Capaldo Fails to Convince

In October 2014, Jeronim Capaldo published a paper (hereafter Capaldo 2014) suggesting TTIP to cause severe negative economic implications for Europe. In a recent paper, we point to the fundamental methodological flaws underlying his attempt to project the economic impact of TTIP. In a response to our paper (hereafter Capaldo 2015), Capaldo defends both the methodology and the results of his original paper. Capaldo (2015) still fails to convince and by this response we seek to clarify what the discussion concerns.

---

1. What is our concern?

Most proposed trade agreements prompt debates about the potential costs and benefits they may incur. TTIP is no exception. There are more than a dozen attempts at quantifying the potential economic costs and benefits of TTIP. They are all associated with the known problems associated with ex ante economic analyses of trade – and they should all be taken for what they are: estimates of the direction of results from a trade agreement. What these numbers can do, at best, is to give a ballpark figure of potential gains or losses.

Most existing estimates of TTIP suggest that its benefits for the European Union (EU) and the United States (US) will outweigh the costs by a solid margin. There is one exception: Capaldo (2014). The author suggests TTIP will lead to net economic losses for the EU in terms of general economic activity and falling personal incomes. His results err on the extreme side of the results from trade agreements – and given the number of translations into several European languages and its wide use in the TTIP debate, it prompted us to figure out how the analysis could lead to results that are vastly different from other studies and, moreover, contrary to the any trade agreement concluded between either developed or developing economies.

Our conclusion is that the model used in Capaldo (2014) is not fit to estimate the effects of a trade agreement on the wider economy.

First, the model used by Capaldo cannot estimate the effects on trade from trade reforms. It cannot predict the scope, profile or the value-chain effects of trade associated with a trade reform. It can only derive macroeconomic effects from changes in net exports. Capaldo’s model does not distinguish between sectors that expand or contract as a consequence of a trade agreement. The model can only deal with trade at a highly aggregated level. It follows that it cannot take account of trade in intermediate products and supply chains, and how these trade flows affect national output. It can only follow a very old version of trade where one product is produced in one country and then consumed in another country. Today, however, one dollar that is lost in exports does not correspond to one dollar that is lost in output.

Second, the model cannot take account of the known benefits from trade, which are linked to improvements in the efficiency of production and the structural adaptation of the economy. The model is a demand-driven macro model: its core and determining factor is effective demand. This does not mean it does not harbour supply-side factors. As we argued in our paper, such a model may have some merits when estimating broader short-term macroeconomic developments in a slack economy. But the model is not designed to take account of the effects of trade on factors like specialisation, productivity, investments or other supply-side effects from trade reforms. These are the effects that most trade economists hold to be the most important effects of trade liberalisation. Therefore, the model’s usefulness for the analysis of trade agreements is poor.

Third, while Capaldo’s methodology is only poorly traceable and highly obscure even to technical experts, the assumptions added to the model by Capaldo are likely to lead to disproportionate consequences when supply-side measures, like trade liberalisation, are introduced since these measures do not affect effective demand in the components of the model. Consider this example. The CEPR (2013) – using a model designed to take account of the effects on trade and output from trade reforms – estimates that the an ambitious TTIP will in the long run increase Europe’s total exports by 5.9 per cent and Europe’s total imports by 5.1 per cent. Capaldo (2014) predicts a net export loss for the EU as a whole. While Capaldo (2014) does not show the result for the EU, or explain the factors in the model that determine the export loss, there are estimates for selected EU countries. For instance, the United Kingdom (UK) will find its net exports decline by 0.95 per cent of GDP.

While the difference between the CEPR (2013) and Capaldo (2014) as far as net exports are concerned are significant, the really surprising part in Capaldo (2014) is how a loss in net exports of less than 1 per cent, leading to a net fall in GDP by 0.07 per cent, amounts to a tremendous loss in labour income of 4,245 euros for the average UK worker. At this point, the fundamental question is: how on earth canthis happen? We argue – but cannot know for sure because we are denied access to both the methodology and the model – that such a disproportionate result can be explained by the assumptions of the modeller and the properties of the model rather than the effects of TTIP on the British economy. Such extreme results demand an explanation – but we cannot find it in Capaldo (2014), Capaldo (2015) or in the technical explanation of the model provided by UNCTAD.

2. Responses to Capaldo (2015)

We do not have many comments to make on Capaldo (2015). None of the concerns we raised in our paper is being properly addressed and explained in credible way. We consider three comments to be useful for the debate about the inappropriateness of Capaldo’s calculations.

