20 December 2021

A recent study claims to show that divestment strategies help to fight climate change. But careful analysis shows how important it is to look beyond the headlines.

On December 7th 2021, Responsible Investor published an article titled: Major study highlights real-economy impacts of climate divestment. The strap line went on to say:

Research on 4,500 investment funds finds “causal link” between divestment and company decarbonisation.

There’s a lively debate on whether divestment or engagement is the better approach to bringing about change in the real economy. Those advocating engagement say that divestment just results in dirty assets being pushed into the hands of less scrupulous owners or less transparent ownership models, without affecting emissions. Those advocating divestment say that engagement means, in reality, cosy conversations with little tangible output and that divestment sends a stronger signal by kicking dirty companies where it hurts: in the finances. I’ve summarised the evidence on either side of this debate elsewhere.

The debate matters because, increasingly, funds are being invested in line with sustainability goals. University endowments, pension funds, and retail investors are pouring money into “sustainable” funds. If these funds are, in fact, not achieving real world change then Tariq Fancy will be proved right: sustainable investing will just be a dangerous placebo with the side effect of enriching the investment industry without bringing about real-world change. If, by contrast, the allocation of funds is contributing to real-world acceleration of corporate action on climate change then finance can fulfil at least some of the hopes expressed for it at COP26. 

The importance of this debate creates a critical role for robust academic research. It’s not obvious whether engagement or divestment is better, or whether either has any impact at all. Careful academic studies can tease out conclusions from the data, correctly controlling for factors that may be confusing the results and separating correlation from causation. The importance of this role is why I now devote most of my professional life to helping create a bridge between academic research and practice in the area of responsible business. 

As someone who believes in the positive impact academic research can have on the real world I’m very supportive of making academic findings accessible and promoting them in the media so they are seen by practitioners. But there’s also a risk. Studies that feed the confirmation bias of a particular side of the argument are more likely to get picked up and promoted. Academics that find newsworthy results may be more likely to get invited to be keynote speakers and to participate in practitioner conferences. A high public profile can increase the chances of being asked to advise on policy issues or other important matters. So there are strong incentives to over-claim eye-catching results that may be consistent with the data, but aren’t really fully supported by it. More innocently, academic findings often contain nuances that can be lost in the simplification of media commentary. 

So scepticism of headlines describing the latest research is warranted. Including in this case. It turns out that detailed scrutiny of the study doesn’t back up the headline. Here’s why. 

What the paper does

The publicly available version on SSRN only contains the abstract, which is unusual. There is a public link to the full paper here, although it will only work until 9th January 2022 (this was provided in the description of a YouTube video produced by the authors describing their results). For those with journal access through an academic library, the full paper has now been published in the Journal of Banking and Finance. 

In summary, the paper does three things:

1. The authors comes up with an approach to identify divestments that are a result of portfolio decarbonisation efforts.

2. The analysis then shows that those divestments lead to share price falls in affected companies.

3. Finally, the analysis shows companies subject to that pressure reduce carbon emissions quicker than others.

The authors’ conclusions draw a thread from the divestment pressure to the carbon emissions, claiming causality.

Watch the data

I have some concerns about the approach used in the paper to identifying decarbonisation trades under Step 1 as it is based on some quite specific choices around cut-offs, data sufficiency and so on. Data mining is always a concern in econometric analysis and it would be comforting if the authors were able to present analysis of the impact of making alternative choices on this first step. There are also changes made to the empirical set-up when they get to the analysis in Step 3. It is slightly disconcerting when methodological horses are changed part way through an analysis. They provide reasons for this, but there is no assessment of the sensitivity of the results to the choices made. 

If raising concerns about data mining sounds like a cheap shot:

• it is, because I have no evidence that the analysis does indeed suffer from data mining problems; but

• it isn’t, because non-other than the eminent Campbell Harvey, in work with co-authors Yan Liu and Heqing Zhu, argues that data-mining problems mean that “…most claimed research findings in financial economics are likely false.”

So scrupulous care in data analysis is the bedrock on which robust research is based, and it is always worth casting a critical eye over the data methodology. 

But let’s suppose for now that my concerns are unfounded and that the results are real conclusion rather than suffering from data-mining. What does the paper find?

Correlation or causation?

Their first result is that stocks subject to divestment by funds looking to decarbonise fall in value. This isn’t news – it is well known that markets aren’t perfectly efficient and that fund flows affect share prices positively and negatively. This has been shown in numerous other papers.

The more important conclusion is that those companies from which funds divest in decarbonisation trades reduce emissions faster than other companies. That may be true, but does the analysis support the “causal” claim made in the strapline of the Responsible Investor article. 

At the heart of many academic papers is the old chestnut of correlation versus causation. The authors use lots of fancy sounding statistical jargon to try to persuade the reader that their results have a causal interpretation. But at the heart of the analysis is a high risk of omitted variables. 

