---

Abstract: An overview

Introduction

This first substory aims to show the general information we were able to extract from the SEC dataset.

Used Dataset: SEC asset manager portfolios

Research Question

What asset managers have the portfolios with the greatest/least social impact?

Results

General Overview and Cleaning

In the histogram above, we observe the frequency at which US dollars (in millions) are invested per investor. Investors in this context, as in the rest of this data story, are large asset manager investors with holdings of at least $100M. We can see an overwhelming preference by these investors to invest under one million dollars. The frequency of investments over one million dollars is difficult to see, so we remove outliers before continuing to explore this data in the graphs below.

In the boxplot above, we observe the distribution of US dollars invested per investor, after the removal of outliers. We can see that the median investment is still under one million dollars; however, we can see that the interquartile range of the boxplot ranges from around half a million to three million dollars. This means that investments of less than one million dollars are as common as those from half a million to three million dollars. Thus, by removing outliers we can see there is still a significant number of investments being made using amounts of half a million to three million dollars.

In the histogram above, we observe the frequency at which US dollars (in millions) are invested per investor, with outliers removed. Unlike in the previous histogram above, we can more clearly see the frequency at which amounts of US dollars greater than one million dollars are invested. we see that as the amount in US dollars increases, the frequency at which those amounts are invested tends to decrease. The frequency of US dolar amounts invested goes up slightly at five million and six to seven million dollars, but does not surpass the frequency of US dollar amounts invested at three million dollars.

In the boxplot above, we can see the distribution of Just Capital company social impact scores. We see, in the overall social impact score, that most companies have a score of between 40 and 60, out of 100 points, with the median being 50 points. Observe that outliers in the overall company social impact score exist above 60 points, and that conponent social impact scores (customer, environment, jobs, etc social impact component scores that feed into the overall social impact score) also have medians of around 50 points.

In the histogram above, we can see the frequncy of weighted portfolio-wide social impact scores per investor.

The weighted portfolio social score per investor is created by aggregating the individual scores of investments held by an investor on a weighted average basis according to the portfolio percentage of each investment. For example, an investor with a portfolio of stock A and stock B, where the investor’s portfolio is 40% stock A and 60% stock B, and stock A and B have social impact scores of 50 and 60 would have a weighted portfolio social impact score of 56 (40% x 50 + 60% x 60 = 56).

The histogram shows that most investor portfolios have a social impact score of 14 points. The frequency of investor portfolios with scores of more than 14 points drops significantly (by almost half at 15 points), and continues to drop in a linear fashion until 45 points. Thus, large asset management investors have predominantly low portfolio social impact scores.

2017 Investor Portfolios by Social Impact

We can observe in the 2-D Scatterplot above, focused on 2017 data, the variation in Investor Portfolio social impact amoung specific large ($100M+) asset managers. We can hover our mouse over specific data points to identify specific investors and their scores. We see from the distribution of portfolio social impact scores that most investors have very low portfolio social impact scores (around 14 points). This is followed by smaller groups with scores higher than 20 points. This suggests that there is no clear consensus amoung the asset management community in the need for them to implement social impact maximizing or minimizing portfolio investment strategies.

Discussion

We see that, for example, the lowest scoring investor, Akanthos Capital Management, is focused on a very specialized investment strategy (event-driven investing), which operates independently of long-term social trends such as societal pressures to be more socially impactful.

On the other end of the spectrum, the highest scoring investor, Aetna Inc, is a healthcare company, and is thus intrensically driven to invest based on values aligned with social impact outcomes (i.e., improvements to medical technology, social services, etc).

Conclusion

After examining the specific organizations behind both the most and least socially impactful investors, on a portfolio-wide basis, we can see that the extent of their social impact can be largely explained by investor-specific attributes, such as specific profit-maximizing investment strategies, and organizational values. This suggests that the lack of a clear consensus amoung asset managers towards the need to make explicitly social-impact oriented decisions comes from a lack of motivation to make such decisions. Given the lack of transparency public stakeholders have into the investment activities and social impact of asset managers, it is not surprising for us to see that asset managers are not motivated as a whole to consider social impact explicitly in their investment strategies, even as public stakeholder support of social impact investments grows.

