A field experiment, explained
Can index providers use their voice to drive corporate climate action?
Index providers set the rules that decide which companies index funds hold. In February 2022, one provider’s CEO wrote to the board chairs of 300 randomly drawn companies, restating climate rules that were already public.
A letter with nothing new in it. Did it change what the companies did?
InsideClimate targetsReported emissionsFirm value
A visual summary of Florian Heeb and Julian F. Kölbel (2026), “Index Provider Engagement and Climate Action: A Field Experiment”, working paper, SSRN 4711873, built from the paper’s data. Made with Claude Code and Nityesh Agarwal’s explorable-explanations skill (MIT License).
This visual summary may contain glitches. For anything beyond visual exploration, please refer to the paper.
Authors
Florian Heeb
Leibniz Institute for Financial Research SAFE
Goethe University Frankfurt
MIT Sloan School of Management
Julian F. Kölbel
University of St. Gallen
Swiss Finance Institute
MIT Sloan School of Management
Setup · The letter
What the letter said, in four stamps
The letter itself is not public. The paper describes what it said, and the four stamps tell it in pictures.
Why would a chair care? Funds that track an index follow its rules, so a firm that drops out of an index can expect those funds to sell its shares.
Every rule in the letter was already public. What was new was the delivery and the ask: one letter to the chair, with the company’s name, the consequence spelled out, and a request to commit by the end of 2022.
Side path · The rules
The rules were on file before the letter
The provider’s climate indices, launched in 2020, follow the EU rules for climate benchmarks. On top of those rules, the provider used the SBTi as a gatekeeper: firms without a validated target were phased out on a published schedule.
Pick a firm and an index to see what the schedule meant for it.
In December 2025, near the end of the study, the provider replaced the strict SBTi exclusions with a gradual underweighting of firms without a commitment or target.
Side path · The money
How much money followed these rules?
No one knows exactly. Two kinds of money follow indices like these.
Exchange-traded funds copy an index directly. ETFs tracking EU climate benchmarks from all providers held about USD 18 billion just before the letters and about USD 125 billion by early 2026. ETFs tracking this provider’s climate indices held about USD 45 million in early 2022.
Large investors can also license an index as the benchmark for their own portfolios. The provider’s indices are licensed by financial institutions with several trillion dollars under management, including the large asset owners who helped develop them.
How much of the licensees’ money is benchmarked to these indices is unknown, and the firms likely faced the same uncertainty.
Side path · Science-based targets
Two steps and a 24-month clock
The Science Based Targets initiative is an NGO, unrelated to the index provider. A company first commits by sending it a letter. It then has 24 months to submit a target, which the SBTi validates if it meets criteria designed to align corporate targets with limiting warming to 1.5°C.
Drag the commitment date to see the deadline move.
Keep this clock in mind. It shapes when the letter’s effect shows up.
Side path · The registry
A public record anyone can check
Every commitment and every validated target in the SBTi’s public registry carries its own date. Because the registry is independent of the index provider, the researchers could track every firm in the experiment on their own. By August 2026 it listed more than 14,000 companies.
Since 2023 the SBTi marks commitments that miss the 24-month deadline as removed. The main outcome counts a firm from the day it commits or sets a target, even if a commitment later expires. The robustness slide near the end shows what changes if expired commitments don’t count.
| Company | Committed | Target validated | Status |
|---|---|---|---|
| Firm A | Jun 2022 | Feb 2024 | Target set |
| Firm B | Mar 2023 | – | Committed |
| Firm C | Sep 2022 | – | Removed |
| Firm D | – | – | Not listed |
Illustration with made-up firms. In the main outcome, Firms A, B and C all count from the month they committed.
Setup · Building the sample
Which firms could get a letter?
The letter asked firms to commit to a science-based target, so it only made sense for firms that had none. The researchers started from every firm in the provider’s climate indices in December 2021 and set two groups aside.
- 669 firms that had already committed to or set a target with the SBTi. The letter would have asked them for something they had already done.
- 61 firms on the Climate Action 100+ list. Climate Action 100+ is an investor initiative that engages the 170 largest listed emitters and asks them to set climate targets. The researchers expected a letter to add little to that.
That left 1,227 firms: the sample.
Setup · Why the sample matters
1,227 firms, 39% of the world’s listed market value
Before the letters, the 1,227 firms were worth about USD 30 trillion together, and they emitted roughly 1.35 billion tonnes of CO2-equivalent a year directly (Scope 1).
That is about 39% of the market value and 12% of the direct emissions of all listed companies worldwide. These are large firms that matter for markets and for the climate.
Because Climate Action 100+ firms are left out, some of the world’s largest emitters are not in the experiment.
