Climate Transition Risk Factor

Through the Paris agreement, the international community has committed to keep global average warming below 2℃, along with a more ambitious objective of 1.5℃. In addition to the physical effects of climate change, the economic transformations required to reach this objective will affect (positively or negatively) certain economic sectors more than others (IPCC, 2022). From an investor perspective, these transformations will generate new transition risks and it is therefore necessary to identify the companies that best anticipate regulatory, technological and market developments to manage them.

Transition risks are difficult to estimate using fundamental approaches. First, despite reinforced regulatory requirements and recommendations, persistent gaps in climate-related data remain ^[1]. Secondly, the radical uncertainties associated with transition scenarios are difficult to incorporate into fundamental valuation models ^[2]. As a result, transition risk metrics display a significant degree of diversity ^[3].

Against this backdrop, academics have sought to measure transition risks directly from market prices. This approach relies on the ability of markets to process information in real-time, which reduces the data and model barriers mentioned above. So far, the effort has focused on building climate transition (CT) factors. These factors are designed on the same principle as traditional factors (e.g., size, value): they are portfolios constructed in such a way that their price changes are representative of the dynamics of the stocks affected by the transition risks.

The methodology we present aims to contribute to this literature on price-based analysis of transition risks by addressing two main conceptual issues. The first one is related to the design of a CT factor. While some papers rely solely on carbon intensity, i.e., the greenhouse gas (GHG) emissions of a company divided by its revenues, others use up to ten metrics to build their representative portfolio ^[4]. The type and number of metrics raises questions regarding their current availability, quality, and their relevance to assess long-term transition risks. Our approach departs from previous attempt at producing a CT factor based solely on individual company characteristics. Instead, we utilize what is likely to be the most robust information regarding a company’s exposure to transition risks: its industrial sector. We introduce a new CT factor that relies on i) the climate-policy relevant industrial sectors (CPRS) classification developed by Battiston et al. ^[5], and ii) the carbon intensity to differentiate companies within these CPRS sectors.

The second issue of price-based analysis of transition risks is related to the use of a CT factor in a risk model. Investors have started considering transition risks relatively recently: 2015 is a pivotal year with the Paris Agreement and the warning by Bank of England Governor Mark Carney ^[6]. Because the traditional tests to validate the relevance of a factor rely on long timeframes, CT factors usually do not pass these tests and are therefore not qualified as “proper” risk factors ^[7]; ^[4]. We propose a different approach that focuses on the practical management of transition risks by disentangling the links between a portfolio's exposure to the CT factor and the traditional ones.

Construction of a Climate Transition factor

A climate-transition (CT) factor is meant to capture the exposure of a portfolio to the energy transition by constructing a signal that is positively correlated to companies that might suffer from an abrupt transition, and negatively correlated to companies that might benefit from this transition.

The energy transition has both a sectoral and a company-specific dimension. First, the extent of the transformations brought by the transition depends on the sectors, as the abatement cost of GHG emissions is directly related to the sector technologies (IPCC, 2022). Sector treatment in the construction of the CT factor is therefore a major concern. On the one hand, energy transition cannot be expected to wipe out entire branches of the economy, which argues in favor of a sector-neutral approach. On the other hand, some sectors will be affected more than others by the energy transition. In order to address both concerns, we design our factor as follows.

First, we narrow down the investment universe to climate-policy relevant sectors (CPRS) as defined by Battiston et al. ^[5]. This classification identifies sectors whose primary economic activities “could be affected, either positively or negatively in a disorderly low-carbon transition [...] considering (i) the direct and indirect contribution to GHG emissions; (ii) their relevance for climate policy implementation [...] (iii) their role in the energy value chain” and has been used by several financial regulators to assess the exposure of financial institutions to transition risks (^[8]; ^[9]). This first source of information for the design of our CT factor is both robust (the sectoral affiliation is easily accessible) and forward-looking (the classification is established based on disorderly transition scenarios).

The goal of the second step is to identify companies within these sectors that may benefit (or suffer) from a disorderly transition in the long run. Ideally, this step would be carried out on the basis of multiple company climate-related data. For example, Görgen et al. ^[4] compute a “Brown Green Score” from ten variables containing company specific information related to value chain, adaptability, and public perception. However, these data remain scarce and are not available for all sectors ^[1]. Moreover, Roncalli et al.^[10] showed that the composite indicator built by Görgen et al. ^[4] is well captured by a factor based on the GHG-emissions intensity only. We therefore consider GHG emissions intensity as a robust and still relevant metric for identifying companies exposed to transition risks within the climate-policy relevant sectors.

