Cato Online Forum

Understanding the Economic Models and the Projections They Produce

By Gabriel J. Felbermayr
October 2015

Quantitative Trade Models: Voodoo Economics or Important Policy Tools?

The quantitative modelling of the largest and deepest bilateral trade agreement ever considered is a gargantuan task. Not only does the sheer size of the involved economies and their intensive trade with other countries mean that there will be systemic implications, but a Transatlantic Trade and Investment Partnership agreement is also expected to break new ground in several areas, such as regulatory coherence. Moreover, at the present stage one can only speculate about the nature of the final outcome. Any attempt at ex ante quantification of the economic effects of the deal must therefore include a careful modelling, not only of the transatlantic economies, but also of the rest of the world. It must develop a perspective on how to deal with non-tariff barriers to trade. And it has to reflect realistic expectations about what is achievable.

It is no wonder that modelling attempts by different teams have produced divergent results. The table below shows that early studies differed substantially in their quantitative predictions. The first quantitative studies were presented by the Ifo Institute (commissioned work by the German Federal Ministry for Economic Affairs and for the Bertelsmann Foundation) and CEPR (commissioned work for the EU Commission).

More recently, there has been some convergence. The CEPR study was updated by Egger et al. (2015), using a different empirical strategy but the same model. Similarly, Felbermayr et al. (2015) updated the Ifo study. These papers are published in Economic Policy, a peer-reviewed scientific journal. Aichele et al. (2014) extends and applies to TTIP a framework introduced by Caliendo and Parro (2015) for the ex post analysis of NAFTA.

The wide range of different results has left some observers with the impression that the existing studies are purely speculative. In this context, the German vice-chancellor Sigmar Gabriel has used the term Voodoo economics. Yet, quantitative assessments are crucial. They help policymakers decide which amongst the many possible bilateral trade agreements to pursue and how much political capital to invest. And they help net out the effects of countervailing forces, such as trade diversion and trade creation.

Table 1 Simulated effects of TTIP (deep integration) in various studies

CEPR (2013)*

Ifo (2013)

Felbermayr et al. (2015)

Egger et al. (2015)

Aichele et al. (2014)

Spillovers:

YES

NO

NO

YES

NO

YES

NO

YES

USA

0.48

13.4

4.89

7.05

0.97

1.13

2.68

3.37

EU

0.39

n.a.

3.94

7.76

2.27

2.97

2.12

2.65

Germany

n.a.

4.7

3.48

7.12

1.43

2.32

2.6

3.1

France

n.a.

2.6

3.46

7.21

1.33

1.88

2.2

2.6

UK

n.a.

9.7

5.14

9.01

1.88

2.22

2.3

2.8

Italy

n.a.

4.9

3.86

7.66

1.46

2.23

1.2

1.7

Spain

n.a.

6.6

5.56

9.59

0.75

1.37

1.2

1.8

Third countries

0.14

n.a.

-0.92

0.8

n.a.

n.a.

-0.03

1.21

Turkey

0.08

-2.5

-1.56

0.14

-0.75

1.8

0.1

1.16

Japan

0.19

-5.9

-0.51

0.61

-0.19

0.09

-0.1

0.4

China

0.03

-0.4

-0.5

0.8

-0.27

0.26

-0.23

1.14

Indonesia

0.89

-0.2

-0.09

n.a.

n.a.

n.a.

-0.1

0.9

Low Income

0.2

n.a.

n.a.

n.a.

0.15

0.02

n.a.

n.a.

Structural

NO

YES

NO

NO

MIX

MIX

YES

YES

Base year

2022

2007

2012

2012

2011

2011

2007

2007

Aggregation

Micro

Macro

Macro

Macro

Micro

Micro

Macro

Macro

* Due to missing country-level detail, Turkey is proxied by Non-TTIP Mediterranean, Indonesia by ASEAN, and Japan by Other OECD, high income. Base year 2022 is a projection. Macro-level aggregation implies one-sector structure. “Structural” refers to models whose parameters are at least partly estimated on the basis of structural relationships generated by the model and using the exact baseline data.

