Raw ESG Variables Outperform Aggregated Scores for Financial Risk

If you want to know whether a company’s stock is likely to swing sharply, a single ESG score may hide the clues. This paper argues that the useful signals are often buried in the raw environmental, social, and governance variables collected from company reports. The authors built a framework for ESG data with a hierarchical structure and many more variables than observations, then used it to pick the variables most tied to financial risk, measured as logarithmic volatility of return. They found that these selected raw variables were significantly more relevant to risk than aggregated ESG scores, and that they added information beyond traditional financial factors. The framework also held up in out-of-sample tests. Using company data from several sectors in the US economy, the study identifies which ESG risk variables matter for large and small companies within each sector.

A single ESG score can make a company look neat. The raw fields underneath can tell a harsher story. If you want to know whether a stock will swing hard, that hidden layer matters. ESG means environmental, social, and governance data. Companies build it from many small clues in reports. One blended score can smooth away those clues. This study looks at that buried layer instead. It asks which raw ESG variables track financial risk best. Financial risk here means logarithmic volatility of return. That is how sharply returns move over time. The surprise is simple. The raw clues beat the tidy summary.

What the score hides

Across sectors in the US economy, the framework found better clues in the raw data. The chosen variables lined up more closely with financial risk than ESG scores did. That difference was significant. The raw picks also added information beyond usual financial factors. That means they explained something the usual market numbers did not. The framework did not stop at one firm type. It also separated large companies from small ones inside each sector. That matters because risk can hide in different places for different firm sizes. The results held up in out-of-sample checks. That means fresh data, not the data used to build the model, still fit the pattern. So the advantage was not a one-off fit to one sample.

How the raw fields get sorted

The framework starts with raw ESG data from company reports. Data vendors like LSEG, Bloomberg, and MSCI collect those fields. They then place them inside a hierarchy. A hierarchy is a tree-like order of groups and parts. That structure matters when there are many more variables than observations. Observations are the companies or records you can compare. The method screens the raw fields against logarithmic volatility of return. It keeps the ones that best line up with risk. Then it checks the picks on out-of-sample data. That is fresh data that the model did not use to learn. This helps show that the pattern can travel beyond one fit.

Why the raw clues matter

This matters because one ESG score can hide the parts that move risk. The raw fields give a closer view of where trouble may sit. That helps analysts look past the headline total. It also helps them separate ESG signal from ordinary financial signal. The framework points to sector-specific clues. It also points to company-size differences inside each sector. So the same score can mean different things for large and small firms. That is a useful warning for anyone who treats ESG as one clean number.

Where the search goes next

The surprise here is that the tidy score loses to the messy inputs. That shifts the first question in ESG work. The question is no longer what is the score. It becomes which raw fields drive risk in this sector. The framework makes that search sharper for large and small firms. It also keeps the out-of-sample check in view. That matters because a good fit on old data can still mislead. So the blended score becomes a shortcut. It is no longer the end of the search. The raw clues stay on the table.