Herds on Green Meadows – The Decarbonization of Institutional Portfolios

We analyze an emerging sustainable trend in asset management: the decarbonization of institutional portfolios. By using broad institutional ownership data we show that investors exhibit herding behavior in the sense of decarbonization. They are inclined to follow their own or other investors’ buys in green stocks and sales in brown stocks over adjacent quarters. In addition, we find that Hedge Funds as well as Investment Advisors lead the herd by executing trades in the sense of decarbonization. This is in line with expectations, since Investment Advisors are often bound to social norms, which nowadays include awareness for environmental aspects. For the aspired achievement of market-wide decarbonization, investors leading the herd should be encouraged to further decarbonize their portfolios in order to trigger follow-up trades.


Introduction
The combat of climate change has evolved as a global challenge for the entire society. We observe changing awareness and behavioral attitudes of society towards environmental-related issues. Besides, governments already take action or prepare to do so in order to mitigate the effects of climate change. In addition, the economy is in a state of transformation towards more environmental-friendly processes and activities.
The overarching political global target that represents climate change-induced action is the reduction of global warming to well below 2°C above pre-industrial levels. However, not only politics and economics are concerned with environmental changes and transformations, but also the financial market. It has to adapt to changes and take a key role in the achievement of sustainable targets. For example, the EU Action Plan on Financing Sustainable Growth of the European Commission expects the financial industry to be a part of the solution towards a greener economy (European Commission, 2018).
We analyze one of the emerging sustainable trends in asset management: the decarbonization of institutional portfolios. Our aim is to answer the question whether the decarbonization movement is observable in the financial market or whether engaging institutions are merely an exceptional phenomenon. For this purpose, we address different aspects of portfolio decarbonization using the largest institutional ownership dataset in literature so far and combine existing methodologies for herding measurement.
Since greenhouse gas emissions are one of the main causes of global warming, their reduction constitutes the main target function for politics and society to combat climate change and achieve the global 2°C target. In its traditional form, decarbonization describes the action to divest from carbon-intensive ("brown") assets and invest in low-carbon ("green") assets instead (PDC, 2015). In recent times, it encompasses far more than greenhouse gas emissions -the goal should be to align the portfolio with the climate economy of the future.
The decarbonization or divestment movement has already been initiated back in 2011, as students called for climate action in the management of university endowments. 1 Since then, commitments to divest have continued to grow rapidly, also among institutional investors, such as insurers, pension funds, and sovereign wealth funds.
The Portfolio Decarbonization Coalition (PDC) is a driving force to increase popularity of portfolio decarbonization. Its overall aim is to "drive greenhouse gas emissions reductions by mobilizing a critical mass of institutional investors committed to gradually decarbonizing their portfolios" (PDC, 2015). We base our research questions on the target setting of the PDC, since it encompasses all aspects of our research.
First, we associate the mobilization of a critical mass with herding behavior. We follow the definition of Sias (2004) who terms herding as institutional investors following each other into and out of the same securities over adjacent periods. Our first research question thus examines whether institutional investors in general demonstrate herding behavior. We show that they in fact do. To be more specific, investors tend to follow the trades of other investors rather than following their own. For this purpose, we use the most comprehensive ownership dataset so far applied in literature.
Second, we analyze whether herding behavior is also present in the context of decarbonization by developing a unique approach to unveil such a pattern. We define decarbonization herding as investors following their own or others' buy trades in green stocks and their own or others' sell trades in brown stocks, respectively. On the contrary, carbonization herding is defined as investors following their own or others' buy trades in brown stocks and sell trades in green stocks. Our results show that decarbonization herding exists and is of higher importance than carbonization herding (in the sense of a significant positive difference between decarbonization and carbonization herding).
Third, we examine which institutional investor groups follow or even promote this herding behavior. We expect that professionally managed funds, such as pension funds and mutual funds, are promoting the overall decarbonization herding, since they are often bound to social norms (Hong and Kacperzyck, 2009;Bolton and Kacperczyk, 2019). In fact, we find that Hedge Funds and Investments Advisors, which include Mutual Funds and Pension Funds, make up the largest part of herding and engage highly in decarbonization herding.
Electronic copy available at: https://ssrn.com/abstract=3461683 Institutional investors obviously exhibit decarbonization herding. In literature, there are various explanations on why herding should occur. 2 We briefly summarize the motives for herding and explain why investors might engage in decarbonization herding.
