A Carbon Wealth Tax: Modelling, Empirics, and Policy

2.1 Carbon Pricing Mechanisms

Carbon pricing consists of two alternative policies: carbon taxation and cap-and-trade systems (Metcalf and Stock 2020; Stiglitz et al. 2017). They diverge in that while carbon taxes are levied on the use of fossil fuels, cap-and-trade sets a limit on emissions and allows producers to trade their “right to pollute”. In other words, carbon tax sets a price target and lets the quantity float, whereas the cap-and-trade system does the opposite (Stavins 2019).

The preference for one scheme over the other varies in the literature. Prominent names advocating for carbon taxes include Nordhaus (2008). The arguments range from the simplicity of the tax system (Stavins 2019) to avoiding the potential price volatility associated with a quantity constraint that may further deter the investment process. It has also been pointed out that a price-setting policy works better in a scenario with non-linear climate damages caused by carbon emissions and linear mitigation benefits. The opposite vision (Stavins 2019) argues that quantitative goals, such as emission abatements, are better tackled by cap-and-trade, due to their reliance on quantitative emission allowances. However, from the distributional and cost viewpoints, both devices produce equivalent outcomes (Stavins 2019).

Empirical work has found that carbon pricing has a discernible, albeit limited, impact on emissions. Among the multi-country studies, Best, Burke, and Jotzo (2020), using panel data on carbon emissions for 142 countries, found a 2 % decrease in emissions after the adoption of carbon pricing. On the other hand, Haites et al. (2018) investigates all the 55 jurisdictions that had at least one form of carbon policy in 2015. He found that the carbon tax reduced emissions concerning business-as-usual (BAU) scenarios, but not in absolute terms. Emissions Trading Systems (ETS) schemes have fared better, reducing more CO₂ while costing less. Metcalf and Stock (2020), estimating an SVAR model for 31 European countries, found that the scheme led to a CO₂ reduction between 3.8 % and 6.5 % according to the model specifications.

As for scheme-specific evaluations, there is mixed evidence on the European Union ETS effectiveness. One work finds a decrease in emissions between 2 % and 6.3 % during the program’s first 2 phases (Narassimhan et al. 2017), although accounting for the economic crisis may reduce this number substantially (Bel and Joseph 2015). Contrarily, Biancalani et al. (2024), using a methodology of matrix completion method, find a more substantial decrease of 15.4 %. Looking at other jurisdictions, Pretis (2019), using a difference-in-difference model for British Columbia, found that despite a 5 % reduction in transportation emissions, carbon taxation failed to impact aggregate pollution. Martin, De Preux, and Wagner (2014) investigating England’s Climate Change Levy, determined that it lowered CO₂ emission by 8.4 % and 22.6 % in firms subject to the levy compared to other firms. Lin and Li (2011), analyzing northern European countries, found that in 4 out of 5 of them, reductions remained between 0.5 % and 1.7 %.

It is clear from the above that, although the point estimates vary considerably, they all fall short of the reductions required to mitigate climate change. It has to be taken into account the fact that the EU ETS does not cover all sectors due to, among other reasons, political restrictions. Currently, it covers domestic aviation, heavy industry, such as steel, aluminum and chemistry, and power sectors (International Carbon Action Partnership 2024). The 4th phase of the program, that will set the rule for the system until 2030, will increase the scope to cover also sectors such as maritime transportation and buildings, while also increasing the program’s reduction goals (European Commission 2024). Thus, the emission results are driven in an important way by political factors, and are not a shortcoming of the instrument itself.^[2]

One of the features of carbon pricing that was developed to lower the political resistance against it was the use of revenues to lower decarbonization costs for firms and reversing the distributional effects. Therefore, the literature also analyzed additional benefits coming from the potential uses of carbon revenues. Indeed, revenues associated with carbon pricing are significant and increasingly so. Earlier literature showed, for example in 2013, that they totaled USD $27 billion, in 2017, USD $32 billion (Haites et al. 2018). For 2013, 70 % of ETS revenue was used for green subsidies, while 9 % was used to lower other taxation. For the carbon tax revenue, these figures were 15 % and 44 %, respectively (Haites et al. 2018).

The so-called second dividend, or revenue recycling, literature stresses that carbon revenues are a vital feature in securing efficiency in the policy choices (Bovenberg and Goulder 1996; Goulder, Parry, and Burtraw 1996). Provided that the second dividend is strong enough, a carbon tax could be adopted with zero net cost for the economy and also avoids the precise calculation of the environmental benefits of the carbon tax (Bovenberg and Goulder 2002). Historically, there were two forms of recycling: subsidizing green investments (Nordhaus 2013), or reducing other kinds of inefficient taxes (Bovenberg and Goulder 1996). In this respect, the current taxation mix is crucial, and a carbon tax can improve efficiency provided that it shifts the tax burden from inefficient, high marginal excess burdens to low marginal excess burdens.^[3]

So far, the majority of studies have suggested the use of carbon revenue to cut capital taxation (Bovenberg and Goulder 2002). Parry and Williams III (2012) find that the efficiency gain can be higher if the recycling scheme produces a shift of taxation from capital to labor. Some authors have gone even further to claim that the benefits of recycling exist only if this shift away from capital taxation takes place (Bovenberg and Goulder 1996). Alternatively, Metcalf (2007) proposes using carbon tax revenue to decrease income tax by issuing tax credits equal to payroll taxes. According to the authors, this tax mechanism is sufficient to improve the tax’s progressiveness. Overall, one main finding is that policy design is instrumental in shaping the distributional effects of carbon taxation.

2.2 Wealth Taxation

The literature has considered the use of carbon revenue to decrease capital taxation partly because economic theory modeling had traditionally maintained that optimality conditions include a zero rate for capital tax. However, there have been recent challenges to this result, which raises the question of using carbon pricing not to lower capital taxation, but to raise it.

A striking conclusion from the classical taxation models is that capital taxation should be zero. For example, Atkinson and Stiglitz (1976) used a life-cycle model to show that taxing only labor is always more efficient than taxing a mix of labor and capital. Moreover, the canonical models of Chamley (1986) and Judd (1985) showed that the steady-state optimal capital taxation is zero when the long-run capital supply is infinitely elastic.

However, recent developments have cast doubts on such results. Saez and Stantcheva (2018) show how incorporating wealth into the utility function produces heterogeneity in wealth (unrelated to heterogeneity in labor earnings), invalidating the zero capital tax result of Atkinson and Stiglitz (1976). Building on the same assumptions of the canonical models, Straub and Werning (2020) proved that intertemporal elasticity below one is already sufficient to produce positive capital taxation. Guvenen et al. (2019), in turn, studied an economy in which agents, because of their idiosyncratic abilities, can extract different returns from the assets. This heterogeneity is enough to yield a rationale for wealth taxation since it penalizes the idleness of the asset holder.

Yet, empirically, it is a well-documented fact that developed countries have extensively lowered corporate taxation in recent decades. On average, the statutory corporate income tax rate was around 33 % in 2000, dropping to less than 25 % in 2020 (OECD 2021). This downward trend was mirrored by an upward trend in wealth concentration. For the US, China, UK, and France, Alvaredo et al. (2017) found that since 1990 there is a clear upward trend in the share of wealth for the 1 % bracket.

Scheuer and Slemrod (2021) highlight the challenges associated with wealth taxation, noting that currently, only three OECD countries continue to impose such taxes. The decline in wealth taxation is largely attributed to its unintended consequences, including shifts in saving behavior, portfolio adjustments, and significant levels of tax avoidance and evasion. These challenges raise concerns about the effectiveness and enforceability of wealth taxes, particularly in the context of highly mobile capital. However, the carbon wealth tax (CWT), by targeting returns on carbon-intensive assets rather than net wealth in general, circumvents some of these issues by focusing on taxing revenue flows rather than accumulated wealth. Nevertheless, problems related to avoidance and evasion remain, necessitating strong international coordination, as further discussed in Section 5.

Thus, the historically low levels of corporate tax, together with the recent theoretical developments, support our consideration that the increase in carbon pricing required by climate change may assume the form of additional taxation on asset returns.

2.3 Tax Implementation and Incidence

Usually, a wealth tax’s base is the net worth of individuals or companies. This is the case in Saez and Zucman (2019) and Jakobsen et al. (2020), whose tax base is the household’s net wealth, including all financial and non-financial assets net of liabilities. There is also the proposal of Guvenen et al. (2019), where the tax base consists of all assets in the economy (thus ignoring the liabilities).

In the present work, we consider wealth taxation as an additional tax on (brown) capital returns. Although different in form, actually, the tax on a stock of wealth and on the flow of income from wealth can be equivalent. This is so because a sufficiently high capital income tax has the same effect as a small tax on the entire wealth. To see why, consider the following after-tax wealth equations from Guvenen et al. (2019), where w _i is the individual’s wealth, r is the return on wealth, τ _k is a capital revenue tax, and τ _w is a wealth tax. It is possible to write the following:

Combining Eq. (1) and Eq. (2) gives us a mapping from the capital income tax into wealth tax:

Therefore, there is always a (high) level of capital income taxation that corresponds to tax levied directly on wealth (Auerbach 2008).

There are important reasons that justify opting for taxing capital returns. First, the implementation is straightforward, since capital taxation is already adopted in a majority of countries, whereas net-worth taxation is much less so. Secondly, it overcomes opacity issues. As pointed out by Kopczuk and Mankiw (2019), net-worth taxation is not based on observable arm’s-length transactions, which would hinder the government’s oversight and give incentive to under-reporting.^[4] Finally, this option has been considered in policy discussions to tax wealth. For example, the 2023 Budget of the U.S. Government includes the introduction of a “tax on billionaires”, which relies on a special tax on investment gains for individuals whose net worth is above the $100 million threshold (Office of Management and Budget 2022).

Next, we discuss the crucial issue of tax incidence. The classic model of Harberger (1962) showed that, in a two-sector general equilibrium model, the sector in which the tax is levied is not necessarily the one that ends up paying for it. The result was that the capital taxation would be borne by the two forms of capital in proportion to their relative size (Auerbach 2006). In the case of CWT, this problem is less stringent. First, brown sectors are the greater part of the capital stock, so the incidence would still fall on the targeted sector. Secondly, the low substitutability between brown and green capital hinders tax shifting. Moreover, to the extent that substitutability happens, with brown capital moving to the green, tax-free sector, that is actually what is intended in a green transition context.

A potential shortcoming in our proposal is related to the tax’s efficiency. All wealth tax schemes are subject to prompt capital flight to foreign countries, thus increasing tax evasion problems. To the extent that the CWT is adopted by only one or few countries, it would be subject to the same criticism. However, capital flight depends crucially on the elasticity of capital supply. In this respect, there is at least some evidence that the elasticity is not high (Saez and Stantcheva 2018).

2.4 Identification of Green and Brown Assets at the Firm Level

Finally, the CWT relies fundamentally on the discrimination between green and brown assets. The literature on climate economics provides us with two ways of identifying green and brown activities: one at the sector level and the other at the firm level.

The sector-level identification is relevant for a broad range of climate topics, such as the impact of carbon pricing and green subsidies in technological efforts (Acemoglu et al. 2012) the study of the production function’s substitutability between clean and dirty energy (Malikov, Sun, and Kumbhakar 2018; Papageorgiou, Saam, and Schulte 2017), and forecasting the net impact of the green transition in employment levels and patterns (Markandya, Arto, González-Eguino, and Román 2016).

In practice, it is usual to assess the sectors’ carbon intensity using the Input-Output framework, which combines information of economic activity with the CO₂ emission per sector. Publicly available databases, such as the World Input-Output Database (Timmer et al. 2015) compile information spanning several years and almost 40 countries. Energy use data are collected from different sources, including the International Energy Agency, OECD, and Eurostat. Additionally, some countries’ statistics bureaus publish environmental tables in their national accounts. Notably, Germany has an accurate and detailed table, which has been used, for instance by Kato et al. (2013) and Mittnik and Semmler (2022), to differentiate between dirty and green sectors. Advantages of this approach include its broader scope since it (in principle) covers all the companies’ activities in each economic sector. Moreover, input-output methods, such as the Hilferding–Hirschman, allows the calculation of the downstream and upstream emissions associated with a specific economic activity.

However, in the context of taxation, the sectoral classification is not sufficient. For one, taxes are usually levied on firms, not sectors. Moreover, carbon pricing schemes aim at aligning incentives towards the adoption of mitigation and adaptation actions. In that case, the relevant differentiation is not between polluting and non-polluting sectors but firms. Fortunately, there is a second identification approach that assesses directly how much carbon emissions are associated with the activities of each company.

In advanced countries, governments are moving in the direction of mandating firms to disclose data on their environmental actions, in particular carbon emissions. Table 1 summarizes the carbon disclosure efforts in the European Union, the United Kingdom, and the United States. We refer to the Appendix for a fuller discussion of each action. In general, they target publicly listed or large companies, mandating the disclosure of climate-relevant information in the next few years. Importantly, in the case of the UK and the US, the carbon metrics underlying this regulation are the 3-scope methodology calculated by the GHG Protocol or the Carbon Disclosure Project. It is the same metrics that we rely on in our empirical strategy to classify green and carbon-intensive capital, which we detail in Section 4.1. As more and more firms use these metrics, it is expected that they will advance in terms of standardization, comparability, and availability of CO₂ emission metrics. Thus, reinforced by the expansion of ESG finance, we expect that carbon emission metrics at the firm level will be widely available information in the near future.

In summary, a CWT on capital returns in the brown sectors and firms is feasible and reasonably efficient, in the sense that it is levied broadly on the targeted factor.

4.1 Data

We estimated the returns on equity in the following way. First, we obtained daily equity prices from the S&P 500, the market index that tracks the 500 largest companies listed in the United States. We transformed it into a monthly series by selecting the last day of each month. The period covered is from January 2010 until September 2021, totaling 141 observations. Next, the individual company prices were aggregated into portfolios using the respective S&P 500 weights. Specifically, we computed the portfolio value as:

where P _t is the total portfolio value, ω _i is the individual S&P 500 weights, and ϵ_i,t are the individual stock prices. Finally, the resulting portfolio prices were used to obtain monthly logarithmic returns.

We classified firms as clean or dirty based on two metrics of the MSCI ESG index, which evaluates the environmental performance of over 8,000 companies and bonds. In this, we follow other works that have used ESG criteria to assess a portfolio’s carbon footprint (Bolton and Kacperczyk 2021; Jondeau, Mojon, and Pereira da Silva 2021; Schoenmaker 2021). The first metric is the “Carbon Emissions Weight,” which captures the importance of the company’s sector to environmental issues. The second metric is the “Carbon Emissions Score”, which grades from 0 to 10 the companies individually by their carbon emissions and climate mitigation efforts. This criterion relies on self-reported and Carbon Disclosure Project data.

First, we used the weighting index to keep only the 25 % highest weights in the dataset. Our use of sectoral emission data as thresholds to determine portfolios follows the strategy in Jondeau, Mojon, and Pereira da Silva (2021). In practice, this narrowed our sample to companies operating in the most crucial sectors of climate efforts. Next, we further subset the companies based on their performance grades. We classified the 30 % lowest-ranked firms as dirty, and symmetrically, the 30 % higher rates were classified as green.

Finally, we matched the two groups (dirty and green companies) with the S&P 500 to obtain equity prices. In the end, we were left with 38 green companies and 40 dirty companies. The list is reported in Tables 2 and 3 below. Although cumbersome, this process prevents us from considering a company as green if it emits a smaller amount of CO₂ just because it operates in a low-emitting sector. Instead, we categorize a green company as one with low emissions in high-polluting sectors. Finally, we performed a robustness check comparing the firms selected through our process to an alternative approach based on input-output data (see Appendix A) (Table 4).^[5]

As discussed in the previous section, asset returns are a central variable in the dynamic portfolio model. Therefore, we obtained harmonic estimations of returns on green and brown assets following the methodology in Semmler and Hsiao (2011), which relies on the fast Fourier transform (FFT) of the time series (see also Chiarella et al. (2016)). The advantage of FFT is that it captures low-frequency movements on the returns, subtracting the effect of short-term noises that are often irrelevant in portfolio allocation decisions. Appendix B reports the sine-cosine coefficient and the sum of squared errors related to the harmonic estimations. Figure 1 plots both estimations of the low-frequency behavior of green and brown assets. Consistent with other findings, brown assets are more volatile. Notably, green returns are more resilient to economic downturns.

Figure 1:

Harmonic estimations for asset returns. Note: This figure appeared first in Neves and Semmler (2024).

The dynamic portfolio optimization problem is solved numerically using the nonlinear model predictive control (NMPC) algorithm introduced by Grüne and Pannek (2012) and Grüne, Semmler, and Stieler (2015). We run the simulation for 40 periods for different tax regimes. In the business-as-usual (BAU) scenario, no tax is imposed. To assess the CWT’s impact, we run the model for after-tax brown returns (Equation (9)) and after-subsidies green returns (Equation (10)). We investigated scenarios with a rate of 20 % and 40 % to capture the influence of the magnitude. Finally, to assess the effects of the classic excise carbon tax, we run the model for each state equation (Equations (6) and (6’)) with rates of 30 % and 50 %. We present the main findings in the next section.

4.2 Results and Discussion

Our first finding is that, in a dynamic portfolio model, the excise tax that aims at penalizing emission-intensive consumption plays a limited role in changing the wealth dynamics. The underlying reason is that within a classic carbon tax environment, the investor lowers their consumption level so that the saved share of wealth remains the same. This adaptation behavior can be seen in Figure 2, where we plotted how much of the wealth the investor opts to consume over the periods in three different tax rates: 0 %, 30 %, and 50 %. The reduction in consumption virtually matches the hike in the rates so that the share of wealth not consumed stays the same. Importantly, given that asset returns are unchanged, the allocation decisions remain the same.

Figure 2:

Consumption decision in the classic carbon tax.

A lower disposable income following a hike in tax rates is a familiar feature as provided by the classic taxation models Barro (1974); Ljungqvist and Sargent (2018). However, whereas in the canonical settings households can counterbalance this by smoothing consumption intertemporally through government bonds, in our model, this channel is shut down by the balanced-budget assumption. Empirically, there is robust evidence of a low smoothing behavior by households, or what is labeled “excess sensitivity of consumption,” which further reinforces our finding (Flavin 1981; Jappelli and Pistaferri 2010; Mankiw 2000; Mankiw 2000, Flavin 1981, Jappelli and Pistaferri 2010). The same seems to happen in carbon pricing. Kanzig (2021) studied the European carbon market and showed that households lower their consumption level in the face of a carbon tax shock.

Thus, precisely because of the pass-on effect, the excise carbon taxation has a (very) limited impact on wealth dynamics and investment patterns. On the other hand, that is precisely what is achieved by taxing brown capital directly. In that sense, the CWT is an innovative policy instrument since it addresses a variable – equity returns – that has so far remained untouched by climate policies. Figure 3 evaluates how the investor’s total wealth behaves over 40 periods under different tax regimes. In the business-as-usual scenario, wealth decreases more or less steadily as consumption outpaces the investment’s returns. Between periods 10 and 15, the decline is less pronounced as returns on both assets increase substantially (see Figure 4 below).

Figure 3:

Wealth dynamics under different tax schemes. Note: As observable from the above figure we have chosen a parameterization of the model such that wealth will be depleted in finite time.

Figure 4:

Dynamics of returns on assets. (a) Carbon-Intensive Assets. (b) Green Assets.

A CWT is capable of altering the wealth dynamic through its direct impact on capital returns, but the level of taxation matters. At 20 %, there are only slight changes in the wealth trajectory. By lowering brown returns, CWT makes investors shift to green investments early on, and the lower volatility of the latter contributes to a slightly lesser decline. On the other hand, at 40 %, the taxation is stronger and meaningful alterations happen. As expected, taxing asset returns impacts the trajectory negatively in the first few periods. Around period 5, the trajectories diverge, with taxation decreasing more smoothly. Such early movements, moreover, have long-lasting effects on wealth. Until the end, wealth under the more aggressive tax regimes remains higher, indicating a lasting improvement in the wealth path. On the other hand, we also note that subsidies for green technology have transitory effects. Its strongest influence is felt in the middle periods, where the share of green investments is greater.

Indeed, a CWT alters wealth paths precisely because of its ability to change asset returns. Figure 4 plot the total calibrated behavior of brown and green yields, respectively. Common to both assets, we see that the BAU scenario is changed substantially in the early periods. This is a consequence of the fact that the CWT’s tax base – wealth – is larger in the beginning, and diminishes as wealth is consumed away in the later periods. As expected, CWT decreases brown asset returns proportionally to its taxation levels, with subsidies having the opposite effect on green returns.

The ultimate impact of taxation regimes on the green transition can be seen in Figure 5, which plots the behavior of π₂, the share of the risky portfolio allocated to green assets. We note that as the CWT increases, π₂ is higher in a greater number of periods. Again, the effect is stronger for earlier periods because of the aforementioned tax-base effect. However, we stress that in a green transition context, this is a very meaningful result, given the importance of a fast action on green investment in spurring technological innovations.

Figure 5:

Portfolio allocation under three levels of taxation.