ESG Data: Evidence of materiality and alpha from continuous, high-frequency data

Evidence that ESG data is moving from the ‘not yet financial’ to ‘emerging financial’

There’s a growing body of evidence that environmental, social, and corporate governance (ESG) measures of public companies are linked to medium and long-term higher performance using various financial and operational metrics. In some cases there is evidence of finding ‘alpha’, that is, risk adjusted outperformance with regard to a particular benchmark, such as the S&P 500. Yet none of these studies and findings is from data that is generated continuously day in, day out.

Working with colleagues from Saint Mary’s College of California and from TruValue Labs (TVL) (1) we found, in a sub-sample, ‘alpha’ (using the S&P 500 as our benchmark) from TVL’s continuously generated, near-real-time data.

In what we called a minority of ‘extreme cases’ this finding was predictive at a very high level of statistical significance, a 99% confidence interval. The study is initial, limited and exploratory. Initial in that there is more to explore, limited in that it covers only one year, and exploratory as it is the first study to analyze this type of data.

The title of the paper says it all: Computer-generated high-frequency corporate environmental data: an event study with predictive power. We looked at 2015 TVL data for three environmental categories (land, water, air). For what we defined as ‘extreme events’ (quickly, ones with statistically significant high or low score levels in a small window of time), there is predictive power from TV data movement correlated with equity price. The significance of this is twofold: short-term ‘alpha’ findings in near-real-time, as well as evidence for market confirmation of the materiality of ESG factors.

Our study is an event study. Event studies cull an event or a series of events to measure how the event (for example, a firm’s new technology which dramatically reduced pollution or a failure of such technology) impacts the firm’s stock price. Most of these studies are relatively short term, say within a week or two.
Event studies in the ESG space have of necessity been retrospective because no timely data has been previously available. The potential of continuously generated, near real-time data is that it can be used for prospective rather than (just) retrospective analysis and action.

We think that our study’s findings are not only a near-real-time market-based materiality indicator, but also suggest, somewhat counter intuitively, that short-term ‘alpha’ has long-term investment implications aside from giving legitimacy to the ESG value proposition. Obviously the long-term is composed of a continuously moving short-term. While there may be ‘alpha’ potential, to the degree that it regresses to the mean, the real value add gets factored in to stock price, to the degree that markets are efficient. The short term can provide a ‘health check’ on whether firms are either making progress, and can point to emerging longer-term trends in economic and not just financial terms.

What we found and how we found it

We focused on five sustainability-environment scores using TVL’s ‘InsightScore’ dataset: air, land, water data, an average of these three, in addition to the aggregate TVL score which includes all fourteen of TVL’s categories.

The study asks whether the stock market reacts to firms that have better or worse ESG performance. We designed the events as the observed statistically significantly high or low (positive or negative) score levels for a firm, and then investigated the stock returns around the event dates.In order to do this, we calculated the mean and standard deviation of each sustainability score in the complete sample year of 2015, and constructed the 99% confidence intervals (which turned out to be mean ± 2.58 * standard deviation) to estimate population proportions with positive and negative extreme values (2).
From these data we found 59 positive events and 39 negative events.
We used daily observations finding that the value in Day T is highly related to the values in previous day—and some extreme values that happen in subsequent days. In other words, if a value in Day T is identified as an extreme value and an event date, it’s highly likely that the value in the following days also have extreme values. We used the event (T=0) examining the ten days following and the three days before (-3,10). We check these for robustness and they worked well.
We looked at stock return, cumulative return and compounded cumulative return. We then focused on abnormal return (AR) and abnormal cumulative return (CAR) (3)

Let’s look at one example for AR and CAR. The first is for AR for the environment atmosphere category (See download on the left, TruValue1.pdf).

We see the positive CR in the T+10 day window was 4.39% and 5.08% for the period T-3 to T+10, that is, 13 days total. In this example the market responded to positive event data. The negative was less strong, but did predict some negative movement for T+9 days. Among issues that await future study are the detailed events causing both movement and this level of prediction.

The second example is again from environment atmosphere, this time using CAR (See second download on the left, TruValue2.pdf).

For the negative we see a similar trend as with AR, but a stronger one for CAR.
These two examples were the strongest from the study, but most other individual environmental categories followed similar patterns, albeit with more complexity.

From just under a million data points, we isolated 98 instances of highly statistically significant events for 2015. Averaging these across the year suggests that there was one significant event about every 2.57 trading days (on the NYSE), and thus (again on average) one significant event for every approximately 3,600 observations per trading day.
The use of near-real-time data has at least two implications. The first suggests it is possible for ‘extreme events’ to have a very high level of predictive ability for both positive and negative information correlated with equity movement. Such ‘alpha’ findings can be generated and become actionable in near real time
The second implication finds the correlation confirms what other studies have found: that data traditionally called “non-financial” but more accurately “not yet financial” or “emerging financial” is in fact material, and its materiality can be predicted and identified in close to real time. That is important evidence for the case that ESG data is moving from the not yet financial to emerging financial, and will matter more and more as investors realize its value and incorporate it into their work.

Jim Hawley is Head of ESG Research at TruValue Labs and Professor at Saint Mary’s College of California. He is contactable via email here and here. (Footnotes overleaf)

(1) Greg Bala, TruValue Labs, and professors Yung-Jae Lee and Tina Zhang.
(2) All data is from 2015 (365 days) with usable observations from the S&P 500 of 497 firms, with nearly 200,000 observations over all firms per category, resulting in nearly a million total observations when including the environmental average and the overall TVL score.
(3) Stock return is defined as the change in the daily adjust-close price divided by the adjust-close price of the stock prior day. Cumulative return is computed as a cumulative sum of stock return for each firm over a specified time window. Cumulative abnormal return is found from the event date that is the date associated with extreme values of sustainability scores.