this issue?

 

Response 3. Since we now use revealed trade preference indices (RTP_ij) based on total trade (exports plus imports) with 1,816 valid observations, we do not face this issue anymore, given that, by definition, RTP_ij = RTP_ji.

 

 

4. Section 4: If the average RTP stands to -0.349, it means that, to compute it, you consider the full networks of world countries, not just the 71 used in dataset of country pairs. Am I right? Please, specify it.

 

Response 4. Even if our RTP formula is influenced by the entire network of world trade, our 1,816 valid observations refer only to the 71 BRI countries. The fact that their average RTP index is negative (-0.298) shows that their mutual trade is less intense than implied by the geographic neutrality threshold.

 

 

5. Equation 4: Please correct. It is beta_1*Contig_ij

 

Response 5. Corrected.

 

 

6. Page 9 (line 297). I suspect that you have two eta parameters and that part of formula is: phi_1*eta_i + phi_2 *eta_j (one for the country of origin, the other for the country of destination). Moreover, I think that eta parameters capture the effect of GDP and population, so, why do you not include GDP and population as covariates in the regression?

 

Response 6. In this revised version of the article, we have introduced the set of country dummies only in the second equation; however, we still consider a single set of country dummies. As specified in note 5, “since we look at bilateral revealed trade preferences, we do not distinguish between importers and exporters because both countries in each pair have both roles. Therefore, we resort to a single set of country dummies to control for specific country effects other than infrastructure endowment”.

Including GDP and population as covariates in the regression would be inconsistent with our approach, as explained in the text (see section 2, lines 143-155), as country size (as measured by the total value of trade with the rest of the world) is already included in our RTP formula.

Reviewer 2 Report

The paper uses cross-section data for 2016 and evaluates the trade intensity against the Belt and Road initiative. The paper is extremely well written and structured. The methods and models are justified well. The conclusions are in line with the models used.

Probably there is the issue that the paper's conclusions are based on one year only. Some justification is in order here. 

 

 

 

 

Author Response

Thanks for your kind comments.
The novelty of our approach, which is based on the use of trade intensity indices in the gravity model, appears in our opinion clearly in a cross-section analysis limited to one year only, which allows to estimate to what extent the quality of existing infrastructure explains the untapped bilateral trade potential of countries participating in the Belt and Road Initiative (BRI).
The choice of 2016 (three years after the launch of the BRI) has been justified in the revised version of our paper, as this is the year that minimizes the number of missing observations.
A future extension of our work could be a full-fledged dynamic panel analysis, but this will only be possible when the actual effects of the BRI will materialize over a sufficiently long time span.

Round 2

Reviewer 1 Report

I'm satisfied with authors' responses to my comments and I think the new version of the paper worth to be published