Where Studies Agree

Putting quantitative details aside, there is a large degree of consensus across these seemingly different studies:

  • Virtually all of them agree that TTIP would increase real per capita income in the European Union and the United States; this is not a priori obvious due to the so called Viner ambiguity. When studies present country-level detail for the EU, they agree that all member states win. This, too, is not a trivial prediction due to within-EU trade diversion effects.
  • Studies also tend to agree that gains are not huge, but that they are larger than what one could obtain from other realistically pursuable trade agreements.
  • All studies concur that the bulk of gains derives from measures in the area of non-tariff barriers, i.e., the simplification of bureaucratic procedures, regulatory convergence, and the optimization of rules.
  • All of them conclude that some third countries would lose from the initiative, unless trade costs for outsiders go down as well, due to what is called spillover effects, e.g., by establishing world standards and thus lowering trade costs for outsiders, too. In this case, both outsiders and insiders would see gains from the agreement go up.

Where Studies Differ

Studies differ across a variety of dimensions. A better understanding of the various origins of these differences would undoubtedly improve the quality of discussion on the pro and cons of TTIP, as it sheds light into the relative importance of modelling assumptions and thus on the economic mechanics at work. Three of them are particularly important:

  • model structure,
  • measurement of trade costs and non-tariff trade barriers,
  • scenario definition.

The Role of Model Structure

Quantitative trade models have been used to assess the potential gains from trade policy reforms since enough computing power has become available to solve complex large-scale non-linear systems of equations. The literature proposes various computable general equilibrium (CGE) models which typically describe a multi-country multi-sector world economy under the assumptions of perfect competition and full employment. With improved computing power and better data availability, these models have grown larger and more complex.1 They have come under criticism for two main reasons: first, because of their black-box nature, and, second, because of their failure to conform to ex post assessments of trade agreements.

That models describing an increasingly complex world economy are not overly tractable is no wonder. And yet, the question arises, whether more complex and presumably more realistic models are better suited to provide insights to inform policy makers and the public. In modern economics, whether theories are good or bad depends not per se on whether all their assumptions are realistic, but on how useful the models are in answering the questions that they are designed to address, and whether their predictions conform to data. It is in this respect that standard CGE models have come into criticism. For example, Kehoe (2005) shows that such models have dramatically underestimated the trade flow effects of prominent trade agreements such as the North American Free Trade Agreement (NAFTA). Moreover, CGE models need information on a large set of parameters; these are often borrowed from various empirical studies that rely on samples and time periods different than the ones that define the baseline data of the CGE model. Moreover, the underlying econometric models may not be consistent with the theoretical framework that forms the CGE model.

This has led to the development of “new quantitative trade theory” (NQQT, Ottaviano, 2014); the seminal articles are Eaton and Kortum (2002) and Anderson and van Wincoop (2003); Costinot and Rodgriguez-Clare (2014) provide an overview. What these new approaches have in common is (i) a simpler, and thus more tractable, model structure which allows solving the model in discrete changes rather than in levels, (ii) the use of structural relationships generated by the model (such as the gravity equation) to econometrically identify the key parameters (such as trade elasticities), and (iii) scenario definitions for ex ante analysis that are based on the estimates of the treatment effects of comparable existing policies.

Multi-sector models have the advantage that they provide insights into the sectoral effects of policy changes. Moreover, they capture differences in the structure of comparative advantage of countries; this is important in assessing the strength of trade diversion effects. If one is interested in long-run aggregate effects, the advantages of multi-sector models are less obvious, as the structure of comparative advantage cannot be assumed to be fixed.2

In NQTT, multi-sector models are still the exception; (see for surveys Head and Mayer, 2014, and Costinot and Rodríguez-Clare, 2014). Extending the one-sector economy to many sectors is straight-forward theoretically, but additional assumptions are needed and difficult data issues arise. For example, one has to assume the elasticities of substitution between sectors. Most of the multiple sector gravity models fix the share of income spent on a given sector, thus ruling out structural change. Single-sector approaches are agnostic about these changes.

Similarly, multi-sector models must make assumptions on factor mobility between sectors. With the usual assumption of perfect mobility between sectors multi-sector models are comparable to one-sector models in terms of the employment of workers: they are always ideally allocated. Therefore, the single sector view corresponds to the long-run where all structural adjustments have taken place. Additionally, it is consistent with structural changes in the economy that may happen due to TTIP, both in TTIP-member and non-member third countries.