One of the simplest motives for herding is that investors follow trends. In recent times, we observe an increased awareness for sustainability in society as well as the call for financing more sustainable projects and stocks (e.g., the Fridays for future movement and the EU Action Plan on Financing Sustainable Growth). Besides, investors are dependent on retail flows. Since society strives for more sustainable and greener investments due to its increased sustainable awareness, retail flows into sustainable financial products automatically increase forcing investors to invest them in green products. Moreover, we observe informational cascades, i.e. investors do not make use of their own noisy signals, but infer information from previous investors' trades. Besides demand also the offer of sustainable products increases inducing institutional investors to follow the actions of others. This might also be attractive to investors due to reputational issues: if investors face a reputational loss when acting differently from the herd they are more inclined to follow others' trades. Increasing popularity of sustainability ratings for investment products and demand for disclosure of sustainability criteria in investment vehicles render this herding motive more than plausible. Furthermore, investors are often attracted to stocks with certain characteristics (characteristic herding). If investors choose to invest green they have a limited investment universe at hand, thus increasing the probability of herding. Finally, investigative herding suggests that investors follow correlated signals at different times, which might reflect the process of incorporating information into prices.
Climate change and respective policy action might involve risks for firm values (e.g., Dietz et al., 2016;Campiglio et al., 2018;Hong et al., 2019). These new risk sources have to be incorporated into market prices.
The remainder of the paper is structured as follows. Section 2 summarizes the related literature.
In Section 3, we explain the data. Section 4 presents the methodology used for the determination of overall herding in the financial market and the respective results, whereas Section 5 introduces our measures of (de)carbonization herding and shows our results. Section 6 gives an overview of additional robustness tests and Section 7 concludes. 5 2 Related literature This paper is related to three different strands of literature: herding, sustainability and divestment, and the intersection of institutional investment and sustainability. Lakonishok et al. (1992) develop a simultaneous herding measure on stock level. Their results point to the assumption that money managers do not herd at a large and meaningful way, but rather follow a variety of trading strategies. Grinblatt et al. (1995) and Wermers (1999) study herding behavior of mutual funds. In their baseline tests, they use the herding measure of Lakonishok et al. (1992), however, they do not find high levels of herding in mutual funds. Sias (2004) measures herding as the cross-sectional correlation between institutional demand of subsequent quarters instead of testing cross-sectional temporal dependence of trades within a period as in Lakonishok et al. (1992). The author finds a positive correlation between institutional demand over adjacent quarters. Furthermore, he shows that herding is attributable both to investors following their own last-quarter's trades and others' last-quarter's trades. Sias (2009) andCeliker et al. (2015) use the Lakonishok et al. (1992) and the Sias (2004) methodology to answer the question whether institutional investors and mutual funds, respectively, follow each other into and out of the same industry. Both papers find strong evidence for industry herding. Popescu and Xu (2018) extend the methodology of Sias (2004) and compute measures for buy-leading and sell-leading of mutual funds. Lastly, Jiang and Verardo (2018) develop a herding measure on fund level that controls for investment styles and institutional preferences.
The literature covering sustainability takes on various facets. The literature on divestment, however, is still scarce. Trinks et al. (2017) compare the performance of investment portfolios with and without fossil fuel firms. They find that fossil-free portfolios do not significantly underperform the market portfolio, whereas portfolios with fossil fuel stocks show higher mean returns which might indicate that a risk premium for this kind of stocks exists. Davies and Van Wesep (2018) show in a quantification exercise of their economic model that executives benefit on average from hypothetical divestment campaigns. Choi et al. (forthcoming) demonstrate that carbon-intensive firms perform worse than low-carbon firms in abnormally warm weather. In their survey of institutional investors, Krueger et al. (forthcoming) find out that divestment is the least frequently used approach to manage climate risks.
Lastly, we present literature on the intersection of institutional ownership and sustainability considerations. Graves and Waddock (1994) find that institutional ownership measured by the Electronic copy available at: https://ssrn.com/abstract=3461683 number of institutions owning shares increases when corporate social performance improves.
