In GTAP, there is a technology change variable. It operates directly on the quantity of trade – either enhancing it (trade facilitation) or eroding it (iceberg melt) (this is Professor Thomas Hertel's words when we exchanged some thoughts on this in emails); on the other hand, there is a variable representing cost index for international transportation which reflects the cost of the input into the transport activity itself.  This index is indirectly impacted by a transportation margin calculated as the CIF price over the custom import value (a. Custom import value (CIV) is defined as the price actually paid for the merchandise excluding import duties, freight, insurance, and other charges incurred in the importing of merchandise to its destination; b. This is because, in GTAP, higher transportation margin leads to larger share of margin in total costs of getting goods from source to origin.). It operates similarly as the Samuelsonian iceberg equaling CIF over FOB prices.

The GTAP technical document explains that there seems to be a relationship of this margin with the role of geographic distance by looking at its value (lower likely means geographically closer), and for example, some pairs of bilateral regions such as the US and Canada and US and others, but this leaves a mystery for other pairs such as with New Zealand and Japan that are not explained alone by distance. It shows that there is a strong evidence of the relationship between this margin and importance of distance for some goods, such as live animals.

The following is the import demand equation in GTAP derived in percent change form:

    qxs(i,r,s)
        = -ams(i,r,s) + qim(i,s)
        - ESUBM(i) * [pms(i,r,s) - ams(i,r,s) - pim(i,s)];

• qxs is export sales of commodity i from r to region s
• ams is the iceberg cost (or technology change) variable
• qim is aggregate imports of i in region s
• ESUBM is region-generic elasticity of substitution among imports of i in Armington structure
• pms is domestic price for good i supplied from r to region s
• pim is market price of composite import i in region r

​(In GTAP syntax, lowercase letters mean percent change of a variable, uppercase letters represent parameters or data both read from files, green color means endogenous variable, and red means exogenous variable that is “shockable”).

pms(i,r,s) = tm(i,s) + tms(i,r,s) + pcif(i,r,s);

• Both “tm”s are tariff shocks, one is source-generic, the other is by source-destination
• pcif is CIF price

The following equation links export and import prices for commodities:

• FOBSHR is the FOB share (goods price at FOB) of total import values
• TRNSHR is the transport share (transportation costs of the goods) of total import values
• pfob is FOB price
• ptrans is the transport margin or cost index for transport of i from r to s

In my view, the iceberg can be  melt twice  based on the equations above: one is counterfactually operated through “ams” which is an iceberg shock (directly on quantity) that is equivalent to either a de facto positive or negative productivity shock (i.e., importers ask less or more than what they receive); the other is operated through “ptrans” indirectly impacted by the data that GTAP collected from COMTRADE/Census which seems to have already explained some of the geographic variations (both natural and artificial). For example, a longer distance between bilateral pairs would suggest higher value of transportation margin and ptrans therefore higher pcif and pms which impedes bilateral trade. See the same import demand equation below but with - ESUBM(i) * [pms(i,r,s) highlighted.

    qxs(i,r,s)
        = -ams(i,r,s) + qim(i,s)
        - ESUBM(i) * [pms(i,r,s) - ams(i,r,s) - pim(i,s)];

​Again, ESUBM are the region-generic elasticities of substitution among imports of goods i, which are positive. Note that this equation has some gravity sense (in the sense that the change of bilateral trade flows on the LHS is explained by some variables and variables on the RHS). The role of distance and other variations reflected by geographic locations are absorbed by pms through the transportation margin, but the role of elasticity of trade costs with respect to distance and other factors such as border frictions (as in NQT literature) is disappeared.

Which one should we see more as iceberg in GTAP from the perspective of a GTAPper? Is that REALLY ams (alternatively seen as a productivity shock; see Walmsley and Minor (2020)) or ptrans (the transport margin)? Ptrans is a variable representing cost index for international transportation which reflects the cost of the input into the transport activity. This index is indirectly impacted by a transportation margin calculated as the CIF price over the custom import value and seems that both natural and artificial components are in it so that the role of distance is left unexplainable. It seems that if we can let the distance explain part of the geographic variations while taking the model to the same data to restate the logic of elasticity of trade costs, we would probably end up bridging the gap with geographic quant literature to some extent.  Alternatively, calibrating rho’s in EK literatures to the aggregate GTAP data and model specification might lead to something similar. This is because that in NQT people do not shock the entire iceberg but only part of it; meanwhile other structural parameters are endogenously impacted by the distance. In the big picture, the intuitions of the two literatures in these frictions are not far-off, but the interpretations, methods/approaches, and applications (especially in the sense of counterfactuals) are different.

Why do we need to calibrate? Before answering this question, perhaps we want to first revisit the fundamental points of estimation and calibration. Last year I presented a guest lecture in Dr. Hertel’s AgEc 618, so many thoughts I wrote in the slides, while have been evolved, seem to be still valid. In the structural (econometric) estimation, what we essentially do is to extract the average information in the data or observables, then the key point of subsequent calibration is to replicate the data with the exact fit in the model.