Another reason to include multiple sectors would be to take into account the linkages between upstream and downstream producers. However, input-output databases assume that production technology is constant, effectively ruling out adjustments of the slicing up of the global value chain due to trade liberalization, which is at the heart of international across-sector linkages.
Typically, single sector models result in larger trade diversion effects and, consequently, in more negative welfare effects for third countries. The reason is that they do not take into account that outsiders may produce very different goods than insiders to a trade agreement, which naturally limits trade diversion. However, the trade structure itself may change due to a trade agreement or due to ongoing technological diffusion. Thus, single-sector models may usefully represent the long-run where technology adjusts.

In contrast, market structure does not make much of a difference. Felbermayr et al. (2015) use a monopolist competition model, which is, in many respects, isomorphic to a model with perfect competition; see Costinot and Rodroguez-Clare (2014).

The Role of Trade Costs and Non-Tariff Barriers (NTBs)

Traditional CGE models feature two types of observable trade costs: tariffs (or tariff equivalents of quantitative restrictions) and transportation costs as contained in input-output statistics. Modern approaches include the costs of NTBs. Putting aside definitional issues in the context of NTBs, there are basically two approaches to measuring NTBs: A bottom-up approach, and a top-down approach. CEPR (2013) use the former strategy. It requires an enormous effort on data-collection and expertise to construct an NTB measure from surveys sent to firms and translated into tariff equivalents by researchers. Berden et al. (2009) summarize the evidence of bottom-up estimates on NTBs for the transatlantic trade relationship. The bottom-up approach requires accurate data for every single bilateral trade link covered in the model. With trade costs set by observational data, researchers have to calibrate expenditure shares to match the model to observed trade data.

In contrast, the top-down approach does not postulate that NTBs can be directly measured, but infers them by fitting bilateral trade costs (often in so called ‘iceberg’ form) to the model such that it replicates the baseline trade matrix. Trade costs calibrated this way are typically much larger than those that one can directly observe. This implies that trade policy can have a much bigger potential impact, as there are larger barriers to be removed.

The CEPR model differs from the others in Table 1 in that it allows for some NTBs to be not resource consuming but rent-creating (such as a quota would). The other studies assume NTBs consume resources. Reducing such wasteful barriers releases larger economic gains than rent-creating barriers, as there is a direct resource saving effect.

Scenario Definition

Probably the quantitatively most important difference across studies is how researchers set the counterfactual scenario. Typically, the idea is to do a simulation-based counterfactual experiment: what happens if the model economy, calibrated to the observed (or projected) baseline data, is modified such that trade costs across the Atlantic are reduced? All studies assume that tariffs are fully eliminated. However, there is considerable variation concerning NTBs. The CEPR study uses an expert-defined trade cost reduction schedule; the other studies in Table 1 use a data-defined scenario.

The data-based approach uses the estimated effect of existing regional trade agreements (RTAs) from an econometric analysis of a gravity model. The assumption is that TTIP lowers EU-US trade costs as much as existing trade agreements have between their member countries. This strategy has the advantage that it does not need to specify by how much NTBs would fall in the proposed agreement, but instead relies on past observed effects of similar RTAs. This approach is attractive because it is easy to implement – at least on the aggregate level, but it may lead to under- or over-estimation. On the one hand, the official ambition for TTIP is to go deeper than the average existing RTA, which will lead to an underestimation of the potential trade and welfare effects of TTIP when relying on the average effect of RTAs in the past. On the other hand, it is possible that the easy barriers to trade have long been removed across the Atlantic, which will then lead to an overestimation, because many past RTAs substantially lowered tariffs and NTBs. Egger et al. (2015) and Aichele et al. (2014) employ this strategy in a multi-sector setup; this requires the estimation of a large number of separate RTA coefficients, each for one sector. Studies differ in how to deal with the endogeneity of RTAs, what type of RTA to use to inform the exercise, and how to deal with parameter uncertainty. Finally, scenarios may differ regarding the assumption of spillovers, see above. As a rule of thumb, across model with similar setups (e.g., Egger et al. (2015) and Aichele et al. (2014)), differences in quantitative results are almost entirely driven by differences in the imposed trade cost reductions.