Setup · The randomization
300 letters, drawn by lot
This is the randomization at the heart of the experiment. The provider had capacity for 300 letters, and the researchers drew the 300 recipients from the 1,227 firms at random, within each country. Every firm in a country had the same chance of getting one, about one in four.
The firms that got a letter form the treatment group; the firms that did not form the control group. Because chance alone decided, a later difference between them is the letter’s effect, give or take luck.
Why country by country? National policies strongly predict whether firms set climate targets. Drawing within each country keeps the two groups alike on that.
That makes 300 firms in the treatment group and 927 in the control group. 16 of the 1,227 firms (3 of them letter firms) turned out to have an SBTi commitment or target by the time the letters arrived. They stay in the analysis, except in the target estimates that adjust for firm characteristics, which leave them out.
Side path · Balance
Alike before the letter
If the draw worked, the two groups should look alike before any letter went out. They do. On every characteristic the paper checks, the averages are close.
A test of all of them together: p = 0.73.
A p-value this high means differences this small are just what a random draw produces. Because chance alone decided who got a letter, the groups should also be alike on traits nobody measured.
Hover or tap a row to see how much firms vary within each group.
Results · What we find
That was the setup. Now: what did the letter change?
From here on, every result compares the 300 firms that got a letter with the 927 that did not (for emissions, the 1,120 firms with data in every year). Because chance alone decided who got one, a gap between the two groups is the letter’s effect, give or take luck.
Result 1 · Climate targets
28.0% against 21.6%
By January 2024, two years after the letter, 84 of the 300 firms in the treatment group had committed to or set a science-based target, against 200 of the 927 in the control group. That is a gap of 6.4 percentage points, or about 30% more firms acting.
Chance decided who got a letter, so the gap, ringed in gold, is the letter’s effect, give or take luck. The ruler under the targets shows how much give: the range of effects the data cannot rule out.
With firm characteristics adjusted for, and the 16 firms that had already acted before the letters left out, the estimate is 7.4 points.
Result 1 · Two steps
Commitments first, targets later
The SBTi works in two steps. A firm commits, then has 24 months to submit a target for validation. The letter asked firms to commit by the end of 2022.
The letter’s effect follows that clock. In the first year it mostly produced commitments: by January 2023, 3.5 points more letter firms had committed without yet setting a target. In the second year validated targets followed: by January 2024, 12.0% of letter firms had one, against 7.6% of the others, a gap of 4.4 points (computed before rounding).
Lines show the gap between the treatment and control groups at each quarter, with 90% confidence bands. The paper’s Figure 2 shows the same pattern with firm characteristics adjusted for.
Side path · Luck or letter?
Suppose the letter did nothing
Then the 284 firms that had acted by January 2024 would have acted anyway. Who got a letter would be just a label, and you could deal it out again.
Re-deal the letters the way the original draw was made, country by country. Japan has 208 firms and 51 letters, so each re-deal picks a fresh 51 of the 208, and so on for every country. Each re-deal gives an estimate that luck alone produced.
Every re-deal here is one of the paper’s own 2,000, scored with its most precise estimate: firm characteristics adjusted for, and the 16 firms that had acted before the letters left out (1,181 firms). For January 2024 that estimate is 7.4 points.
Side path · Many shuffles
What luck usually looks like
One re-deal tells you little. Thousands show you the range of estimates luck produces.
In the paper’s 2,000 re-deals, an estimate at least as large as the real 7.4 points came up about 5 times in 1,000 (p = 0.005). The run here replays those same re-deals in a random order, so once all 2,000 are in, it lands on the paper’s count. Reset it to watch them arrive in another order.
Side path · Other dates
Earlier and later
Pick another date, and switch between the raw gap (no controls) and the estimate with firm characteristics adjusted for (the 16 prior adopters left out).
One year after the letter, the raw gap was 3.0 points, and luck produces one that large about 20 times in 100 (p = 0.203). With controls the estimate was 4.1 points, matched by luck about 5 times in 100 (p = 0.050). From January 2024 on, luck matches the real effect at most about 6 times in 100 without controls and 3 with them.
The paper uses this test throughout. For the regressions it re-deals the letters 2,000 times and re-estimates the model on each deal.
Side path · The registry fills up
Watch the registry fill up
Each target is a firm in the experiment: blue with a letter (treatment), gray without (control). An outlined target means the firm has committed to the SBTi; a filled one means its target has been validated.
Press play, or drag through the four years after the letter.
Side path · Four years
The gap over four years
The gap grows through the first two years, peaks at 7.5 points in April 2024, and is still 5.9 points four years after the letter (significant at the 10% level).
