C Resultssupplements
C.1 Complementaryresultfigures
We present first the details of the shock used in subsection 4.3, based on the market reaction to transition risk. The mapping adopted to generate the shock is of the form
(16)
where
,
and ?2 > 0 are
real parameters. The function is chosen so that the resulting return quantile
appear sensible and in line with the narrative of the shock. The resulting
shock is represented in figure 16a.
For the market shock based on physical risk that is used in subsection 4.4, the mapping used to define the shock is of the form
, (17)
with ?0 ? (0,1), ?1 ? (?0,1] and ?2 ? R+. The resulting shock is represented in figure 16b.
C.2 Resultsfromamarketshockonbrownerfirmsbysector
We investigate in this section an alternative narrative to the one of subsection 4.3, where instead of shocking securities based on their absolute carbon intensity, we consider how polluting they are relative to other firms from the same sector. The narrative is now of a shock that stems from a green investing endeavour by sophisticated investors who use carbon emission information to foster the least polluting companies of each sector, and penalize the most polluting ones. Alternatively, this could stem from an application of more stringent carbon pricing whereby part of the additional cost is passed downstream but such that the most polluting firms within each sector are forced to internalize more of the cost in order to remain competitive compared to less polluting firms.
In practice, using the sector decomposition as shown in figure 14a, we can assign to each firm a sector position SP?, given by the empirical cumulative distribution function of carbon emissions in the sector of ?. Thus, a value close to 0 means that the issuer of ? is part of the greenest firms in its sector, and among the brownest for a SP? close to 1. Note that for about 38% of the sample the sector information is missing, and these firms are grouped together as if forming another sector. Thus, we define the market shock from a quantile mapping function that combines both the within-sector position and the original carbon intensity:
??
= (Q? ? DSec)(?), D
, (18)
with ? ? [0,1] the parameter that governs the relative importance of the within-sector position. For our application we pick ? = 0.8, in order to give predominance to the within-sector position.
The shock generated is represented in figure 18a, with a separation between the debt and equity classes. As we would expect, the correlation between the carbon intensity and the shock has become weaker than it was in the initial transition shock of figure 16a. Indeed, a number of securities that are not at the tail of the overall distribution carbon-wise, are now very stressed as they represent the worst polluters in a given sector. Corresponding results are presented in figure 18b. We observe that results present similarities to the initial transition shock, but investment inflows to green funds result in a mild amplification, even for the browner funds.
C.3 Resultsfromashockongreenassets
Recent data on investment funds show that those labelled ESG or SRI are prominent drivers of the industry growth. This growth of sustainability-labelled funds, which are mostly included in these two categories (where not wrongly assimilated to them), in its steady aspect alone has raised the concern that green stocks are overvalued[33] because they are part of a pool of assets that is too small. Thus, it is argued in van der Beck [97] that strong flows to ESG funds have artificially driven up the price of their holdings over the past years, also in line with a concern-driven” performance of green securities by Pástor et al. [79]. The implication that is advanced is that reduced future flows to sustainable funds could harm the price of the highly ESG-rated securities from a lesser purchasing pressure, although this argument is contingent on a broadly unchanged economic framework. For instance, a strong commitment to a rapidly increasing carbon price in the medium term could change the perspective, as could legal liabilities of firms related to ESG criteria. Nonetheless, given the uncertainty surrounding these developments, a market correction of the price of current sustainable assets” i.e. a green bubble” cannot be disregarded.
We model the occurrence of a shock on green assets is modelled as the opposite of the transition risk market shock (although with different parameters to avoid a large positive shock on brown assets), i.e. being influenced by both within-sector considerations and absolute carbon emissions.
Thus, we get
?? = (Q? ? DGB)(?), DGB
The shock thus generated is represented in figure 19a, where we observe a shape opposite to the previous ones, i.e. a net loss on the greenest assets and small gains on assets that are relatively brown. Interestingly, we observe from the results in figure 19b that the consequence on funds are quite homogeneous relative to the previous transitory shocks. That is, greener funds tend to lose more, but losses of brown funds are comparable. Differences between the groups are even smaller at the point of considering the second-round effects. Moreover, the initial shock that we derive here is of relatively small magnitude. The key difference with the transitory shocks is that shortterm dynamics are the only ones likely to materialize. Therefore, this indicates that damages from a green bubble” would propagate to a large part of the investment funds system but with a limited impact.