However, the effect loses its significance when institutional ownership is measured by the percentage of shares owned by institutions. Thus, the authors conclude that there is no penalty in institutional ownership if corporate social performance is increased. Hong and Kacperzyk (2009) focus on so-called "sin stocks" and show that sin stocks have a significantly lower ownership ratio. Likewise, Chava (2014) comes to the conclusion that firms with environmental concerns exhibit lower institutional ownership compared to firms without such concerns. In turn, Bolton and Kacperczyk (2019) find out that insurance companies, investment advisors, and pension funds own less high-emission firms. Since these institutions are more constrained by social norms, they are more likely to shun high-carbon firms. Besides, Dyck et al. (2019) analyze the impact of institutional investors on the environmental and social performance of a firm. They show that institutional ownership positively impacts environmental and social performance.
With this paper, we contribute to the herding literature by utilizing the most comprehensive ownership dataset in literature and reinforcing previous results. Additionally, we focus on herding in the sense of decarbonization. For this purpose, we combine three research areasherding, institutional investors, and sustainability -to derive important insights into the sustainability trend in asset management. With our unique approach of measuring (de)carbonization herding, we answer the research question whether institutional investors tend to follow their own or others' trades that point to the decarbonization of portfolios. Our results confirm that herding in the financial market is driven by decarbonization trades. Thus, we demonstrate that institutional investors indeed tend to follow the decarbonization movement in the financial market and are inclined to pursue sustainable trends in asset management.

Sample construction and summary statistics
We obtain data from different databases. The ownership data is sourced from the Refinitiv Ownership and Profiles database. 3 This is the most complete global shares database providing insight into the ownership structure of over 70,000 publicly traded equities in over 70 markets.
The data is sourced from stock exchanges, regulatory bodies, institutions, and different financial reports, e.g. 13F filings. It covers different types of investors, e.g. Investment Advisors, Governmental Agencies, Foundations, and Individual Investors from all over the world. The database reports for each quarter the shares held and the respective market value of the investors' positions at the respective firms as well as the percentage held.
We run several filters to fit the ownership data to our needs. In order to prove herding behavior for investors, we need information on the trading activity of individual investors. We define an investor's trading activity as purchasing a stock if the percentage held increased over the quarter. In the opposite case, the investor's activity is identified as selling the stock. However, the ownership data obtained from Refinitiv contains the ownership structure of stocks and not trading data, which means that we obtain information on which shareholders hold the particular stock in their portfolio and to what extent. To fully determine the investor's trading activity, we complement our dataset to control for interim and final sales. We exclude all investors if they do not trade at least one stock over the observation period and we exclude investors with less than eight quarters of ownership data.
We categorize stocks as brown and green using ESG data of the Refinitiv ESG 4 database. With information for over 7,000 firms, this database covers the most important stocks traded on global stock markets and therefore serves as a good proxy for the worldwide investment universe. Refinitiv uses over 400 ESG metrics and assesses firms on the basis of their sustainable use of resources, emissions, environmental production processes, human rights, and management structures. It is currently perceived as one of the best available sources for firms' environmental, social, and governance information. Due to the fact that we want to analyze (de)carbonization herding, we focus only on one of the three pillars, namely the environmental pillar score (E-score). Hereby, firms are ranked according to their resource usage, emission reduction efforts, and their degree of environmental innovation.
We do not solely focus on carbon emissions to identify brown and green stocks for many reasons. First of all, data availability on carbon emissions is rather scarce. Furthermore, investors choose different strategies to actually decarbonize their portfolios (e.g., exclusionary criteria, carbon footprints, and engagement; PDC, 2017). This makes it difficult to measure decarbonization herding based on one specific stock criterion. Besides, due to recent developments, financial market participants demand forward-looking assessments for sustainability. PDC (2017) emphasizes that investors have to systematically determine and consider climate-related risks and opportunities. Hence, "portfolio decarbonization refers to systematic efforts by investors to align their investment portfolios with the goals of a lowcarbon economy" (PDC, 2017). This subsumes not only the reduction of the carbon footprint, but also, e.g., an increased investment in renewable energies and capital withdrawal from high energy consumption activities (PDC, 2017). For this reason, we choose to label firms as brown or green based on the environmental pillar score, since it comprises a more forward-looking assessment of sustainability and a wider scope than just carbon emissions. 5 [Insert Table 1 about here] Table 1 reports the descriptive statistics for our sample. From January 2002 to September 2017 we observe 5,669 distinct firms, for which we were able to obtain ownership data on a quarterly basis and for which ESG data is available. The number of firms in this study increases each year until it reaches its peak in 2016 with 5,064 stocks. Every year, we calculate the median E-Score of all firms and identify firms with an E-Score greater (smaller) than the median as green (brown), which leads to half of the stocks being classified as green and brown, respectively, each year. Each quarter, we identify an investor as a buyer if her ownership in the respective stock increases compared to the last quarter and as a seller if it decreases. The number of identified trades increases nearly sevenfold from 2002 to 2016. 6 The most trades can be observed in 2016 with around 4.5 million trades. Over our sample period, we identify slightly more buys than sales with an average buy to overall trades ratio of around 53%, which is in line with existing literature (Wermers, 1999;Choi and Sias, 2009).
