YvesBlue set out to create a more transparent approach to scoring ESG Data. The YvesBlue Public Equity ESG score consists of a purely quantitative and succinct set of ESG metrics for each pillar sourced from 450+ Refinitiv indicators. It can be used for comparison or in conjunction with other scoring methodologies or just used on its own. It is also for combining with other scoring and analytics provided by YvesBlue. The following describes the steps taken to provide the pillar and overall scores at the instrument and portfolio levels.
Data that gets used in the final scoring process can be in one of three states:
Reported: Company has reported the data point
Backfilled: We have pulled forward previous years' data to fill in
Modeled: If there is no data to pull forward, the value has been estimated using our medians methodology described below
The Calculation Steps:
I. Pillar Selection, Data Clean and Prep
II. Normalize
III. Backfill
IV. Model
V. Standardize
VI. Pillar and Total ESG Score
VII. Scoring Process
I. ESG Pillar Metric Selection and Data Cleaning and Prep:
II. Normalize
The reported raw metrics are converted to USD and normalized per $1 million of company revenue unless the metric does not scale with company activity.
III. Backfill
Sometimes a current year’s worth of data for a company has not come in yet or the company is a few periods behind in reporting. The data is filled in with the most recent reported data up to 5 years.
IV. Model:
For companies that have not reported anything over the last 5 years for a given metric, there is a process in place to give the best estimation. The value is modeled by creating an estimation set or group of companies that provide the greatest amount of data per company for the field in question.
YB creates an estimation set (excluding micro-cap companies) based on reported medians in three contexts: activity classification, country, and market-cap grouping [micro, small, mid, large, and mega-cap companies].
For a given indicator, if there are enough companies reporting in that cross-section, that median will be used as the fill-in value. However, if there are not enough companies reporting that indicator, the model uses a wider cross-section to get more data, e.g., activity classification combined with enterprise value level.
The process will widen until there are at least 32 companies reporting in that metric to provide a statistically rational median. With this approach, we can model more companies giving us a coverage set that includes companies that have only one modeled value all the way to companies that have most of their values being modeled.
V. Standardize:
The standard score, also called the Z-score, helps to make unlike data usable for operations like averaging or summing across.
VI. Pillar and Total ESG Score:
The Z-scores are then ready to be equally weighted and then averaged across each pillar to provide the E, S, and G pillar scores. After which, the three pillar scores are then averaged to give the total ESG score.
VII. Scoring Process Deep Dive:
Below is a high-level view of the contexts we develop in order to model for missing values. All companies in our database are categorized into estimation sets or “contexts” that allow us to apply a median value for a missing datapoint.
Example of the entire process using Tesla Inc.
Data that gets used in the final scoring process can be in one of three states:
Reported: Company has reported the data point
Backfilled: We have pulled forward previous years' data to fill in
Modeled: If there is no data to pull forward, the value has been estimated using our medians methodology described below
Polarity: When a high value requires a low score
Modeling: When a field is not reported by a company and there is no data from previous years to fill it in, the medial value for companies of the same industrial activity and market cap is used. Where there are not at least 32 companies in that basket, the process moves to take the median from a bigger basket of companies, escalating and widening the context until there is an appropriate number from which to derive the median.
Normalizing: Finally, the values for each indicator in each pillar, whether they be reported, backfilled, or medians, are averaged, z-scored, outliers are controlled and given a sector rank, and the z-score is normalized to 10.
[*] Micro, small, mid and large, and mega-cap companies.
* Challenges with this data include:
Each company’s coverage through Refinitiv can vary between extracts due to regular corporate actions.
Not all changes in identifiers are able to be mapped between extracts.
This leads to some companies being dropped from our analysis.
Reporting periods and cadences can differ between companies.
Different fiscal year definitions cause measures to represent various periods in time from company to company.
Values for the latest fiscal year might take up to a year for Refinitiv to capture, rendering current-year reporting spotty and unreliable.
Changes to company reporting definitions and standards can lead to historical revisions.
This necessitates that we reprocess the full five years of history with each ESG refresh to ensure internal consistency.
Confidence Intervals: Confidence intervals are commonly used to express how valid a number may be within plus or minus a few points. YvesBlue confidence intervals (to be rolled out soon) are very simply based on the number of missing values that had to be imputed for a given company.