Finally, the choice of base year is important, too. From 2005 to 2015, the relative weight of the transatlantic economy in world GDP has fallen by almost 10 percentage points, and this trend is to continue. Therefore, the earlier the base year, the larger are the potential gains from TTIP as the relative importance of the economy affected by the agreement is higher. This is particularly visible if one compares Ifo (2013) to Felbermayr et al. (2015). The latter study differs from the former in only two respects: it uses 2012 as the baseline year instead of 2007; and it uses a somewhat larger country sample (173 instead of 126 countries). The same scenario generates substantially larger effects in the former than in the latter.

Conclusions

Studies differ in many respects; the discussion above highlighted the most important of them. Differences in the modelling strategy lead to differences in projected outcomes. While this is irritating for the policy makers, comparing models and assumptions allows a better understanding of underlying economic mechanisms. However, at the end the most important question is: how large are the expected trade cost reductions that TTIP can generate? Modelers expect with impatience a fully negotiated text to guide their quantitative analysis.

References

Aichele, R., Felbermayr, G., and Heiland, I. (2014), “Going Deep: The Trade and Welfare Effects of TTIP”, CESifo Working Paper No. 5150
Anderson, J. E. and van Wincoop, E. (2003), “Gravity with Gravitas: A Solution to the Border Puzzle”, American Economic Review, 93(1): 170-192.
Berden, K., Francois, J., Thelle, M., Wymenga, P., and Tamminen, S. (2009), “Non-Tariff Measures in EU-US Trade and Investment- An Economic Analysis”, Report OJ 2007/S 180-219493 for the European Commission: Directorate-General for Trade.
Caliendo, L. and Parro, F. (2015), “Estimates of Trade and Welfare Effects of NAFTA”, Review of Economic Studies 82(1): 1-44.
CEPR (2013), “Reducing Transatlantic Barriers to Trade and Investment: An Economic Assessment”, Study for the EU Commission.
Head, K. and Mayer, T. (2014), “Gravity Equations: Workhorse, Toolkit, and Cookbook”, Chapter 3 in Gopinath, Helpman and Rogoff, Elsevier (eds.) The Handbook of International Economics Vol. 4, 131-195.
Costinot, A. and Rodriguez-Clare, A. (2014), “Trade Theory with Numbers: Quantifying the Consequences of Globalization”, Chapter 4 in Gopinath, Helpman and Rogoff, Elsevier (eds.) The Handbook of International Economics, 197-261.
Eaton, J. and Kortum, S. (2005), “Technology, Geography, and Trade”, Econometrica 70(5): 1741-1779.
Egger, P., Francois, J., Machin, M., Nelson, D. (2015), “Non-tariff Trade Barriers, Integration, and the Transatlantic Economy”, Economic Policy 30(83): 539-584.
Felbermayr, G., Heid, B., Larch, L., und Yalcin, E. (2015), “Macroeconomic Potentials of Transatlantic Free Trade: A High Resolution Perspective for Europe and the World”, Economic Policy 30(83): 491-537.
Ifo (2013), “Dimensionen und Auswirkungen eines Freihandelsabkommens zwischen der EU und den USA, ifo Forschungsberichte 62.
Kehoe, T. J. (2005), “An Evaluation of the Performance of Applied General Equilibrium Models of the Impact of NAFTA”, in Timothy J. Kehoe, T.N. Srinivasan and John Whalley (eds.) Frontiers in Applied General Equilibrium Modelling: Essays in Honor of Herbert Scarf, Cambridge University Press, 2005, 341–77.
Ottaviano, G. I. P., 2014, “European Integration and the Gains from Trade”, CEP Discussion Paper No 1301.

Notes:

1 Still, while computing is no real bottleneck anymore, certain extensions, such as the estimation of confidence intervals, are still very time consuming and, therefore, rarely undertaken.
2 CEPR (2013), Aichele et al. (2014) and Egger et al. (2015) have used a multiple-sector model which is in the tradition of the older CGE literature. The Ifo (2013) study and Felbermayr et al. (2015) have instead used a much simpler single-sector model, but applied the methods used in NQTT.

The opinions expressed here are solely those of the author and do not necessarily reflect the views of the Cato Institute. This essay was prepared as part of a special Cato online forum on The Economics, Geopolitics, and Architecture of the Transatlantic Trade and Investment Partnership.

Gabriel J. Felbermayr is a professor of economics at the University of Munich and director of the Ifo Center for International Economics.