Due to the fact that we distinguish between green and brown stocks, we also consider the respective buy and sell ratios of these two stock groups. The buy and sell ratio of brown stocks are on average higher than the buy and sell ratio of green stocks. The decarbonization ratio (sum of the buy ratio of green stocks and the sell ratio of brown stocks) and the carbonization ratio (sum of the buy ratio of brown stocks and the sell ratio of green stocks) do not exhibit large differences with the decarbonization ratio being slightly higher on average.
[Insert Table 2 about here] 5 According to Refinitiv, the environmental score "reflects how well a company uses best management practices to avoid environmental risks and capitalize on environmental opportunities" (Thomson Reuters, 2015). The score thus is appropriate for our needs to identify stocks for portfolio decarbonization. 6 The number of trades in 2017 decreases due to the fact that we obtain ownership data only until September 2017. We are therefore missing one quarter in 2017. Table 2 contains statistics about the investor types. For the 5,669 firms, we obtain ownership information for over 137,976 investors and we classify each of them into one of eight investor types. 7 To the best of our knowledge, the ownership dataset in this study is, so far, the largest one used to study herding behavior. Just like the number of stocks, the number of investors

Herding in the financial market
At first, we analyze in how far the financial market exhibits herding behavior. For this purpose, we combine the methodologies of Sias (2004) and Popescu and Xu (2018). In this way, we do not only connect different measurement methods to derive more detailed insights into herding behavior, but also do so by using the largest institutional ownership dataset so far in the literature.
For each stock in each quarter, we calculate the buy ratio, defined as the number of all purchases in that stock divided by all trades in that specific stock.
As seen in Table 1, the average buy ratio over all stocks over the whole sample period is approximately 53%.
Following Sias (2004), we further calculate the standardized buy ratio of stock i as where is the cross-sectional average of the buy ratio in quarter t over all stocks i and is the cross-sectional standard deviation in quarter t over all stocks i. (2004), we define herding as investors following themselves or each other into or out of the same stock over consecutive quarters. In order to prove herding behavior, the crosssectional correlation of the institutional investors' demand for a share between two consecutive quarters is calculated. This cross-sectional correlation is equivalent to the coefficient from the regression of the standardized institutional demand for share i in quarter t + 1 on the standardized institutional demand for share i in quarter t.

As in Sias
Due to the standardized factors, the constant is zero and the coefficient simplifies to Herding can derive from different sources. Therefore, Sias (2004) decomposes the beta coefficient into two sources of herding: investors following their own trades, i.e. they trade the same stock in the same direction in the following period, and other investors following the trade of the investor, i.e. in the following period, another investor trades the same stock in the same direction.
∆ , , ∆ , of the correlation that results from investors following their own trades in consecutive quarters, i.e. buying (selling) a share in t and buying (selling) it again in t + 1. We expect this term to be positive if investor trading is persistent over time, i.e. investors trade the same stock in the same direction a quarter later. If investors tend to avoid executing the same trades as before, the term becomes negative. If there is no systematic correlation between the trading activities of investors in two consecutive quarters, the term becomes zero. The second part of the crosssectional correlation in Equation (5) results from other investors following the trades of the respective investor. The same expectations regarding the direction of the correlation apply here.
This method therefore helps to answer the question whether investors follow themselves or others when trading the same stocks in the following quarter.  (2004), Sias (2009), andCeliker et al. (2015). This illustrates clearly that on average the financial market shows herding behavior.
In a next step, we decompose the herding measure into different sources of herding. One arises from investors following their own trades, which we call self-herding, and the other from other investors following the trades of the respective investor, which we refer to as following-herding.