C.4 Shockmethodologycomparison
It is worth comparing the methodology employed here, for the generation of input shocks in particular, to the case study on investment funds conducted in Carlin et al. [26]. While it does not model contagion, [26] is one of the closest works to ours, given its focus sector, its emphasis on the importance of short-term climate stress tests, and having low-carbon assets with positive shocked returns. Using our notation, their methodology could be represented by the following equation:
?? = B(TIC(fMSCI(LCT?),Sec(?),Country(?))), (20)
which is composed of the following:
1. The equity price shock function fMSCI is calibrated using MSCI Low Carbon Transition (LCT) scores.
2. Company-level baseline shocks fMSCI(LCT?) are calculated, where LCT? is the LCT score of the company issuing ?. Industry-level shocks are then derived as weighted averages from the firms they include.
3. Industry-based and country-based caps/floors are applied (function TIC), including a minimum shock to all companies based in a major oil-producing country.
4. Shocks are bucketed to a discrete grid (function B).
The price changes then affect investment funds on their portfolio, similarly to our first-round impact.
Aside of the fact that we have further dynamics in our model following the initial shock, we can identify two features in which our shock generation differs conceptually from [26]. The first one is that their shocks do not depend on the financial properties of the securities, relative to the way we interact carbon emission intensities with the distribution of returns, and in particular the historical variance observed. This reflects a difference in narrative, whereby we try to better capture a plausible immediate reaction of financial markets, while [26] may be based on a change in fundamentals that moves the market over a somewhat longer time frame.
Second,
the choice of applying relatively uniform shocks within each sector is
something that is not present in our main simulations presented in Section 4. It is opposed but complementary to the
narrative that supports our additional exercise presented in Appendix C.2. Where [26]
assume that the market would react primarily on sector information, we assumed
that it would respond given a more accurate data access and favour best-in-class
companies. The relative plausibility of both alternatives is not settled and
may depend on factors such as the extent to which carbon price increases would
be passed on to clients downstream, or the degree of competition existing
within a sector.
(a) Average carbon emissions by sector (b) Average physical exposure by country
Figure14: Decomposition of climate data along dimensions of interest.
Left panel: carbon emissions on scopes 1 and 2 (in log scale on the x-axis) by sectors of firms (on the y-axis). Only sectors with more than 50 entities are kept. Additionally, data analysis not presented here shows that there are also marked differences between countries with regard to carbon intensity as some of them concentrate a larger number of polluting industries. Right panel: some important variations between countries (on the y-axis) are observed with regard to climate physical risk exposure (on the x-axis). Only countries with more than 100 entities are kept. The blue dashed line represents the euro area average.
Sources: Urgentem, 427 and authors’ calculations.
|
|
|
(a) Carbon intensity categorization (b) Physical risk categorization
Figure15: Grouping into deciles of risk as used in the simulation framework.
Left panel: all funds whose securities with available carbon ratings represent more than 50% of their portfolio by market value distributed into different deciles. The allocation is based on the fund-level carbon intensities defined by equation (10). Category 1 represents the greenest 10% of classified funds, category 2 the 10% to 20% greenest, etc, and NC includes all non-classified funds (falling below the 50% threshold). The column with mean carbon intensity is an unweighted average across all funds in a category. Note that, although all securities have a positive carbon emissions associated, some funds may have a negative weighted average because they short polluting securities.
Right panel: all funds whose securities with available climate physical risk exposure represent more than 20% of their portfolio by market value distributed into different deciles. The allocation is based on the fund-level physical risk exposures defined by equation (11). Category 1 represents the 10% least exposed of classified funds, category 2 the 10% to 20% least exposed, etc, and NC includes all non-classified funds (falling below the 20% threshold). The column with mean physical risk is an unweighted average across all funds in a category. Source: Urgentem, 427, Refinitiv and authors’ computations.
i y
i i y i i y
(a) Transition risk market shock
(b) Physical risk market shock
Figure 16: Bi-dimensional histogram of the market shocks used in 4.3 and 4.4, with a decomposition between asset types. The x-axis gives the carbon intensity of the assets while the y-axis gives the shocked returns. The colour of a square corresponds to the number of securities that present the corresponding risk/shock profile. Source: authors’ calculations.
J J l J J l J J l J J J l J J l J J l J 7 8 9 7 8 9
(a) Transition risk market shock (b) Physical risk market shock
Figure17: Time series of indirect severity.
Left panel: indirect severity for different months, based on the same transition risk market shock that is used in section 4.3, and represented in figure 16a. Right panel: indirect severity for different months, based on the same physical risk market shock that is used in section 4.4, and represented in figure 16b. For both, the indirect severity is defined in section 4.1 relative to a uniform shock with similar aggregate direct impact. Source: authors’ calculations.
(a) Transition risk market shock
(b) Decomposition of outputs
Figure18: Shock and results for the simulation penalizing worst in class” securities. For the upper panel, the x-axis gives the carbon intensity of the assets while the y-axis gives the shocked returns. The colour of a square corresponds to the number of securities that present the corresponding risk/shock profile. For the lower panel, the x-axis corresponds to groups of funds arranged by deciles according to the values of their carbon-weighted assets.
The y-axis corresponds to the size of gains and flows relative to the initial value of assets, all aggregated by group. Source: authors’ calculations.
(a) Market shock
(b) Decomposition of outputs
Figure19: Shock and results for the green bubble” simulation.
For the upper panel, the x-axis gives the carbon intensity of the assets while the y-axis gives the shocked returns. The colour of a square corresponds to the number of securities that present the corresponding risk/shock profile. For the lower panel, the x-axis corresponds to groups of funds arranged by deciles according to the values of their carbon-weighted assets. The y-axis corresponds to the size of gains and flows relative to the initial value of assets, all aggregated by group.
Source: authors’ calculations.
Acknowledgements
We thank Irene Monasterolo for her comments on earlier versions of the manuscript, an anonymous referee for the ECB Working Paper Series for their suggestions, participants of the Steering Financial Markets in the Sustainable Transition” conference and of the ECB Climate Change Centre seminar series for their constructive questions and comments.
RG acknowledges the financial support of the Erasmus+/Knowledge Alliance project [GrEnFin, grant number 612408-EPP-1-2019-1EPPKA2-KA], and the technical support of the European Central Bank. Any views expressed are those of the authors alone and do not necessarily represent the views of the ECB or the Eurosystem.
Régis Gourdel
Vienna University of Economics and Business, Vienna, Austria; email: regis.gourdel@wu.ac.at
Matthias Sydow
European Central Bank, Frankfurt am Main, Germany; email: matthias.sydow@ecb.europa.eu
© European Central Bank, 2022
Postal address 60640 Frankfurt am Main, Germany
Telephone +49 69 1344 0
Website www.ecb.europa.eu
All rights reserved. Any reproduction, publication and reprint in the form of a different publication, whether printed or produced electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.
This paper can be downloaded without charge from www.ecb.europa.eu, from the Social Science Research Network electronic library or from RePEc: Research Papers in Economics. Information on all of the papers published in the ECB Working Paper Series can be found on the ECB’s website.
PDF ISBN 978-92-899-5469-3 ISSN 1725-2806 doi:10.2866/258109 QB-AR-22-122-EN-N
[1] As investment funds appear vulnerable to climate shocks, they were included in the long-term portfolio losses projections of ECB/ESRB [40], which is itself in line with Battiston et al. [12].
[2] Examples include Poledna et al. [81], Roncoroni et al. [85], and Wiersema et al. [100], although the counterparty risk transmission channel does not necessarily rely on an Eisenberg and Noe-type clearing mechanism as we do.
[3] More precisely, in the case of a vector a we have a+i = max(ai,0) and ai- = min(ai,0).
[4] Note that holdings of tradable securities are allowed to be negative, i.e. funds can short them, while values for crossholdings are all non-negative. Moreover, when j is a close-end fund we have ?i,Ri,j = 0, as shares of j are counted as tradable securities.
[5] In the case of sales or redemptions, a decoupling occurs such that the NAV does not change but the TNA does.
[6] Such iteration is at the core of Chrétien et al. [33], and used in Semieniuk et al. [89], where the new information of lower prices creates a shock to portfolios, based solely on financial entities posting new information about their value.
[7] Consider that it means the total assets of these funds, except fund shares within the group, equals its total external debt. This is unlikely, first because funds’ profits are primarily meant to go to shareholders and not (only) to creditors, and second because most investment funds have a low leverage, often by law.
[8] The presence of a feedback loop with other agents, contemporaneous to the one taking place between funds, could invalid equation (3) in a larger financial system.
[9] The assumption ?i > -1 ensures that there is no complete run from investors, and therefore no default consequent to this step.
[10] Grill et al. [50] relate how this was used in the face of the market shock induced by the COVID-19 pandemic. They observe a stronger usage by more vulnerable funds, i.e. leveraged, illiquid, or with little cash holdings.
[11] Although we depart from it, Zeng [101] provides an important model explaining why funds would try to replenish their cash buffers in the period following redemptions. Chernenko and Sunderam [32] and Huang [55] provide further background regarding the cash management of funds.