The preregistration named January 1, 2023 as the outcome date. The raw gap was then 3.0 points, not statistically significant (p = 0.203); with firm characteristics adjusted for, it was 4.1 points (p = 0.050). The raw gap became significant in the second year, as more letter firms than control firms joined the registry.
The shaded band shows the range of effects the data cannot rule out (90% confidence). Switch to the paper’s estimates with firm characteristics adjusted for.
Result 2 · Reported emissions
Targets are promises. What about tonnes?
A target is a promise about the future. Reported emissions are a separate outcome: did the letter also change the tonnes firms report? The experiment can say what the letter did to each, not whether the targets caused the cuts. In 2021, the last year reported before the letter, the two groups’ smokestacks were about the same size.
Your bet: by the fourth reporting year after the letter, how did the letter firms’ cloud compare with the others’?
Reported emissions are direct emissions plus those from purchased energy (Scope 1 and 2), from S&P Trucost, for the 1,120 firms with data in every year from 2018 to 2025. Trucost estimates emissions for firms that do not fully disclose them.
Result 2 · Year by year
Reported emissions 18% lower by year four
Drag through the reporting years. The top panel shows each group’s mean reported emissions per firm; the bottom panel is a difference-in-differences: the letter firms’ change since 2021 minus the control group’s. No pre-trends: before the letter it stays near zero (joint test p = 0.64). After it the gap opens: 4% in 2022 (not significant), then 14%, 17% and 18%. These are ; part of the cut may come from selling plants.
90% bands; p-values from re-dealing the letters. Poisson pseudo-maximum likelihood with firm and year fixed effects; equal weights per firm give similar estimates.
Side path · Reporting years
Which years count as after the letter?
Firms report a year’s emissions in the following year. Reporting year 2022 covers fiscal year 2021, which was largely over when the letter arrived in February 2022.
Those figures were not yet reported when the letter arrived, and the letter could still have affected them, so the paper counts reporting year 2022 as after the letter. Counting it as before instead leaves the pooled result similar and significant.
Side path · Stress tests
Try to break it
Before believing a result, try to break it. Each switch adds another version of the estimate next to the main one: the gap averaged over the four years after the letter, with its 90% band. The p-values come from re-dealing the letters.
The emissions data come from S&P Trucost, which estimates emissions for firms that do not fully disclose them. Start there.
Side path · Scopes
Most of the cut is in direct emissions
Direct emissions (Scope 1), from sources a firm owns or controls, fall by about 18% on average over the four years, against about 13% for Scope 1 and 2 together. They make up 75% of the tonnes measured here.
Emissions from purchased energy (Scope 2) also point down, but not significantly. Upstream emissions (Scope 3), which are hard to measure and compare across firms, show no detectable effect.
90% bands; p-values from re-dealing the letters.
Side path · A puzzle
One firm in thirteen, 18% lower reported emissions
At its peak the letter’s effect on targets was 7.5 points, roughly one letter firm in thirteen committing or setting a target because of it. Yet the letter group’s reported emissions ended up 18% lower. How can so few firms move so many reported tonnes?
The paper offers three possible explanations, and they can all be at work at once. Pick one to start.
Side path · Possible explanation A
A few big emitters
Counting firms treats a small firm and a giant alike. Flip the switch to weight each firm by its emissions before the letter.
Weighted this way, the targets gap rises from 6.4 to 14.2 points. Firms that went on to cut their reported emissions by half or more held 24.5% of the letter group’s emissions before the letter, against 7.5% of the others’. Above-median emitters, with 98% of the sample’s emissions before the letter, show a significant effect on targets, though the two halves do not differ significantly.
Side path · Possible explanation B
Cuts without a target
The letter asked for a target, but EU climate benchmarks also reward lower emissions directly. A transition benchmark must be at least 30% less carbon-intensive than its parent index, a Paris-aligned one at least 50%, and both must cut intensity by at least 7% a year. Providers get there by choosing and weighting constituents.
So a firm could read the letter as a broader signal that investors care about its carbon, and cut without setting a target. The letter may also have encouraged firms that would have set a target anyway to follow through with meaningful emission reductions.
Side path · Possible explanation C
Quiet cuts
Public sustainability commitments have become politically costlier since 2022, according to studies the paper cites. New SBTi commitments by firms in the experiment, with and without a letter, fell from 118 in 2022 to 19 in 2025, and in December 2025 the index provider itself relaxed its SBTi exclusions.
A firm caught between anti-ESG pressure and climate-screened investors may cut emissions without ever appearing in the registry. Some call this greenhushing. Switch the lens to see what each kind of data records. The experiment has no direct test of it.