Both components make a significant contribution to the average herding measure. Around 5% of average herding results from self-herding, which is in line with the hypothesis that institutional investors split bigger investments into several trades over some period to minimize their price impact (Chakravarty, 2001). However, as seen in Panel A, most of the herding, nearly 95%, derives from following-herding. Choi and Sias (2009) have already shown this pattern when taking aggregated stock industries into account rather than individual stocks. A slightly increased herding measure and the same pattern also occur when only stocks traded by more than 50 or 100 investors are considered.
[Insert Table 3 Equation (6) is a more detailed version of Equation (5), whereby a distinction is made between herding resulting from a purchase or a sale by assigning a strict value to , , , , , i.e. 1 for purchase and 0 for sale for the trade of investor n (m) of stock i in quarter t. This decomposition enables us to show whether investors mainly exhibit self-or following-herding when the triggering trade is a buy or a sale.
In Panel B of Table 3, the sum of self-herding resulting from buys (0.0032) and sales (0.0129) has to be equal to the general self-herding measure (0.0161) from Panel A, whereas the same applies to following-herding. Furthermore, the sum of all four components corresponds to the general average herding coefficient of 0.3203. The results show that herding is triggered by both buys and sales. In unreported analyses, we show that the differences between the buy and sell herding for self-herding as well as for following-herding are statistically significant. More important, all four herding sources have a significant share on the overall herding measure. The largest portion of herding is triggered by the sale of a stock (54.69%), whereby for the most part (50.66%), this sale functioned as a signal for other investors to sell this stock as well in the following quarter. The results are consistent with existing studies and are robust when considering only stocks that are traded by more than 50 or 100 investors.

(De)carbonization Herding
In the last section, we have shown that the financial market proxied by our large dataset exhibits herding behavior. This section develops the methodology derived before further to break down the herding measure in (de)carbonization herding. Henceforth, we distinguish between self-herding and following-herding induced by purchases or sales of green and brown stocks, i.e. we differentiate whether the triggering buy or sale in Electronic copy available at: https://ssrn.com/abstract=3461683 quarter t takes place in a green or a brown stock. For this purpose, we extend Equation (6) once again. 8 To answer our main research question, we differentiate between herding in the sense of decarbonization and carbonization.
We define decarbonization herding as the behavior of investors to buy a green stock which was recently bought by the investor herself or some other investor, as well as to sell a brown stock which was recently sold by the investor herself or some other investor. Thus, we capture whether triggering trades in the spirit of decarbonization cause the same investor or other investors to decarbonize their portfolios as well in the subsequent quarter and thus induce herding behavior.
The (de)carbonization herding measures are computed as follows. -ℎ , , , , where , and , indicate the number of stocks identified as green or brown.
The decarbonization following-herding equals the sum of the following-herding resulting from the buy of a green stock and the following-herding resulting from the sale of a brown stock. In contrast, herding in the sense of carbonization would be the exact opposite, i.e. the portion of the herding measure resulting from the same investor or other investors increasing their shares in a brown stock by following the respective investor's buy of a brown stock and decreasing their shares in a green stock following the sale of the green stock.
, , where , and , indicate the number of stocks identified as green or brown.
As shown before, we observe significant herding behavior in the financial market. If there was no preference for neither green nor brown stocks when following own trades or trades of other investors, the four carbonization and decarbonization herding measures introduced above should have approximately the same value. We show that this is not the case.
[Insert Table 4 about here] To get a better understanding about herding in our sample, Figure 1 plots the decarbonization and carbonization herding measures for self-and following-herding. As already seen in Table   4, self-herding for decarbonization as well as carbonization is rather small and on average close to zero over time. In contrast, following-herding is much more volatile and reveals some interesting interrelations over time. Even though the average carbonization following-herding is close to 0 (0.0040), it is negative most of the time from 2006 to 2012. As defined before, carbonization herding covers following trades on sales of green stocks and purchases of brown stocks. Thus, a negative carbonization herding coefficient relates to investors who tend to buy green stocks that others have previously sold and sell brown stocks that other investors have previously bought. Therefore, the following investors decarbonize their portfolios. Taking this into account, the actual decarbonization effect is even greater than initially assumed. Until 2006, the differences between decarbonization and carbonization herding are not as clear cut as in the following period. In 2006, the decarbonization following-herding leaps from -13% to +30% and the decarbonization and carbonization following-herding diverge strongly from this moment on. We interpret this as a movement of investors who pay more attention to whether other investors decarbonize their portfolios and then follow this strategy. Around the years 2010 and 2011, we can see the highest decarbonization following-herding with over 60% and the most negative carbonization following-herding around -26%.