[12] It is symmetric compared to most of the related literature that does not consider purchases, because they have a shock that is negative across all assets [see e.g. 24], and thus are purely asymmetric. Note that we can make the reaction easily asymmetric by defining two coefficients d? and h? instead of a single d? used on both sides.
[13] On the contrary, trading against constrained funds has been identified as a profitable strategy [38]. A configuration where funds in our sample benefit from this effect is possible, e.g. if they are sufficiently shielded from negative shocks but have some brown positions that they can expand by taking advantage of browner funds’ distress.
[14] Comprehensive results for other months are generally not presented here but are available upon request.
[15] More precisely, the filtration applied to the data in order to select funds has four components: funds needs to have at least some tradable holdings (standard stocks or debt securities) or redeemable holdings (the fund shares issued by other funds included), they should have a positive initial equity, this equity cannot be smaller than the total market value of the shares they issued that are reported in the portfolios of other funds, and finally they must not belong to a group that would cause regularity to be breached.
[16] This includes direct carbon pricing, stronger fines in case of breaches to environmental regulations, costlier risk management, etc.
[17] Given an asset ? for which we know the country and economic sector of the issuer, if the set of firms in that countrysector intersection is larger than 20, we average their carbon intensities and use it as a proxy for ?, otherwise we use the average of the whole sector across countries. Sector information is taken from NACE codes, at the two-digit level.
[18] See for instance Janssen et al. [63] for the bias of carbon intensity to shifts in exchange rates.
[19] Recent work at ESMA has already emphasized this interconnection between brown funds and its implication in case of shocks [4]. Following the framework of Cont and Wagalath [36], one could suspect that even if green and brown assets were fundamentally uncorrelated or negatively correlated, the distressed selling by funds would induce a positive excess correlation between them.
[20] Nonetheless, one could explain this discrepancy by the lack of maturity of green financing, and we cannot exclude that sustained flows to green funds eventually modify the sub-sector topology. Thus, the appearance of hubs among green securities is a plausible future stability threat.
[21] Given an asset ? that cannot be mapped to the Four Twenty Seven dataset, we may know its country and postcode from the Register of Institutions and Affiliates Data (RIAD) database. In that case, we proxy its exposure as the average from firms sharing the same postcode, with the condition that there exist at least five of them. 22See Battiston et al. [10] for a stress test of these securities.
[22] The market bias toward more carbon intensive firms is already known in the literature and affects other institutions such as central banks [71].
[23] Some scenarios such as the ones used by the EBA also feature shocks on securities with little granularity, i.e. defined at the country level.
[24] The largest flow-performance coefficient of 0.8 that is used here is larger than our calibrated coefficients for negative returns given in 4.3, but smaller than the ones used for positive returns.
[25] The fire sales step introduced in 2.5 is consistent with it, as the extension of portfolios takes priority over the purchase of new assets.
[26] The lower bound allows for complete defaults while preventing prices to go negative, and as we do not model large positive shock the upper bound prevents large returns that could be driven by data imperfections. Moreover, this roughly corresponds to the 1% highest absolute monthly returns.
[27] The stochastic approach to physical risk is also in line with what Budnik [23] advocates for, in the case of climate banking stress tests.
[28] The lack of attention to the issue of physical damages is in line with findings from Naran et al. [77] that adaptation finance still mostly comes from the public sector.
[29] See for instance Popescu et al. [82], which reviews the current options for investment funds to do so, and Afota et al. [1] that details the action of Banque de France with regard to its own portfolio, also replicable to a large extent by private financial actors.
[30] The same result can be obtained from fields other than linear algebra. First, it is known in input-output analysis, with an equivalent proposition in Karlin [67, theorem 8.3.2], a result that describes cases where the Leontief matrix is well-defined, also related to the Hawkins-Simon conditions [see for instance 72]. It can also be obtained from spectral graph theory. Then, X can be identified as a strict generalized random walk Laplacian in the sense of Veerman and Lyons [99], meaning a matrix of the form In - a · S where a ? Rn, with a < In and S ? Rn×n is a (row) stochastic matrix, corresponding to the normalized adjacency matrix of the graph of the fund system.
[31] It is more general than the case of a strictly dominant diagonal and encompasses it. Note that these notions are often defined from taking row-wise sums, but invertibility is unchanged when considering the transpose.
[32] We have excluded assets that are completely illiquid at the point of splitting between A and B, so that fire sales can be considered for all assets captured in A.
[33] Opposite opinions on the matter have been expressed by Aramonte and Zabai [6], looking in particular at the valuations of ESG funds, and by Jourde and Stalla-Bourdillon [66], comparing green to brown assets.
RSS - všechny zprávy
Vložit zprávy na www stránky