• Make the model accessible to the public

Capaldo (2014) is using a model that is not used for estimating the effects of trade agreements. We have scanned the literature for studies using the GPM to estimate trade agreements, but did not found a single attempt. Therefore, there should be a solid defence for why this model – and not standard trade models – does better at predicting the effects of trade on the economy. Neither Capaldo (2014) nor Capaldo (2015) provide that defence. We argued in our paper that if you look at the history of other trade agreements of similar size and depth, Capaldo’s results do not at all support the arguments used to give legitimacy to the GPM as an appropriate model to estimate the effects of trade agreements.

We have been denied access to the model used by Capaldo (2014). The model is owned and run by UNCTAD. Capaldo (2015) argues that the technical explanation given by UNCTAD (Cripps and Izurieta, 2014) publicly provides enough detail to either replicate the model or to understand in detail how it works and how different components of the model operate vis-à-vis each other when faced with new economic circumstances. It is almost impossible, however, for any researcher to conduct sensitivity analyses that are based on paperwork.

CGE models, like those offered by the Global trade Analysis Project (GTAP), are publicly available for policymakers, individual researchers and research groups. Dozens of researchers worldwide have tested these models for the sake of sensitivity analysis and helped to improve databases, model components, functional relationships and elasticities.

UNCTAD has given Capaldo exclusive access to the model. As a publicly financed institution, why does UNCTAD discriminate between different potential users of the model? Capaldo (2014) is supposed to be an academic analysis published by an academic institute. As such, its methodology should be open for standard peer review. That implies that other academics are provided the opportunity to replicate Capaldo’s research by making public the model, the assumptions, the methodology and the data.

• Make the methodology transparent and the data accessible to the public

Capaldo (2014) also creates a lot of ambiguity for any reader trying to understand what he has done in order to prepare and process the data in the model. For instance, Capaldo (2014) states:

“In order to implement the TTIP scenario, we assume that the volume of trade among TTIP countries will initially expand at the pace indicated by the existing studies. However, we do not rely on these studies for changes in net exports, which ultimately determine any changes in GDP. Instead, we calculate net exports changes taking into account the global feedbacks built into the GPM. Therefore, our simulation clarifies the implications of the “consensus” pattern of trade in terms of GDP, income distribution and non-TTIP trade. In the GPM, the impact of a given increase in trade is different from other models. As indicated above, such change affects the distribution of income ultimately feeding back into total demand and income.” Capaldo (2015, p. 12)

A footnote is accompanying the first sentence in that quote saying:

“The GPM does not include data on tariffs, so we cannot calculate the tariff equivalent of a reduction in trade costs and its impact on exports. Thus we take the approach of checking the implications of the changes in trade that have been estimated by previous studies. We express these increases in terms of each country’s share in the import market of the others rather than in terms of export and import levels.” Capaldo (2014, p. 12)

We already explained in our previous paper that Capaldo’s methodological exercises are quite adventurous and will have consequences for the way the model will work. This is particularly true for the re-calculation of trade into trade shares on ‘each others’ market, as obscurely pointed out by Capaldo. The point is that this presentation of what has happened to the data is so poor that no scholar would have a chance to replicate Capaldo’s exercise, provided they could get access to the model.

• Explain the impact of labour income shares on economic output

What is it in TTIP that causes extreme income losses despite, as in the case of UK, an insignificant effect on GDP and marginal employment changes? In fact, what is obscure for trade shares is also true for labour income shares. The way Capaldo looks at labour income shares lies at the heart of the extremeness of his results. Capaldo models changes in labour income shares in a way that makes his final results out of proportion. As outlined above, Capaldo proclaims that the loss in the UK’s net exports of less than 1 per cent causes the UK’s GDP to fall by 0.07 per cent, which amounts to a tremendous loss in labour income of 4,245 euros for the average UK worker.

Capaldo does not at all explain assumed causalities or any historical relationships between trade and labour income shares. Not does he give explanation of the impact of changes in labour income shares on economic activity. According to his own understanding, losses in national employment cause labour incomes to fall in the economies he is looking at:

“[t]he loss of employment would further accelerate the reduction of incomes that has contributed to the EU’s current stagnation. Indeed labor income will continue its steady decrease as a share of total income, weakening consumption and residential investment while likely exacerbating social tensions.” Capaldo (2014, p. 15, bold added)

Capaldo simulates a strong shift in the composition of national income in terms of rising capital income and falling labour income. For clarification, in the Global Policy Model, labour income includes the national accounting categories ‘compensation of employees’ and ‘mixed income’. Profits are represented by the ‘operating surplus’. Employment includes employees, self-employed and family workers (UNCTAD 2014).