[Insert Figure 1 about here] So far, we have proven that there is a general herding behavior in the financial market and moreover that investors pay attention to whether other investors decarbonize their portfolios.
With our next analysis, we try to find out which investor type leads this decarbonization movement, which investor type carries out the buys and sales, which follows itself, and which are followed in the consecutive quarter by other investors. We therefore calculate the proportion of the overall herding attributable to one specific investor type. We then calculate for each investor type the share of decarbonization and carbonization herding as we have shown in Table   4 for the entire sample of investors. The sum of all individual parts across investor types must be the same as the overall herding measure from the beginning (0.3203).
[Insert Table 5 about here] For each type, we can confirm that the proportion of decarbonization herding is significantly higher than the one of carbonization herding. The only exception are Foundations, where we only find an insignificant but still positive difference. The difference is mostly driven by investors following other investors rather than themselves. Some substantial part of overall herding is driven by the trades of Individual Investors (5.10%) and Banks and Trusts (4.65%).
The trades of Investment Advisors and Hedge Funds trigger the major part of overall herding in the financial market (53.08% / 34.43%), which is not surprising due to the high number of trades executed by these two investor types. 51.98% of overall herding is driven by triggering trades of Investment Advisors in the spirit of decarbonization. This means, if an Investment Advisor buys a green stock or sells a brown stock, it is more likely that another investor or the Advisor herself will follow her than if the Advisor traded exactly the opposite way in the first place.
To further test which investor type demonstrates the most decarbonization herding, we redo the analyses above, but now we consider the aggregated investor types rather than the underlying individual investors. Therefore, we consider every investor type as a single investment portfolio. By doing this, we account for the different number of trades of the individual investor types.
[Insert Table 6 about here] Table 6 Panel A shows a pattern similar to Table 3 Panel B, where we can see the overall herding measure as well as the differentiation between herding induced by buys or sales. Due to the aggregation on investor type level, the overall level of herding is lower (0.0636) compared to the previous herding measure on investor level (0.3203). Besides, the proportion of self-herding is much larger than before. 32.73% of herding results from an investor type selling the same stock over consecutive quarters, therefore splitting its sales over consecutive quarters. This can be regarded as intra-group herding (herding within an investor type). The overall following-herding now accounts for only 44.89%, whereas on investor level about 94.98% of the overall herding measure is determined by following-herding. Investor type selfherding accounts for 55.11%, thus being more important for overall herding than followingherding. This leads to the conclusion that a lot of investors trade the same stocks the same direction over consecutive quarters as investors of the same investor type, e.g. Investment Advisors tend to follow the trades of other Investment Advisors.
Panel B is similar to Table 5 and shows the breakdown of the investor type level herding in terms of decarbonization and carbonization. Similar to before, for each type, except for Governments, a significant larger proportion of overall herding results from decarbonization trades. Hedge Funds (36.62%) and Investment Advisors (33.03%) still account for the largest part of overall herding, however, the share of Investment Advisors is lower than before (33.03.% vs. 53.08%). In turn, the percentage of the other investor types, especially the proportions of Banks and Trusts as well as Individual Investors, trigger more follow-up trades compared to the investor level measure in Table 5. When only taking self-herding into account, we find that Individual Investors as well as Private Equity firms are more likely to follow trades in the sense of carbonization rather than decarbonization. For following-herding, we find the same pattern as before, with herding in the sense of decarbonization being more pronounced.
Overall, Investment Advisors thus seem to promote decarbonization herding besides Hedge Funds. This result confirms the hypothesis that Investment Advisors are often bound to social norms and have to meet certain investment goals. Additionally, they might be prone to reputational herding (e.g., Scharfstein and Stein, 1990;Dasgupta et al., 2011).

