برآورد نشت کربن و بهره وری از تنظیمات مرزی در تعادل عمومی - آیا تجمع سکتوری اهمیت دارد؟

Estimates of the carbon leakage resulting from sub-global climate policies tend to be lower when using economy-wide general equilibrium models than what technology-specific and bottom-up models suggest. In order to test whether this difference is due to excessive sectoral aggregation, I exploit disaggregated data and estimate unobserved values to create a dataset with rich industrial sector detail. The bias caused by sectoral aggregation is estimated by calibrating a computable general equilibrium model to this dataset and comparing results with those generated from more aggregated data. A stylized unilateral carbon pricing policy is simulated. Results show that aggregated calibrations overestimate industrial output loss and underestimate the increase in the CO2 embodied in imports. The efficiency of border carbon adjustments at reducing leakage is also underestimated. However, I find that general equilibrium estimates of carbon prices and economy-wide leakage rates are mostly unaffected by the degree of industrial aggregation. Highlights ► The number of sectors in a computable general equilibrium model is expanded. ► The direction and magnitude of aggregation bias is estimated. ► Aggregated calibrations underestimate industrial trade response to carbon pricing. ► Leakage rates are mostly unaffected. ► Border carbon adjustments are perceived to be more effective at reducing leakage.

As some countries put a price on CO2 emissions, it is feared that production of carbon-intensive goods will be shifted to countries which have not introduced mitigation measures, inducing CO2 emission leakage. A range of solutions including emission-right rebating schemes or border carbon adjustments (BCAs) — which can include carbon-intensity based tariffs, export rebates, or both — is discussed at the policy-making and academic levels. The set of countries and regions which have implemented or plan to implement market-based emissions reduction policies is rapidly changing. At the time of writing, the most important emissions trading schemes (ETS) include the European Union Emissions Trading Scheme, the Regional Greenhouse Gas Initiative, the New Zealand Emissions Trading Scheme and California's Global Warming Solutions Act. Some non-annex 1 countries such as China and South Korea are also considering emission reductions. Some policy proposals in the European Union (EU) and the United States (US) make explicit mention of BCAs (Asselt and Brewer, 2010). For example, the now-defunct H.R. 2454 Bill passed in 2009 by the US's house of representatives included provisions for the introduction of carbon tariffs. EU policy also contains provisions to support the sectors deemed to be exposed to carbon leakage and California has included the economically similar policy of including imported electricity within its cap. A number of studies estimate the extent of carbon leakage, the magnitude of necessary border adjustments and their leakage-reducing potential. Many rely on multi-sectoral, multi-regional general equilibrium models which allow both qualitative and quantitative assessments of economy-wide changes induced by climate policy. These models tend to predict modest leakage rates in the 10% to 30% range. Similar models are also used to estimate the efficiency of Border Carbon Adjustments at reducing this emissions leakage, and find them to be relatively ineffective. However, currently available studies are characterized by a high degree of sectoral aggregation, which might be ignoring significant leakage rates in carbon-intensive sectors. Indeed, another strand of literature relying on sector-specific partial equilibrium models generally predicts higher rates. The approach taken in this paper is to increase sectoral detail within a computable general equilibrium model of the world economy. The objectives are to investigate how available data can be used to provide a more precise sectoral distribution of impacts, re-assess leakage rates and the effectiveness of BCAs and quantify the “aggregation bias” which may be caused by relying on overly aggregated calibrations. Increasing sectoral detail is relevant from a policy perspective as BCAs are discussed at a fairly fine degree of aggregation. In the US, the Environmental Protection Agency (US-EPA, 2009)1 has identified 44 sectors as presumptively eligible for allowance rebates, being particularly energy and trade intensive. In the EU, about 105 sectors have been singled out as “exposed to a significant risk of carbon leakage” (European Commission, 2009). These are identified, respectively, at the North American Industrial Classification System (NAICS) 6-digit level, and the NACE-4 classification system — levels of aggregation close to what this paper works with as it includes twenty‐three 6-digit and twelve 4-digit NAICS sectors. None of the general equilibrium studies in the literature use models which are capable of generating such a level of detail. Indeed, a majority of existing models are based on different aggregations of the same global trade and production dataset, Purdue University's Global Trade Analysis Project (GTAP), which has a coarse description of industrial sectors. In this paper, I create a series of GTAP datasets based on different aggregations of the 16 industrial sectors available in cgtap. In addition, I develop gtap-mecs, a micro-consistent dataset covering the whole world economy which uses GTAP as a starting point and expands its industrial coverage