Capaldo’s results explode in size due to shifts in income compositions that are, however, not explained any further. In addition, Capaldo’s projections are in conflict with country-specific real world developments over the past 20 years (for which we checked the data, see Figure 1).

Between 1995 and 2011 (most up-to-date data for adjusted labour income shares provided by the International Labour Organisation, which is Capaldo’s benchmark), the share in labour-income in France actually increased from 58.1 per cent to 58.8 per cent. At the same time, the share in labour-income in the United Kingdom increased from 62.1 per cent to 63.6 per cent. For Germany, it indeed declined by 4 percentage points, while it slightly decreased by 1 percentage point for Italy. The numbers clearly show that the trend Capaldo highlights in his paper does not correspond with the record for the countries he focuses on. Notably, this trend does not hold for the past two decades – a time in which Europe faced strong import pressure from low wage countries including China and other emerging market economies. Notably, absolute labour income kept rising over the same period.

Figure 1: Development of labour income shares

Source: International Labour Organisation (ILO), labour income share in Gross Domestic Product (GDP) – adjusted, most up-to-data data (2011); Capaldo (2014) forecast.

According to the reasoning of Capaldo,

“[t]he flipside of this decrease is an increase in the share of profits and rents in total income, indicating that proportionally there would be a transfer of income from labor to capital. The largest reductions will take place in UK (with 7 per cent of GDP transferred from labor to profit income), France (8 per cent), Germany and Northern Europe (4 per cent).” Capaldo (2014, p. 15, bold added)

The chain of causality, as briefly outlined by Capaldo, appears to be based on logical grounds: 1) less employment causes 2) less economic activity, which causes 3) incomes to fall. However, the shifts in labour income shares are fundamentally disproportionate to what Capaldo types out for losses resulting from changes in employment.

The following calculations shall illustrate how disproportionate Capaldo’s calculations are and how the paper hides from readers the technical causalities that define the results.

We have tried to uncover the link between Capaldo’s estimated changes in national employment and proclaimed changes in national labour income shares for France, Germany, Italy, and the UK (the countries for which numbers are reported in Capaldo’s original paper). The calculations are presented in Table 1.

For sensitivity considerations, we use two indicators of national employment as a starting point: (I) total national employment based on the employment-to-population ratio as provided by the World Bank and (II) total national employment figures as provided by Eurostat. Based on Capaldo’s estimated employment changes, aggregated labour-income would fall by 3.25 billion USD in France, 3.57 billion USD in Germany, 59 million USD in Italy and 80 million USD in the United Kingdom. At maximum, this would correspond to an annual income loss of 126 USD per worker in France, 90 USD per worker in Germany, 2.60 USD per worker in Italy and 2.60 USD per worker in the United Kingdom.

Following the way Capaldo presents his results, he overestimates the impact of his own numbers by a very large factor.

When we take into account Capaldo’s estimated changes in the shares of labour income as part of total national income, we move very close to Capaldo’s final results. This constitutes the biggest flaw in Capaldo’s analysis: Capaldo links TTIP to falling labour income shares in EU member states, but he does not at all explain the precise relationship between the impact of TTIP on trade, industrial economic activity and falling shares in labour income, let alone that he deliberately ignores the ILO’s numbers on past trends in labour income shares for the countries he is looking at.

Capaldo argues that

[t]he most important difference between the GPM and the CGE models […] is that, in the GPM, the full-employment assumption is replaced by the Keynesian principle of “effective demand” […]. This means that the level of economic activity is driven by aggregate demand rather than productive efficiency. Consequently, a cost-cutting trade reform may have adverse effects on the economy if the “costs” that it “cuts” are the labor incomes that support aggregate demand. Unlike in CGE models, changes in income distribution contribute to determining the level of economic activity. Capaldo (2014, p. 10, bold added)

As we already outlined in our first paper, Capaldo’s results contradict with the UN’s communiqués on the benefits that would results from what the UN calls ‘global rebalancing’. In addition to that, Capaldo’s reasoning also conflicts with UNCTAD’s observations for the relationship between labour income shares and GDP.

Capaldo refers to UNCTAD’s 2014 report on global trade to be his primary data source for setting up his baseline scenarios. According to this report,

“[…] declining labour incomes affect revenues of households that have a higher propensity to spend, further eroding, in the aggregate, consumption and investment demand. This eventually has an adverse impact on imports, and thus on the exports of the whole set of countries.” UNCTAD (2014a, p. 19)

Following the understanding of UNCTAD (2014a), falling labour income shares would cause less demand for imports, which would, everything else being equal, cause rising net exports. This would be exactly the contrary of what Capaldo claims for the path of EU countries’ net exports.

In addition, UNCTAD itself clarifies that the relationship between falling labour income shares is ambiguous. Moreover, UNCTAD itself draws attention to the fact that falling labour income shares do not cause GDP to fall:

“[…] the most recent evolution of real unit labour costs and GDP growth in Europe suggests some ambiguity about the relationship between these variables. Despite a seeming pause in wage compression in Europe, though no real growth in wages, GDP seems to be gaining traction. Likewise, there were periods in the past when GDP growth in developed countries remained relatively strong even though labour income shares in GDP were falling, or only marginally rising.” UNCTAD (2014a, p. 19, bold added)

Table 1: Sanity check of Capaldo’s income loss calculations

• What exactly is modelled by the GPM?

The basic test of a model used to estimate the effects of a trade agreement is whether it can capture the effects of changes in tariffs and non-tariff barriers on trade. The model used by Capaldo (2014) cannot, and Capaldo (2015) argues: “While it is true that tariffs and non-tariff measures are not explicitly measured in the UN model, it is not clear why they should be.”

One would think the answer should be evident: if the model cannot predict the effects on trade from a trade agreement, there is a flaw in the basic model closure that should be analysed. The more the model relies on macroeconomic aggregates and the further away from the actual effects on trade the model operates, the bigger are the risks that it will describe other effects than those emerging from the changes on the scale and composition of trade as a result of a trade agreement.

If Capaldo wants to improve the understanding of how TTIP in his view will affect trade, he should at least provide the model results for how trade will expand for the US and selected EU countries, or groups thereof, and allow the reader to compare them with the results from other models. To our understanding, in the model applied by Capaldo (2014) trade liberalisation always is a zero-sum game. Rising net exports in one region (here the US) must, by definition, correspond to falling net exports in other regions (here the EU). In the model, changes in net exports are directly linked to changes in economic output. Therefore output gains in one region automatically correspond to output losses in other regions. The model cannot account for rising levels of bilateral commerce and economic activity occurring in all regions signing a trade agreement. Capaldo’s model, by definition, arrives at winners and losers.

To our understanding, Capaldo’s work is marked by deliberate avoidance of the economic benefits that arise from trade liberalisation. Capaldo (2015) rather aims to move away the focus of the discussion towards income distribution and inequality. That is a good discussion, but it needs to be separated from a discussion about models that might be able to estimate distributional variables, but not the effects of trade from trade agreements. And it is separate from Capaldo’s own work on TTIP. Capaldo still cannot show that it is TTIP rather than model properties (or assumptions put into the model) that explain the negative effects on employment and income that he claims TTIP will entail.

Based on above considerations, Capaldo should allow for peer review and, therefore, address the following issues in detail:

• Concerning the initial expansion of trade from existing studies, it would be necessary to know: Which studies? Which years? What is the data source?

• These studies are not used for calculating the changes in net exports: why not? What data has been used to derive net exports?

• What is the role of falling labour income share? Why do labour income shares drive the model’s results? How does TTIP affect labour income shares? Why does TTIP not affect labour income shares in the US?

• What is the relationship between trade and labour income shares, and labour income shares and GDP?

• Compared to the US, the assumptions on investment and government spending are adverse for the EU. Why?

• What data has been used to calculate import shares? What is the level of import shares? How do they find their way to the model? Why is that data transformation more appropriate then data from existing studies?

• The global feedbacks in the GPM used to calculate net exports: What is meant by feedbacks and how exactly do they operate? If the analysis is based on a selective use of bilateral trade expansion from existing studies (and not the total expansion of trade from existing studies), how do these two sources of data match?

• Since trade data are taken from different simulations that are based on different behavioural assumptions of the economy, how are these data made compatible?

And so the list could continue.

The key point is: release the model and the datasets used so interested scholars can test the robustness of the results!

---

References

Capaldo, J., 2015, Overcooked Free-Trade Dogmas in the Debate on TTIP, Global Development and Environment Institute, Tufts University, Medford MA

Capaldo, J., 2014, The Trans-Atlantic Trade and Investment Partnership: European Disintegration, Unemployment and Instability, Global Development and Environment Institute Working Paper NO. 14-03.

UNCTAD, 2014, The UN Global Policy Model (GPM): Technical Description, authored by Francis Cripps and Alex Izurieta.

UNCTAD, 2014a, Trade and Development Report 2014.