• Client: EPFL - ADA Course
• Date: December 2018
• Service: Assets

---

Abstract: Financial asset is a non-physical asset whose value is derived from a contractual claim, such as bank deposits, bonds, and stocks. Financial assets are usually more liquid than other tangible assets, such as commodities or real estate, and may be traded on financial markets

Introduction & Definitions

This data substory aims to understand the types of investments asset management investors favor, and to examine the motivations behind asset manager’s investments into the best and worst social impact investments.

The Ethical Puzzle

Although the definition of ethical investing could be stated as simply purchasing investments from issuers who act ethically, investors who seek to achieve this must first establish a set of criteria that can be used to determine what actions and practices are considered ethical to them. For example, the Bible and other religious books provide rules and principles for people to live by, and those who believe what those books say use them as guides to evaluate things in the world around them. Religious texts and other moral teachings, however, do not always provide crystal-clear guidance for practical investment decisions. Although creating or finding the ethical criteria to use as the basis for evaluating investments may often be relatively simple, effectively applying that criteria to actual investment choices can be more difficult. For instance, the Bible condemns drunkenness in several passages, but does this then make it unconditionally wrong to purchase securities issued by makers of alcoholic beverages? Since not all consumers of alcohol get drunk, then the investor will have to decide for him or herself whether this investment truly supports immoral behavior. Virtually all issuers of securities or investments most likely do something that at least a substantial percentage of their investors do not approve of (how many investors who buy Treasury securities approve of everything that is done by the government?). Of course, this begs the question of which practices and policies are tolerable for investors and which are not.

Assets

According to the International Financial Reporting Standards (IFRS), a financial asset can be:

• Cash or cash equivalent,
• Equity instruments of another entity,
• Contractual right to receive cash or another financial asset from another entity or to exchange financial assets or financial liabilities with another entity under conditions that are potentially favourable to the entity,
• A contract that will or may be settled in the entity’s own equity instruments and is either a non-derivative for which the entity is or may be obliged to receive a variable number of the entity’s own equity instruments, or a derivative that will or may be settled other than by exchange of a fixed amount of cash or another financial asset for a fixed number of the entity’s own equity instruments.

Used Dataset: SEC asset manager portfolios

Research Questions

• What are the most popular “ethical” or “unethical” investments held by asset managers?

Results

Popularity of High Social Impact Stocks

The heatmap shows the most and least popular “ethical” investments held by asset managers. It shows the number of investors in 2017 for each stock that scored within the top 10% of Just Capital Social Impact stocks, where stocks scoring in the top 10% of Just Capital Social Impact scores are labeled “ethical”. Said slightly differently, the heatmap displays the social impact (scored by Just Capital) of each stock within the top 10% of Just Capital Social Impact Scores, as well as the number of large institutional investors (with at least 100M in US 13F type securities) that own that stock. More popular stocks are represented by lighter colours, whereas less popular stocks are represented with darker colours. Although the top 10% of Just Capital Social Impact Scores corresponds to 87 stocks, only 43 stocks are shown in this heatmap. This is because half of these stocks are not held by any large asset manager investor.

Popularity of Low Social Impact Stocks

The heatmap follows an identical format to the previous heatmap, and shows the most and least popular “unethical” investments held by asset managers. These stocks correspond to bottom 10% of Just Capital Social Impact Scores.

Discussion

High Impact Stock Popularity Observations

We can see that technology companies, as an industry category, are the most popular “ethical” investments held by asset managers. A large number of stocks shown in the Popularity of High Social Impact Stocks plot represent technology companies, including Apple (APPL) with a social impact score of 59.97005 and 1181 large asset manager investors, and Microsoft (MSFT) with a social impact score of 68.55428 and 948 large asset manager investors.

Another popular type of high impact stock are Blue Chip stocks (stocks with a history of reliable returns that are highly trusted by the larger investment industry) such as Johnson & Johnson (JNJ) with a social impact score of 59.94658 and 833 large asset manager investors, and Nike (NKE) also appear.

We also notice that there is an uneven distribution of stocks within the highest social impact stocks. Although the top 10% of social impact stocks have social impact scores ranging from 53 to 70, few stocks exist in the upper half of this range, with scores higher than 63 points.

Low Impact Stock Popularity Observations

We can see that companies employing large, less educated labour forces are the most popular “unethical” investments held by asset managers. As shown in the Popularity of Low Social Impact Stocks plot, these include a large number of manufacturing and retail companies, including Tesla (TSLA) with a score of 45.05744 and 66 large asset manager investors, Goodyear Tire and Rubber Company (GT) with a score of 45.96831 and 15 large asset manager investors, and TJX Companies (TJX) with a score of 44.58723 and 67 large asset manager investors.

Similar to the Popularity of High Social Impact Stocks plot, Blue Chip stocks once again appear. These include stocks such as Sourthern Co (SO) with a social impact score of 44.58723 and 32 large asset manager investors, and Stryker Corporation (SYK).

We notice that large asset manager investors in general hold far fewer stocks in this category, with each stock individual stock having lower numbers of investors. There are also less stocks displayed in the Popularity of Low Social Impact Stocks plot compared to the Popularity of High Social Impact Stocks plot because many stocks (80%) in the bottom 10% of Just Capital Social Impact Scores are not held by any large asset manager investors.

Conclusion

Although the values and beliefs that guide ethically minded investors differ somewhat for each person, there are several specific areas where ethics play a major role. According to our results the most popular “ethical” investments are technology companies, with the most popular “unethical” investments being companies that employ large, uneducated labour forces such as manufacturing and retail companies. Blue Chip stocks also appear widely in both high and low impact stock lists. We observe, as an overall pattern, that large asset manager investors hold greater numbers of high impact stocks. However, we acknowledge that the technology companies that dominate the rankings of high social impact stocks have also dominated the financial markets in terms of returns on investment. Thus, an investor could easily end up selecting a portfolio with more high social impact investments simply based on looking for stocks with high historical financial growth - technology companies such as Amazon and Apple.

• Client: EPFL - ADA Course
• Date: December 2018
• Service: Assets

---

Abstract: Social good project benefit

Introduction

This data substory aims to examine if there exists a trend towards social impact investing between 2013 and 2017.

Used Dataset: SEC asset manager portfolios

Research Question

• Was there a transition toward investing more into social good then into other projects?

Results

Portfolio Social Impact by Investor and Year

The 3-D scatterplot shows the portfolio-wide social impact of institutional investors with over $100M in holdings in a sample of three years between 2013 and 2018 (2013, 2015, and 2017). Portfolio-wide social impact is measured using a score computed as a weighted average of a portfolio’s individual stocks’ social impact. The score is computed using the portfolio weight percentage of each stock declared in SEC 13F filings, and individual stock social impact scores computed by Just Capital.

We can see in this visualization the lack changes in portfolio social impact for the asset management industry as a whole (while some individual investors may have changes in their portfolio social impact, investors as an aggregate still result in roughly the same levels of social impact year over year.

Investor Social Impact in 2013 vs 2017

The Scatterplot displays how the social impact of investor portfolios has changed from 2013 to 2017. In particular, this visualization allows us to see how specific investor portfolio social impact levels have changed year to year by configuring the 5 layers of data traces provided.

The graph has 5 layers/traces that can be optionally displayed or removed from the plot by clicking the respective item in the Legend on the top right hand corner of the plot.

Using these different layers, you can see how the Social Impact of the best and worst Social Impact Investors has changed, or not changed between 2013 and 2017.

For example, if you choose to display the “2017 Investors by Social Impact Scores” in combination with the “Lowest Scoring 2017 Investors in 2013”, you can compare how investors have changed their portfolios in terms of social impact between 2013 and 2017.

Discussion

We can see that investors do not seem to be, as a general trend, shifting towards higher social impact investments. There are very few investors have relatively high social impact scores, with the majority of investors having relatively much lower scores.

We observe that the highest portfolio social impact scores correspond to Mirae Asset Global Investments (2013), a $400B Korean Financial Services Group, Rehmann Capital Advisory Group (2015), a Wealth Management Advisory (for high networth individuals), and AETNA INC /PA/ (2017), a Healthcare Company.

We also observe that the lowest portfolio social impact score (all scores of 0) come from Index Management Solutions, LLC (2013) and Water Asset Management, LCC (2015) - Land and Buildings Investment Management, LCC (2017).

Notice that the highest and lowest scoring investors are all relatively small, LLC, specialized investors focused on very specific verticals such as Water Storage, Real Estate, and investment services. It it likely that larger investors hold more broadly diversified portfolios that reduce their overall correlation to social impact score variables.

Conclusion

It seems like while a small minority of investors have adopted specialized strategies towards increasing social impact, most have not decided that predominantly social impact oriented portfolios related to successful investment returns. There does not seem to be an overall transition towards social impact investing. Although we can definitely see that the the largest, most successful investors are not among the worst at investing for social impact, this may be due to for example owning larger numbers of successful technology company stocks such as Apple and Microsoft that also tend to have produced incredibly high financial returns, instead of deliberate efforts to adopt investment portfolios that are more socially impactful.

We also conclude that large investors have greater pressures to diversify their portfolios of investments (not owning all the same types of stocks) due to the need to reduce investment risk (likelihood of all stocks in a portfolio simultaneously dropping in value). Thus, it is understandable that it is rare for the largest asset managers to have portfolios with the highest or lowest social impact scores. This logic also explains why it is difficult for the largest asset managers, who need to keep their portfolios sufficiantly diversified, to make significant year over year changes to their portfolio-wide social impact.

• Client: EPFL ADA
• Date: December 2018
• Service: History

---

Abstract: During this section, you will get insight on how it was possible to link tweets to sentiments and opinions about companies through sentiment analysis.

Introduction

Public opinion is a important criteria to choose a investement. In fact, each investement will be a part of the identity of the inve stor and public opinion could really sustain some investement and criticize others. It is thus interesting to correlate the greater public’s interest into social good projects expressed on social media platforms such as Twitter and Facebook, or news platforms, with the actual investments that are done. To add the public opinion part, sentiment analysis was done on tweet from 2017. In fact, sentiment analysis is contextual mining of text which identifies and extracts subjective information in source material, and helping a business to understand the social sentiment of their brand, product or service while monitoring online conversations. In this part of the data story, the text classifier process tool was used to analyses an incoming message and tells whether the underlying sentiment is positive, negative our neutral.

Used Dataset: The raw dataset was composed of all the tweets from 2017. First of all, output of the last part where taken and mix with a new dataset which is the Twitter dataset given in the course. It contains the tweets of 2017 and it’s store on the ADA cluster. The documentation of Twitter API was analyzed to understand how to use this database. The group decided to work on a selected part of the entire dataset. In fact, our goal is to score the public’s opinion of compagnies thanks to a sentiment analysis done over tweets of 2017. To do that we decide to take just tweet corresponding to 3 days per month. English tweets were filter as our analysis is made on Americans compagnies. We thus care more about the public opinion of English speaking population as they are more impacted by theses compagnies. For each month, we take always the same days : the 1th, the 11th one and the 21th. Then a preprocessing part corresponding to the cleaning of the tweets and the creation of dictionaries was done. Moreover, we link each tweets to compagny. To do that a easy rule was used : a tweet concerned a company if the name of the company is present in the tweet text or in the tweet tags.

Research Question

• How to make a ranking of “public opinion” regarding a company?
• What is the great public’s opinion regarding the companies and investors ?
• Just capital ranking is one thing. This metrics is complex and tries to take into account a high number of variables. But it just capital score correlated with public opinion ? It means that we will try to understand the public’s opinion regarding the companies that held many investments in social good versus those who do not invest in social good?

Results

Histogram of the different distribution :

Here is the distribution of the several scores. The first thing we can notice is the huge number of investors with no linked tweet. In fact, we have 80 percent of Null values for investors in opposition to compagnies where we have only 30 percent of Null values. This first observation make sense : in fact, compagnies will be more known by lamba citizens because compagnies are closer to their lifes compare to investors. Then, when you click on the legend to remove all categories with fill null values, data becomes clearer. The Vader score distribution is centered around 0.2 whereas FastText score and total score (mean between Vader and FastText score) are centered around 0.1. It means that the majority of the tweets are neutral with some positive connotations. In fact, scores are normalized between -1 and 1 : -1 means that the tweet in totally negative and + 1 means that the tweet is totally positive. Extreme negative or positive tweets are also present in the distribution, and this observation is also more visible for FastText score. It makes sense that the majortiy of tweets about the actors are neutral because theses corresponding tweet are usually neutral news.

Scatter plot showing dependency between Just Capital Scores and the mean of Vader and FastText method :

This graph illustrate the dependency between sentiment analysis score and Just capital score. In fact, the conclusion is that sentiment analysis score is not so correlated with just capital scores.

Graph analysis :

Here is some parameters of the graph plot. There are 4887 compagnies/investors linked with 16914 edges in total. The sparsity of the plot is 0.0052 which means that the graph is really dense. It contains one huge connected component. The largest component has 2517 nodes accounting for 100 % of the nodes and the longest shortest path between any two actors in the giant component is 7 It’s also called the diameter.

Here is the plot of the Degree distribution plot :

We observe that a small number of actors have a really high degree and that the majority of actors have small degree next to 0. It means that there are few really import actors compare to the others.

To finish this small graph analysis, we calculated the Homophily which is like correlation, but translated into graphs. It means how likely two compagnies that have the same attribute(just_capital_score or sentiment_analysis_score) are linked? We found that all homophily coefficient are around 0. In means thaht there is no correlation between factors and nodes.

Graph plot taking into account the degree and the sentiment analysis score :

The size of the node is scale in function of the degree of the node. Moreover, the color is given by the different sentiment analysis scores.

Discussion

Bias :

Unfortunatly, in this sentiment analysis, some bias can be highlighted. First of all, with the twitter dataset consisted of the tweet from 2017 we wanted to gives a indication of the public opinion per compagnies/investors. Therefore, we did not need to take the data from everyday. It was decided that it would be wiser and sufficient to take 10-days samples through the year 2017. This made the calculation feasible in Pandas. Since we were looking for information about investors and companies in the USA, we decided to only look at the tweet in English. This reduces the number of tweet per day by roughly 70% the majority of the Tweet are written in the mother tongue of the user (Chinese, Japanese, German, …). All theses assumptions can create bias.

Moreover, to link tweets and and compagnies/investors names, we search in tweets text and tags the name of the corresponding actors. This method makes senses but as language is a really complex process with a lot of possibilities, some tweets corresponding to one categorical compagny/investor could have not being linked.

To conlude this discussion, fastText method gives labels from 1 (negative sentence) to 5 (positive sentence) which is not a score between -1 and 1. Thus, we decide to add this labels number as integer and do a mean. In fact we choose the mean to penalized well extreme values. This way of creating a score can introduce a bias as the method is not made for that. Scores for actors with one tweet would be less robust than with several tweets.

Conclusion

As we explained, used tweets linked to investors/compagnies to make a ranking of “public opinion” regarding theses actors. Some bias exists, but by kowing them we are able to draw conclusion from the analysis of the results. For exmaple, a lot of investors do not have tweets linked to them. Moreover, we are able to conclude that the majority of theses tweets are neutral with positive connotations. From what we saw, we can conclude that public opinion on actors is decorelated from the their social impact. This is not logical. This fact can be explained due to the fact that twitter is not the best social media for finding social information about investors. Perhaps we should have focus on media. We found a small correlation but not high enough to draw another conclusion.

• Client: EPFL - ADA Course
• Date: December 2018
• Service: Sentiments

---

Abstract: Applying a geographic lens to the fund performance scores is important in order to grasp all the different implications of our study.

Introduction

Thanks to the provious substories. We were able to rank assets according to their “social good” value. All these companies certainly come from different states in the US and we wanted to know if certain regions of the USA have more asseets invested in social good project. For instance one could think that innovative regions like California should have higher investments than states which are know for their large amount of conservative people.

Used Dataset: SEC asset manager portfolios

Research Question

Where are the most ethical investment assets?

Results

First, one could easily see from the graph that some regions of the US do not have any company listed as having assets of sufficiently high value to be considered in our ranking (see the white state on the maps were no information pops up while passing the mouse over the map). The scores of the different states are heterogeneous. There is no cleat trend to observe. Furthermore, one could observe that our hypothesis was wrong. The Sate of Idaho has the highest social score according to our criteria, followed by Indiana, Washington and Colorado. The State of California is only in the midfield of our ranking. The States of Texas, Missouri or Nebraska have score bellow 1.5 in our ranking.

Discussion

Considering the low score of States like Texas, Missouri or Nebraska which are renown for the high influence of their conservative party, we can observe a certain trend that the States in the center of the US struggle more to invest in Social Good assets then the states on the Coast.

Conclusion

It is not just because a given State has many start-ups or a young population that the Companies of a given State will have larger amount of their portfolio invested in social good. It could be interesting to compare the plot that we generated with the last elections in the USA in order to see if there is a trend or not.

• Client: EPFL - ADA
• Date: December 2018
• Service: Geography

---

Abstract: data available outside the us

Coming Soon

One huge improvment of our project could be to generalized this stuy to the entire world. To do that, new dataset need to be found and analyse but having the a worldwide view could allows to understand lot more mechanisms behind theses investments.

• Client: ADA - EPFL
• Date: December 2018
• Service: outreach