from 16 to 51 sectors. gtap-mecs exploits detailed industrial energy use data made available in the US by the Energy Information Administration's Manufacturing Energy Consumption Survey (MECS), as well as input–output data from the Bureau of Economic Analysis (BEA) and disaggregated international trade data. Because energy intensity and input/output data outside of the US is not observed, the calibration relies on identifying assumptions for their estimation. The uncertainty due to these assumptions generates variability in results which is carefully accounted for. Industrial sectors are often both energy intensive and heavily traded: they comprise 69% of total US imports in value and 80% of embodied CO2 imports. They are thus central to the estimation of carbon leakage rates and the effectiveness of BCAs. gtap-mecs includes sectors such as cement or aluminum which are the focus of partial equilibrium studies and are missing from GTAP. Fig. 1 compares the GTAP and gtap-mecs datasets along three dimensions which are important determinants of a sector's relative competitiveness under climate policy: energy intensity (amount of energy inputs per unit of output), import share, and output (all in value terms). As can be seen, there is a wide range of variability in these three dimensions across sectors and across the datasets. Full-size image (32 K) Fig. 1. Comparing GTAP and gtap-mecs sectors (2004 values). Figure options The GTAP and gtap-mecs datasets are used to calibrate a standard static constant-returns model of international trade based on the Armington differentiated goods assumption. In this class of general equilibrium models, increasing the degree of sectoral detail influences results in different ways: heterogeneity in CO2 intensities can change substitution possibilities in final and intermediate demand; a more detailed description of technologies and fuel mixes can affect energy input substitution possibilities and overall abatement costs; finally, disaggregation can affect the scope for import substitution if it changes the sector-level correlation of trade and CO2 intensities. The magnitude of these effects is an empirical matter and its estimation requires calibration. I implement a simple counterfactual policy in which CO2 emissions are reduced in a sub-set of countries. Results generated with the detailed gtap-mecs calibration are compared with those generated by the calibration of the same model to different aggregations of the industrial sectors available in GTAP. I find that, as expected, the range and standard deviations of sectoral impacts increases with disaggregation. The increase in detail can also lead in qualitatively different predictions for some sectors, and changes the relative ranking of impacts across industries. I then estimate aggregation bias at the GTAP-sector detail level: the difference between impacts estimated with a GTAP calibration and the re-aggregated impacts from the disaggregated calibration. The magnitude of this bias is estimated to be large, with considerable differences between sectors both in sign and magnitude. I show how within-sector heterogeneity which is not captured by GTAP is linked to these biases. Importantly, sector-level biases tend to average out, and the amount of aggregation bias which remains at the overall industrial level is moderate. Relative to estimates generated using GTAP, the decrease in industrial output is predicted to be about 40% smaller. This can be partly explained by the larger variance in industrial energy intensities in gtap-mecs than in GTAP. Variables related to trade are more affected by aggregation: the increase in industrial imports is about 50% larger, and the increase in the CO2 emissions embodied in these imports is about 60% larger. This is partly explained by a larger correlation between energy intensities and import intensities in gtap-mecs than in GTAP. However, this effect is not large enough to translate into a large bias in economy-wide leakage rates. Indeed, despite the fact that uncertainty due to unobserved parameters in the calibration prevents the identification of the sign of aggregation bias in leakage rates, results show its magnitude to be small. This is explained by the relatively small size of industry in overall output and CO2 emissions, and because the impact of sub-global policy on fossil fuel prices has a larger influence on overall leakage rates than import substitution. I thus conclude that the high sector-specific leakage rates indicated by bottom-up partial equilibrium studies do not translate into high economy-wide leakage rates when estimated in general equilibrium, even though industrial sectors see larger increases in imports than previously estimated. Industrial detail does matter, however, in the estimation of the efficiency of carbon-intensity based import tariffs (BCAs), which are perceived to reduce about one third more leakage using the disaggregated calibration. The remainder of the paper is structured as follows. After a brief literature survey in 2 and 3 presents the modeling framework. Section 4 describes the various data sources used to build the gtap-mecs dataset, as well as an exposition of the most critical assumptions. Section 4.4 reports descriptive statistics relative to industrial sectors, and compares the ability of GTAP and gtap-mecs to capture heterogeneity. Section 5 describes the scenarios and aggregation levels. Finally, Section 6 presents full calibrated general equilibrium model results as well as estimates of the aggregation bias.