The next question is how to produce the exact fit. Taking a simpler CES case in a general equilibrium model as an example, in the model there is one sigma substitutability parameter, one border cost parameter and one distance elasticity of trade costs that stayed constant across the estimation and subsequent calibration. In this case, the difference in the implementation between econometric estimation and the calibrated exact fit is that the distribution parameters are often permitted to vary bilaterally in the calibration (see, e.g., Balistreri and Hillberry, Economic Inquiry, 2008), whereas in the econometric estimation each origin has a unique solution of distribution parameter presented in the utility function of each buyer or destination’s representative agent. Still with the CES, but in a more complex case on the Melitz model, Balistreri, Hillberry and Rutherford (JIE, 2011) allow bilateral variation in fixed costs to generate the exact fit.

One issue prior to the CDE parameters ever had been directly estimated was that income elasticities of certain goods in developing countries do not change considerably when incomes grow, whereas empirically it must be true that income elasticities will generally decrease as incomes grow, up to different orders of magnitude across goods. The issue thus spurred partial motivation for calibration, but this doesn’t seem to be a problem based on the results in my CDE estimates. Nevertheless, the point of calibrating the CDE is still to produce the exact fit that closely reflects the dataset, therefore broadly it does seem that subsequent calibration is necessary.

The CDE-in-GTAP case is more complicated. In this case, there is one distribution parameter, one expansion parameter and one partial substitution parameter, along with many other parameters in the GTAP model with CDE private household demand. The important question to ask is, which parameters should we believe to be common to all countries, and which parameters do we want to free up to produce an exact fit? Without thinking twice, I would immediately think that, just like the CES, the distribution parameters should be freed up, and that the expansion and substitution parameters are common to all countries. Similar to my responses in a group discussion with Dr. Hertel, this is not because we don’t believe that in the real world each country should not have her own consumer preferences, it is simply because that from the given piece of data information and the ad hoc specification the CDE model does not reveal regionally differentiated information at all, but rather predict different demand orbits and shares of consumption at each level of unit-cost prices. Put it simply, the econometric exercise does not exploit the variations of other variables (i.e., other than income and/or prices) that make the consumer preferences differently illustrated.

However, this choice becomes more complicated when I relate the normalization strategy of distribution parameters to other parameters in the econometric estimation.  In our CDE estimation paper, the change of value of either distribution or expansion parameters would affect the other one as well as the seemingly “floating” utility levels.  Through the design of normalization procedure, we find that our estimation would be robust if we let 1) the sum of the distribution parameters to equal the joint value of the right-hand side of additivity equation (in our case we set this value to equal 1 for the purpose of numerical stability), and 2) the sum of the expansion parameters to equal the number of goods; this means that the expansion parameters as exponents on utility tend to equal one on average across the goods. We also find that these operations tend to subtly improve the scaling of the utility level. So technically, the parameters that should be freed up might be both the distribution and expansion parameters in a way satisfying the same optimization constraints in the estimation.
​
The issue is that, in general, we don’t want the parameters we use to generate an exact fit to be used also in any counterfactual analysis. However, the expansion parameters do enter the counterfactuals. Maybe one way to think about this problem is that for each distribution parameter present in the utility of the destination’s market there is a one-to-one mapping of one expansion parameter wherein the solution to the partial substitution function is unique at each income-price pair. For this reason, we shall seek for appropriate values of distribution parameters that produce the exact fit corresponding to the data, while satisfying both the adding-up constraints for distribution and expansion parameters. 

These days I have been in frequent conversations with people who have good insights into schools of levels and linearization in contemporary CGE models. Some thoughts are derived from our discussions. These are things at least I would think that have created somewhat communication barriers between schools of CGEMs and NQTMs. One thing that initially draws my attention was due to an argument that linearization is indifferent to implicit or explicit preferences ’slash' technologies. But perhaps before we even think about theories and computational strategies we may want to ask how less good were computers if we all step into a time machine and go back to 20-30 years ago. At that time, an empirical solution in levels might have been much more challenging than it is now today. A mathematical algorithm that crawls toward the solution with a series of linear approximations was probably much more robust. With advances and innovations in the technology of computer advancement, however, all have become ancient history. This probably means that for a less extensive collection of CGE models there is no need to do approximations and the exact hat calculus would be the suitable candidate to calculate counterfactuals. This then makes sense that majority of the school of NQTM modelers do not use the GEMPACK software which typically comes up with solution to errors associated with linearization to iterate the linearization over and over until a new solution in levels is achieved. Pioneers of CGEs have developed this strategy to solve large general equilibrium trade models (e.g., huge contributions to CGE due to Johansen, Dixon and Hertel). If the schools of CGEs and NQTMs ever found a way to communicate, then the CGE modelers would likely explain that they actually obtained the levels solution with the so-called iterative linearization procedure, then undoubtedly NQTM modelers would question why CGE modelers do it that way. The fact is probably that the linearization representation with GEMPACK indeed has their own advantages in solving large CGE models (e.g., see IMPACT working paper), while some younger generations of CGE economists have been locked into this tradition, but similar to what the IMPACT working paper (revised version published in Economic Modeling) argues, with contemporary advances in computers and softwares we might tend to have less need to do approximations (e.g., Johansen matrix inversion) if levels solution is becoming inexpensive and more reliable.  ​