This brings up the question of the scope of the emissions to be considered. The objective of our factor is to capture a market signal that is expected to evolve as new information becomes available, in particular when companies' greenhouse gas emissions are updated. It is therefore important that the granular design of the CT factor is based on consistent data regardless of the data provider used by market participants. By comparing seven data-providers, Busch et al. ^[11] highlighted strong inconsistencies in indirect (scope 3) data, whether reported by companies or estimated by external parties. To be as robust as possible, our CT is therefore only built on the GHG emissions intensity on the direct (scope 1) and energy consumption (scope 2) perimeters, for which there are higher levels of data consistency. Not explicitly considering Scope 3 emissions is also consistent with our sectoral approach. As pointed out by Ducoulombier ^[12], reporting standards are not intended to support comparisons between firms, and estimates take insufficient consideration of firm-level circumstances to support intra-sector comparisons. Therefore, a company's Scope 3 emissions data (currently available) is essentially linked to its sector. This information is already taken into account in our factor by the CPRS classification.

Finally, our CT factor is constructed as follows: the long (“brown”) leg is built as an equally weighted (EW) portfolio of the 50% most GHG emissions intensive stocks selected within each of the six CPRS sectors. Symmetrically, the short (“green”) leg is built as an EW portfolio of the 50% least GHG emissions intensive stocks selected within each of the six CPRS sectors. Then, the weight of each leg is set so that the factor is market neutral. In this way, we assume that the CT factor should not contain any market risk. This approach is consistent in the context of asset management, where the market serves as a benchmark for risk.

Use of the Climate Transition factor for risk management

From an asset pricing perspective, a risk factor is worth adding to a model when it helps to better explain the cross-section returns (^[13]). As proposed by Fama (^[14]), we performed an ordinary least-square regression of the CT factor against traditional risk factor models to measure the regression intercept (represented as $\alpha$) $$F_{CT} = \alpha+\sum\limits_f\beta_fF_f+\varepsilon.$$ Here, the factors belong to a set of traditional risk factors. We ran a regression between the CT and traditional factors, realized between 2012 and 2022 using weekly returns. We found that for different sets of risk factors, the intercept of these regressions is close to zero and statistically not significant, as shown in the table below.

This indicates that the CT factor does not possess a premium that is not explained by already existing factors (consistent with Görgen et al. ^[4] and Amenc, Esakia & Goltz ^[7]).

From a risk perspective, a risk factor might be worth adding to a model if it helps to better explain the variance of the returns. To decide whether this is the case, we evaluate the semi-partial correlation between the returns of a sample of funds and a set of risk factors including the CT factor (see Cohen and Cohen, 1975 ^[15]). The semi-partial correlation determines how much of the return variance is explained uniquely by the CT factor. The next table shows that the addition of the CT factor does not improve the model’s ability to explain risk in a significant manner, as it contributes very insignificantly to the model’s R-squared.

These tests confirm that the addition of the CT factor does not improve the performance of existing asset pricing models: the information contained in the transition factor is already spanned by existing financial risk factors ^[16]; ^[7].

However, a risk factor does not need to improve the power of an asset pricing model to be relevant for risk management ^[17]. The CT factor, as we defined it, is indeed representative of stocks that are sensitive to the energy transition. As such, especially for institutional investors with long-term horizons, it is useful to understand how they are exposed to this segment of the market, and to be able to disentangle the resulting risk from other financial risks. To do this, we propose to filter the CT risk from traditional risk factors, $$r_i = \beta_{i,CT}+r^{\prime}_i.$$ Here, the residual return denoted by $r^{\prime}_i$ corresponds to the returns of the $i^{th}$ instrument that have been filtered from the effect of the CT factor. These returns are not exposed to CT risk. The CT risk associated with this instrument is thus given by measuring the part of the risk that is due to the exposure to the CT factor, that is $$\text{Active CT risk} = \sqrt{\text{Var}(r_i-r^{\prime}_i)} = \sqrt{\beta^2_{i,CT}\text{Var}(F_{CT})}.$$ Because we have made the factor market neutral, this risk does not include any market risk, hence the “active” denomination. If we now regress the returns and filtered returns of each instrument on traditional financial risk factors, we obtain two sets of exposures. One set corresponds to the exposures of the $i^{th}$ instrument's returns to the $f^{th}$ traditional risk factor. $$r_i = \sum\limits_f\beta_{if}F_f+\varepsilon_i.$$ The other set (denoted by $\beta^{\prime}_{if}$) corresponds to the exposures of the instrument’s filtered prices to the same traditional risk factors.