Robustness tests
We conduct several additional analyses to test the robustness of our results. First, we divide stocks in green and brown based on a static selection process. A stock is labeled as green if its average environmental score is above the median of all scores in the sample, and brown otherwise. Since we do not require stocks to have a complete time series of scores available, we assign the characteristic of greenness (brownness) to each stock from its existence on. Thus, we are able to extend our data sample to a time period from 2000 to 2017. All of our results remain economically the same. However, this procedure might lead to a look-ahead bias of the sustainability characteristic of a stock.
Secondly, we substitute the environmental score with the overall ESG score to account for other aspects of sustainability. The results remain unchanged, even when including the social and governance dimension of sustainability. This mirrors the fact that the environmental score is highly correlated with the overall ESG score.
Last, we randomly assign stocks several times into the green and brown sample, respectively, regardless of their actual E-score. As expected, we do not find any statistically significant differences between decarbonization and carbonization herding. This verifies that the allocation of stocks based on the E-score does not follow a random criterion but is meaningful for the representation of (de)carbonization.

Conclusion
We analyze the presence of portfolio decarbonization in the financial market. For this purpose, we use the most comprehensive institutional ownership dataset in literature and combine herding measures from Sias (2004) and Popescu and Xu (2018). In the financial market, institutional investors tend to follow others rather than following their own trades regardless whether the triggering trade is a buy or a sale. In addition, we find that institutional investors engage in decarbonization herding. This means they follow their own or other investors' buy trades in green stocks and sell trades in brown stocks, respectively. It is noticeable that decarbonization herding is majorly triggered by following-herding, i.e. investors follow others in portfolio decarbonization. Across all investor types, decarbonization herding is greater and significantly different from carbonization herding. In specific, we show that especially Investment Advisors and Hedge Funds promote decarbonization herding.
Overall, institutional investors seem to be prone to the divestment movement in the sense that they are taking part in it. However, the motives for their herding behavior are still to be analyzed in future research. We give first indications on what might be the reasons for their decarbonization herding, but leave empirical analyses to future research. Furthermore, the impact of decarbonization herding on stock prices of green and brown stocks, respectively, has yet to be determined.
We rather focus on an as-is description of the occurring phenomenon of portfolio decarbonization. In conclusion, for the aspired achievement of market-wide decarbonization, investors leading the herd should be encouraged to further decarbonize their portfolios in order to trigger follow-up trades by others. Our results point to the fact that climate change does not only become critically important in the financial market, but already has arrived in asset management.  This table shows key statistics for the respective year of the dataset used. The dataset contains all stocks for which Refinitiv ESG data and the respective ownership structure data is available in the period from January 2002 to September 2017. It provides counts of stocks, green stocks, and trades, as well as the overall buy (sell) ratio and the buy (sell) ratio for green and brown stocks, respectively. The buy (sell) ratio is calculated as the ratio of the number of buys (sales) to the total number of trades. This table shows tests for herding measures for securities with equal or more than 1, 50, and 100 traders, respectively, for the sample period. Panel A calculates the overall average herding measure as well as the self-and following-herding using the methodology described in Sias (2004). Panel B additionally decomposes herding in buy and sell herding in the sense of Popescu and Xu (2018). The columns labeled as "%" indicate the share of the respective herding source on the average herding measure. Statistical significance is measured by two-sided t-tests and *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively. This table shows the decarbonization and carbonization herding measures as defined in the text for the sample period. The column labeled as "%" indicates the share of the (de)carbonization measure on the average herding measure in the financial market. Statistical significance is measured by two-sided t-tests and *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively. Difference 0.0008*** 0.0000 0.0008***

Figures and Tables
The table shows the average (de)carbonization herding measures per investor type in the sample period. The calculation of the herding measures follows the description in the text. Investors are grouped into investor types as described in the text. For comparison purposes, the overall average herding measure is shown as well. The column labeled as "%" indicates the share of the (de)carbonization herding on the average herding meausre. The rows denoted as "Difference" are the differences between the decarbonization and the carbonization measure per investor type. Statistical significance is measured by two-sided t-tests and *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively.  Sias (2004) and decomposes it in buy and sell herding following Popescu and Xu (2018). Panel B shows the (de)carbonization herding measures per investor type. The rows denoted as "Difference" are the differences between the decarbonization and carbonization herding measure per investor type. Columns labeled as "%" indicate the share of the respective herding measure on the overall average herding measure. Statistical significance is based on two-sided t-tests and *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively.