Dear Mustachians,
I am considering Socially Responsible Investments (SRI).
In particular the iShare ones, based on MSCI “Environmental Social Governance” ratings.
In order to make a decision, I tried to learn what exactly are the ratings, how are they used, and how much do the ETFs based on them cost.
By sharing it, I hope to help readers decide if they would prefer donating to charity, according to their personal values and priorities. I also hope to get feedback and criticism to correct my misunderstandings.
I tried to be descriptive and factual, letting people decide for themselves if the price is worth paying.
ETFs can cost much more than their management on risk-adjusted basis (depending on how they’re taxed, what they borrow, …). That part of my reasoning in particular is quite sketchy…
I’d be very curious to read a thorough review of other SRIs as well.
Introduction to the ratings
The full presentation is in next chapter.
This method aims at producing a score, the MSCI “Environmental Social Governance” (ESG) rating.
- Analysts computing evaluate company actions (“management score”) scaled by business risk (“exposure”) to form key issue scores from 0 to 10. These key issue score indicates how bad/good is the company on a specific aspect.
- Concretely, if exposure is low, the score will be in the 3/10 to 6/10 range. If exposure is high, the score will be in the 0/10 to 10/10 range. Management score determines the actual value within that range.
- They then compute a weighted average of key issue scores according to weights depending on industry type.
Analysts have a lot of flexibility to set the weights. Nevertheless for the two segments I studied (Automobile and Software Services), they seem intuitively reasonable. - It finally gives a rating according to the score’s percentiles in the GICS segments considered.
Detailed presentation of the ratings
If links do not work, try searching them from Google first and clicking on the link there (I think they check the referrer). Feel free to skip if you’re not interested in the details.
Building the ~12 key issue scores
- For a given GICS segment , analysts pick about 12 “key issues” among 37 (full list is figure 1, page 4 in the methodology presentation)
- For example for the Automobile industry the following are scored (from a report on Volkswagen):
- Environment: Opportunities in Clean Tech ; Product Carbon Footprint ; Toxic Emissions & Waste ; Water Stress ; Carbon Emissions
- Social: Product Safety & Quality; Labor Management ; Health & Safety
- Governance: Business Ethics & Fraud ; Corporate Governance ; Anticompetitive Practices
- For example for the Automobile industry the following are scored (from a report on Volkswagen):
- For each key issue, analysts evaluate an “exposure score”. Exposure means “material impact of sin of business”, not as “inclination to sin” (unlike the weights we’ll see later).
- For example, Volkswagen has a high exposure to the “Labor Management” risk because its workforce is highly unionised (page 1 and 12 of this report).
- Volkswagen is also very exposed to “Product Safety & Quality” because “companies’ product portfolios contain segments where the incidence of quality or safety problems is high or there is high associated liability”(page 8 of this report).
- This definition of exposure surprised me at first. More on that later (see step 4).
- For each key issue, analysts assess how the company handled the issue to build a “management score”
- Analysts rate controversies in the last 3 years according to the table shown figure 5, page 8 in the methodology presentation
- Analysts collect a number of metrics relevant for the issue (figure 5, page 8 in the methodology presentation).
- For example, for Volkswagen’s “Product Safety & Quality” they check wether the company certifies its suppliers and how often are they trained (page 9 of this report)
- An undisclosed formula merges (3.1) and (3.2) to produce the management score
- A formula gives the “key issue score” for each of the picked issues by merging exposure (2) and management score (3) (figure 4, page 7 in the methodology presentation).
- The exposure score (2) determines the min and max value for the overall key issue score.
For example, if the exposure is low the score will be in range 3-6. If exposure is high the score will be in range 0- 10. - The management score (3) determines the score value within that range
- This sheds some light on the reasoning behind exposure, which surprised me at first:
- Lack of exposure means that no controversies can be witnessed (eg: no strike with no unions should not be 10/10).
- Lack of exposure also mean that fewer actions can be observed (no water saving measures in Ireland should not be 0/10).
- Companies do not get so much benefit of doubt because of the way ETFs are built (getting from 1/10 to 3/10 by repressing syndicates is unlikely to put you in a good percentile).
- The exposure score (2) determines the min and max value for the overall key issue score.
Building the rating from the ~12 key issue score:
- MSCI computes the percentiles of each key issue scores over past three years in each GICS segment. Percentile 2.5 becomes 0, percentile 97.5 becomes 10. This is the normalised key issue score.
- Similarly to how they only keep ~12 key issue to score, based on the GICS segments, they assign weights representing the “impact” the company has (figure 2, page 5 in the original document).
- Weights are in the range 0%-30%. A weight of 0% means that it’s worth keeping track of but not having material impact. A weight of 30% indicates that the industry is a major contributor to the environmental or social concern.
- Unlike the “exposure score” above which measures risks, the weight aims at proxying the impact this segment has on the world as a whole.
- For example for the Automobile industry the scores are as follow (from a report on Volkswagen):
- Environment (34%): Opportunities in Clean Tech (17%); Product Carbon Footprint (17%)
- Social: Product Safety & Quality (21%); Labor Management (17%)
- Governance (28%): Business Ethics & Fraud (14%) ; Corporate Governance (14%)
- The following have 0% weights: Carbon/Toxic Emissions (by the company itself, not by the product), Water Stress, Anticompetitive practices
- To contrast, for software and services the scores are as follow (from a report on Alphabet(Google), page 1):
- Environment (23%): Opportunities in Clean Tech (18%); Carbon emission (5%)
- Social (41%): Privacy and data security (23%); Human capital development (18%)
- Governance (36%): Corporate Governance (18%); Anticompetitive Practices (9%); Corruption & Instability (9%)
- They compute the weighted average of key issue scores with weights above. They then use a table to transcribe the numerical value into a rating from AAA (best) to CCC (worst).
Cases and observations
-
AAA: Betson , Report from 06-2018
Highest weight is: Corruption and instability (23%). Management score is 3.8/10, which is top quartile.
They observe for example that while the company does have a whistleblower policy (page 8 of Betson’s code of conduct), it does not commit on the legal protections of whistleblowers (“Whistleblower protection with no details on system or legal protection” page 19 ). - A: CISCO, Short tearsheet from 10-2016 (Not much to say, just a tearsheet)
- A: Google, Short tearsheet from 08-2016 (Not much to say, just a tearsheet. I appreciate that they seem to recognise the sophisticated technical measures taken to prevent data leaks, and have flagged its governance. Disclosure: They are my employer)
-
CCC: Equifax, Report from 06-2017
There is a selection bias here because it is put forward by MSCI as one of their success.
Nevertheless, it remains that they did downgrade Equifax in 2016 before the September 2017 scandal. -
CCC: Volkgswagen Report from 06-2018
Also selection bias because it is put forward by MSCI as one of their success. Similarly to above, they did flag it before a major scandal.
I would be interested to read more investigations.
What that implies for an ETF
Properties the ETF’s index should have
- Because ratings are relative to other companies within a segment, scores cannot be meaningfully compared across GICS segments. I cannot make sense of ETFs that attempt doing that (such as DSI).
- Maybe this is one of the reasons why most indexes screen out some industries, as eg: Betson can be “AAA” (maximum grade) if it is better than other betting company.
- The regulatory environment can be very different from region to region, so region is key to determine the level of reporting from companies. This impacts both exposure management scores making comparisons less meaningful across regions.
- The ability to document and report its actions is affected by the company size, and that influences the management score. I expect comparisons to be less meaningful across companies of different sizes.
- One can game the screenings in various ways, bringing profit from sinful activities below the thresholds of 5-15% . For example by merging, or with Hollywood accounting. This could in theory be an issue if screening is important to you. In practice, I don’t expect screened ETFs to become popular enough to make companies pull this kind of tricks. I didn’t count this possibility against ETFs.
ETFs based on MSCI ESG scores and meeting the above
The following ETFs seemed the most sensible.
They are structured by regions, and try to reflect the broader market as to the weights of each sector. They fail to properly account for the size, but this can be partially corrected. They only partially constrain sectors, though in practice they don’t seem to deviate as they try to replicate the parent index.
They are built by trying to optimise the percentile of ESG scores of its constituents, while constraining how different can they be from the “parent” (non ESG) index.
According to the indexes’ methodology (page 12), the constraints are such:
- Predicated tracking error: 0.5% (1% for ESGE)
- Min Constituent Weight: 0.1%
- Constituent Weight: ±2%
- Sector Weights: ± 5%
- Country Weights: ± 5% (not applicable to ESGU)
- Max turnover: ± 10% in May and November, 5% in February and August.
| Management | Ticker | Parent index | TER | Average Spread | Lending to issuer | Median premium/discount | Holdings |
|---|---|---|---|---|---|---|---|
| iShares | ESGU | US | 0.15% | 0.07% | 29 % | 0.07% | 292 |
| iShares | ESGD | Developed World ex-US | 0.20% | 0.08% | 20 % | 0.31% | 442 |
| iShares | ESGE | Emerging Markets | 0.25% | 0.14% | 20 % | 0.46% | 280 |
| iShares | ESML | US Small & Mid Cap | 0.17% | 0.13% | 29 % | 0.11% | 881 |
Note: All numbers in the tables are courtesy of the corresponding www.etf.com page
EASG also seemed very promising, but for some reason I was unable to find much information.
Evaluating the fees
There are many way to describe costs. I chose the following as the most intuitive:
If you would normally invest 10k of your favourite currency in VT, how much extra must you pay now to cancel the fees of these ETFs over the next 10 years?
I will be comparing VT with a mix of 35% ESGU, 35% ESGD, 10% ESGE, 20% ESML. My intention is to produce geographic distribution and sizes less dissimilar to VT.
Note: This part of my reasoning is a bit sketchy, I would love to hear of a cheap ETF that is easier to compare than VT, and the corresponding percentages applicable to the ESGU/D/E/L etfs.
Note: All numbers in the tables are courtesy of the corresponding www.etf.com page
Using Morningstar’s X-ray tool, I confirmed that the sector allocations were somewhat similar:
| Category | Basic Materials | Consumer Cyclical | Financial Services | Real Estate | Communication Services | Energy | Industrials | Technology | Consumer Defensive | Healthcare | Utilities |
|---|---|---|---|---|---|---|---|---|---|---|---|
| VT | 5.49 | 12.06 | 17.48 | 3.60 | 3.38 | 6.44 | 11.40 | 17.99 | 7.82 | 11.44 | 2.88 |
| mix | 5.65 | 12.07 | 17.58 | 4.34 | 3.49 | 5.64 | 11.32 | 17.06 | 7.90 | 11.60 | 3.53 |
Regional allocations were also similar between VT and the mix.
| Management | Ticker | TER | Average Spread | Lending to issuer | Median premium/discount | Holdings |
|---|---|---|---|---|---|---|
| Vanguard | VT | 0.10% | 0.01% | 0% | 0.10% | 7684 |
| iShares | mix | 0.184% | 0.093% | 24.5% | 0.213% | 1895 |
Assuming that investing costs half of the spread + the premium, the “entry bill” on a new investment with VT is 0.10%. With the mix it is 0.259%. The difference is of 0.159%
Assuming VT will have a y/y return rate of 7%, 10 years of TER costs (1.069^10^ - 1.06816^10^)/(1.069^10^)=0.783%.
So the extra money you need to pay is 1/((1-0.159%)*(1-0.783%))-1=0.950%.
Going back to our investing 10k example, an extra 95 of today’s money is a reasonable estimate of what you need to pay to cancel the cumulated fees over 10 years.
The cost
IE: How should you size your donation to Effective Altruism
This is actually a much harder question than the fees. Because the real cost is the difference of returns, after adjusting for risks.
Past returns are easy: on the day I am writing ESGU returned slightly more than VTI, over the past year, the past month, the past three months. But past returns don’t predict future returns.
Risks are even harder. It is likely that the fancy screening reduces some risks, namely ESG risks. But it seems that some properties of the replication algorithm make ESGU somehow pick up riskier stocks according to some other criteria.
Using Morningstar X-Ray’s tool, one can see that the mix contains 2.31% “distressed” stocks, against 1.10% for VT. But how to value that?
| High Yield | Distressed | Hard Asset | Cyclical | Slow Growth | Classic Growth | Aggressive Growth | Speculative Growth | n/a | |
|---|---|---|---|---|---|---|---|---|---|
| VT | 3.26 | 1.10 | 8.31 | 43.16 | 16.38 | 10.81 | 6.51 | 6.00 | 4.47 |
| mix | 2.03 | 2.32 | 9.00 | 44.38 | 15.94 | 9.68 | 5.55 | 6.51 | 4.59 |
I chose sidestep the question with the assumption that the portfolio is broad enough that risks are likely compensated.
IE: The reduction of ESG risks for a given price tag is probably compensated by the raise of some other risks and/or by lower returns and/or by the minimum cost of implementation.
That leaves open the idea that some portion of the fees could be compensated by higher risk adjusted returns. But because minimum costs of implementation are almost certainly lower than what the ETF charges, I don’t expect to recover fees in this way.
So in the end I chose to use the fees to approximate the cost.