$$r^{\prime}_i = \sum\limits_f\beta^{\prime}_{if}F_f+\varepsilon_i.$$ These two sets of exposures yield a new relationship for the CT risk that involves only financial factors. $$\text{Active CT risk} = \sqrt{\text{Var}(r_i-r^{\prime}_i)} = \sqrt{\beta^2_{i,CT}\Sigma_F\beta^T_{i,CT}+\frac{1}{T}\sum\limits_t(\varepsilon_{i,t}-\varepsilon^{\prime}_{i,t})^2}$$ where $\Sigma_F$ is the covariance matrix of the financial risk factors and the components of the vector $\beta_{i,CT}$ are given by $\beta_{if,CT} = \beta_{if}-\beta^{\prime}_{if}$. The CT beta $\beta_{if,CT}$ is the difference between the traditional beta and the filtered beta. It corresponds to the financial exposure of an instrument that comes from the part of its returns that are exposed to CT risk. For instance, if many Value stocks of a portfolio are located in CPRS sectors for instance, the CT beta associated with the Value factor will be high. It also means that if a portfolio’s exposure to CT were to be completely eliminated, its exposure to the $f^{th}$ financial factor would be reduced (or increased) proportionally to $\beta_{i,CT}$. The CT beta thus actually helps to disentangle financial risks from CT risk. There is indeed a straightforward relationship between $\beta_{if,CT}$ and financial factors $$\beta_{if,CT} = \beta_{if} \times \varrho_{f,CT}\frac{\sigma_{CT}}{\sigma_f},$$ where $\varrho_{f,CT}$ is the correlation between the $f^{th}$ financial risk factor and the CT factor.

Importantly, the definition of CT risk connects the CT factor to existing financial risk models. It shows that it is possible to measure transition risks using already available financial time-series models. It only requires the simple step of computing an extra set of betas from filtered returns. The above definition also allows the derivation of a method to minimize the CT risk of a portfolio by solving the following minimum variance problem. $$\text{min}(w^T\Sigma_F^{CT}w)$$ where $\Sigma_F^{CT} = \beta_{CT}\Sigma_f\beta^T_{CT}+D_{\varepsilon}$ and the elements of the diagonal matrix $D_{\varepsilon}$ are equal to $\frac{1}{T}\Sigma_t(\varepsilon_{i,t}-\varepsilon^{\prime}_{CT}+D_{\varepsilon})$.

Because the market-based CT factor is of the same nature as the financial factors, it can be managed with the same tools. This is one of the greatest practical advantages of a market-based estimation of the CT factor from a portfolio management viewpoint.

Holdings-based transition risks

For a holdings-based approach, we compute the percentage of stocks within a portfolio that have a negative exposure to climate transition risk. This is done by analyzing the carbon intensity of each stock within a portfolio compared to the universe median carbon intensity of the relevant climate policy relevant sector. Stocks are classified based on the following criterion: a stock with a lower carbon intensity than the universe median carbon intensity of the respective sector is categorized as “green”. Conversely, a stock that has a higher carbon intensity than the relevant sector’s universe median is categorized as “brown”. The “brown” stocks are considered more likely to be negatively impacted by a transition to a low-carbon economy.

Comparison of holdings-based and market-based measures of transition risk

While the first goal of the CT factor is to capture sensitivity to the long-term transition, such a factor should already allow the identification of funds considered as "green" or "brown". A second test then consists in measuring the extent to which the sensitivity of a fund to the various factors is consistent with its current “green” or “brown” characteristic (estimated by a third party). We consider a sample of “green” funds as the top decile of funds within a universe of 600 funds and ETFs offered in the US market) based on the average share of corporate revenues that contribute positively to the climate mitigation. Conversely, we define a sample of brown funds as those with the highest transition risk score. We show that the betas of the “green” (respectively “brown”) funds to the CT factor are significantly lower (respectively higher) than the average betas of the funds of our universe. This confirms the ability of the CT factor to identify “green” and “brown” funds from their prices.

However, a portfolio categorized as “brown” or “green” based on its CT risk exposure will not always be aligned with the holdings-based classification. This can be attributed to the fact that the market-based approach is based on prices, which represent a large amount of information “digested” by market participants and can therefore capture more information than sector and carbon intensity. Two companies with the same carbon intensity and belonging to the same sector may indeed be impacted differently by the energy transition, for example if the regulations applicable in their respective countries of activity differ (presence or absence of carbon taxes) or if one of them has been the subject of climate change controversy. A market-based approach captures these kinds of differences, whereas a holding-based approach only focuses on a small number of transition-related metrics.

References

Back to top