The paper began with the observation that top-down general equilibrium models used to assess the efficiency of sub-global climate policies and leakage rates often work at low degrees of sectoral aggregation. Bottom-up partial equilibrium modeling exercises which focus on particularly affected sectors find leakage rates which far exceed leakage rates predicted by economy-wide models. Relying on the availability of yet-unused detailed industrial sector data, I have increased the degree of disaggregation in a general equilibrium model. The exercise has allowed the estimation of the sign and magnitude of aggregation bias which studies working with more aggregated datasets may suffer from. Table 11 summarizes the direction and magnitude of the biases which are caused by using either a single-sector industrial aggregation of GTAP (GTAP1) or the 16 industrial sectors in GTAP (GTAP16) instead of the more detailed gtap-mecs dataset. Instead of point estimates, the Table displays a range which reflects the uncertainty caused by assumptions made about energy intensities outside of the US and substitution elasticities. A calibration to GTAP biases industrial output change upwards, but the magnitude of the bias is modest: a GTAP-based calibration can get within 16% to 45% of the value predicted by gtap-mecs (depending on assumptions). The impact of disaggregation on energy input substitution possibilities and thus abatement costs cannot clearly be identified. In particular, I find that using estimated energy input substitution elasticities considerably increases industrial abatement costs relative to the usual Cobb–Douglas assumption, a fact which should motivate further research into the estimation of these elasticities. Table 11. Direction and magnitude of bias caused by excessive aggregation — magnitude computed as the ratio of GTAP results relative to disaggregated GTAP-MECS results. Aggregation bias Biases results: Is robust to assumptions: Magnitude of bias GTAP1 GTAP16 Industrial sector variables (%chg ref) Output upwards yes 21% to 72% 16% to 65% Energy demand upwards no − 15% to 47% − 11% to 56% CO2 emissions downwards yes − 68% to − 80% − 18% to − 63% Imports downwards yes − 68% to − 80% − 23% to − 52% Embodied CO2 in imports downwards no − 55% to − 77% 24% to − 36% Embodied CO2 in imports (at MECS intensities) downwards yes − 97% to − 98% − 18% to − 63% Economy-wide variables Carbon price (ref) downwards no 22% to − 18% 24% to − 17% Leakage (ref) upwards no 20% to − 3% 30% to 5% Leakage (ref_ffp) downwards no − 15% to − 49% 18% to − 29% Leakage (tariff) upwards yes 56% to 14% 51% to 11% Efficiency of BCA (%leakage reduced) downwards yes − 53% to − 58% − 23% to − 32% Tariff revenue downwards no 23% to − 12% 24% to − 11% Note: range of bias corresponds to different energy intensity and elasticity assumptions used in gtap-mecs; magnitude computed as the ratio of GTAP results relative to disaggregated gtap-mecs results. Table options In general, industrial trade response to carbon pricing is estimated to be lower in the aggregated than in the disaggregated models: the increase in industrial imports as well as the CO2 embodied in these imports is biased downwards. The amount of trade-related leakage (estimated by fixing fossil fuel prices) is thus underestimated, although the magnitude of this bias is modest. In particular, this bias is not large enough to substantially affect overall leakage rates, which remain largely unaffected by the level of industrial aggregation. Thus, it seems like the importance of industrial aggregation on leakage rates is less than what may be caused by other modeling assumptions, such as trade structure or industrial organization. In all cases leakage rates remain smaller than what might be interpreted from partial equilibrium studies. In conclusion, we can be reasonnably confident that sectoral aggregation in computable general equilibrium models does not lead to very large mis-estimations of leakage rates. These models can thus inform the policy debate about the overall (economy-wide) efficiency of climate policies. GTAP-based calibrations can provide a good approximation of most variables of interest. In many cases, this is also true even with the inclusion of only a small number of energy intensive sectors (crp, nmm, nfm and i_s). This study has shown, however, that disaggregation beyond the GTAP level does increase the estimates of trade-related leakage and that a higher degree of sectoral detail is important in determining the efficiency of border carbon adjustments: import tariffs in the disaggregated model are perceived to be about one third more efficient.