Side path · What the data can say
One explanation has evidence, two are open
Only the first has direct support in the data, and some of its large cuts may be divestments. The other two are possibilities the paper raises without a direct test. All three can be at work at once.
Side path · On paper or in the air?
Three ways to cut reported tonnes
Here are three ways a firm can lower the emissions it reports. Only one of them lowers the tonnes in the air. Try each.
Toy numbers for one made-up firm.
Side path · What firms said
Large letter firms appear to cite divestment more often
Firms that report to CDP explain why their emissions changed. Nearly all firms in both groups cite emission-reduction activities (92.5% and 92.7%). Counting firms, no reason differs significantly between the groups.
Weighted by emissions before the letter, divestment stands out. Letter firms citing it held 43.5% of their group’s pre-letter emissions, against 21.1% (p = 0.084, suggestive). Only 504 firms explain their changes, so treat this with some caution.
The timing fits reporting choices. The first reporting year after the letter mostly covers months before it, yet already shows a 4% gap (not significant); any response there could only come from how the inventory was compiled.
Result 3 · Firm value
Did shareholders pay for it?
Climate targets and emission cuts can cost money, or they can pay off; the evidence on how such decisions affect firm value is mixed. The chart follows equal-weighted portfolios of the treatment and control groups, from a year before the letter to four years after.
Your bet: compared with firms that got no letter, did the letter firms’ shareholders do better, worse or about the same?
This part of the study was not preregistered.
Result 3 · Returns
No detectable difference
Put the two groups on a balance. Buy the letter firms, sell the firms without a letter, and strip out the market’s ups and downs: what is left is the alpha. If the letter had paid off for shareholders, the beam would tip toward the letter firms; if it had cost them, toward the control group.
Over the full four years the alpha is −0.02 percentage points a year (p = 0.984). Drag through the windows: from six months to four years, the shaded range always includes the level beam, so the alpha is never distinguishable from zero. Other factor models give the same answer.
Value-weighted portfolios and the two emitter halves also show no detectable difference, though their estimates are much less precise. The p-values come from re-dealing the letters.
Side path · What could hide
Could a small effect hide here?
Drag a hypothetical true effect. The curve shows the estimates this study could have produced if that were the truth. The shaded part is the chance the test would have flagged it.
An effect of 1.9 points a year or more would likely have shown up (an 80% chance). Smaller effects, which are plausible for a single letter, could go unnoticed.
Interpretation · Why would it work?
The same rules, addressed to the chair
Firms moved, and firm value showed no detectable change. Yet the letter told firms nothing new: the rules sat in a public methodology, written for anyone who looked. The letter put them in front of one person, the chair, with the company’s name, a date and a consequence. The paper discusses three ways that could matter.
- Scarce attention at the top the authors' favorite explanation
Boards cannot attend to everything. The letter may have put a rule that had gone unnoticed on the board’s agenda. - Costly information
Tracking many investors’ rules may be costly; the letter delivered part of that information at no cost to the firm. - Agency
Managers who favored climate action may have used the letter as outside backing.
These accounts can all hold at once; the experiment identifies the letter’s effect, not the friction it relieved. Not every firm welcomed the letter: some survey replies asked that such matters not be addressed to the chair.
Wrap-up · What it means
Can an index provider drive corporate climate action by restating its public rules?
The answer: yes. In this experiment, it did. One letter to each board chair raised commitments and targets and lowered reported emissions, with no detectable effect on firm value in either direction.
For index providers and asset owners, published rules and direct letters appear to work together, and a letter costs little to send. One caution for rule designers: a firm can lower its reported emissions by redrawing the boundary of what it reports, so rules aimed at physical decarbonization may need to reward physical abatement.
Side path · Limits
What the experiment cannot tell you
Wrap-up · Robustness
Robustness: look for the weak spots yourself
The paper’s main estimates next to the robustness checks the authors report: other samples, weights, estimators, data sources and factor models, for targets, reported emissions and share prices.
Each line shows the range of effects the data cannot rule out (90% confidence). The p-values come from re-dealing the letters, and hollow dots mark p-values of 0.10 or more.
Wrap-up · Write back
Thanks for reading. The whole story is in the paper.
The full paperSSRN4711873 Index Provider Engagement and Climate Action: A Field Experiment Florian Heeb & Julian F. Kölbel Read it on SSRN ↗Florian Heeb (SAFE, Goethe University Frankfurt, MIT Sloan) and Julian F. Kölbel (University of St. Gallen, Swiss Finance Institute, MIT Sloan). Numbers from the current draft. This page was built with Claude Code and the explorable-explanations skill by Nityesh Agarwal (MIT License), in the spirit of Nicky Case’s explorable explanations.