What does this mean? An omitted variable is one that causes both the divestment and the carbon reduction, meaning that the carbon reduction is merely correlated with divestment rather than caused by it. The authors do robustness checks for some plausible omitted variables, for example earnings surprises and historic performance. But there are some very plausible explanations that aren’t tested.

Here’s one example. Suppose it becomes clear that a firm is more vulnerable than thought to low carbon transition (e.g. because a competitor emerges or regulation changes). Then shares in that firm may be sold by investors driving the price down but also the firm may subsequently cut emissions more quickly to try to address their vulnerability. A less carbon-efficient European firm facing an increase in the EU carbon price would fall into this category. In this case it is the change in carbon price that has driven both the divestment and the emissions reduction. Divestment and emissions are correlated but with no causal relationship.

Alternatively, investors may identify that a firm is planning to cut carbon emissions inefficiently and sell out. An example might be a company that is subject to an exceptional level of political or climate activist pressure to reduce carbon emissions quicker than economically optimal for the firm. Seeing this coming the investors may sell out driving down the share price. This may even be happening right now with the oil majors. Since the start of 2020 the share prices of Shell and BP have underperformed Exxon, Total, ConocoPhillips and Chevron by 15% to 20%. It may be that investors are looking at the higher stakeholder pressures faced by Shell and BP and think this will damage returns by causing them to take non-economic actions. Indeed over this period we have seen BP pivot its strategy away from oil and Shell come under pressure from a Dutch court case forcing it to cut emissions quicker than it had planned.  That same stakeholder pressure may cause Shell and BP to cut carbon emissions quicker than other firms. In this way an omitted variable of “differential stakeholder pressure on climate change” can drive both the divestment and the carbon reduction, meaning the relationship between the two is one of correlation rather than causation.

Note that in both these examples the divestment (which can be immediate) will come before the carbon reduction (which takes time). Looking at the timeline might imply causation, but actually it is just correlation due to a shared root cause. 

There are ways of separating correlation from causation, but this needs a so-called instrumental variable or natural experiment which in effect creates something close to a randomised control trial.  The analysis in this paper doesn’t have any such mechanism for determining causation, so we can’t be sure to what extent omitted variables are driving the result. We shouldn’t tie ourselves to requiring definitive proof of causality before taking insight from papers, but in this case there are some plausible omitted variables that have not been addressed. 

What if it is causation?

Just because we can identify some plausible omitted variables doesn’t make it certain that there is no causal link between divestment and decarbonisation. It could be that the omitted variables aren’t a problem. Even if this is the case there are some nuances that the paper does not address.

A causal link from divestment to carbon reductions could work through one of two channels. One is the cost of capital channel emphasised by the authors. Divestment reduces the share price, increasing cost of capital, and the firm then takes corrective action to lower their cost of capital again, or alternatively simply stops investing so much in growth because cost of capital is higher - either way carbon emissions are reduced.

However, in a careful piece of work by top authors Berk and Binsbergen, it is shown that credible divestment levels just aren’t enough to influence cost of capital to a meaningful degree. Now this paper hasn’t yet been published in a peer-reviewed journal, but they are very credible authors and their conclusions are consistent with other work that suggests that in many cases flow-induced share price changes don’t change investment levels in affected firms (or at least not by much and only in specific circumstances). 

This means the more likely channel is the share price incentive channel. Divestment leads to a reduction in share price which affects executives' equity holdings, so they have an incentive to react to get the share price back up again. Note that this channel only works if there are practical actions managers can take to reverse the selling pressure. So it is not effective if fund managers are operating blanket sector exclusions (an oil company can't stop being an oil company) but can be effective if fund managers are operating tilting or best in class approaches whereby they stay invested in sectors but skew investment towards the cleanest companies in each sector. The channel works if there are any company specific factors driving the divestment that can be acted upon.  This creates the incentive for the laggards to improve. 

The paper doesn’t differentiate between blanket divestment or tilting or company-specific divestment. But there are some interesting pointers. First, the study focusses on actively managed funds – index funds are more likely to operate blanket screening approaches. Furthermore, the authors found that divestment was strongest in the most actively managed funds. These facts point towards tilting or firm specific divestment being the more likely approach of investors covered by the study, rather than blanket divestment. 

Creeping causality

So, in conclusion, because of omitted variables this paper doesn’t prove a causal link between divestment and carbon reductions. But its findings are at least consistent with the possibility that tilting or best in class approaches can lead to cuts in carbon emissions by providing incentives for sector laggards to improve. However, the paper certainly does not show that blanket divestment approaches reduce emissions. This nuance is not covered in the paper, which, first, links their findings to the divestment approach more generally and, second, claims causality.

The pervasive concern with the paper is the problem of creeping causality. 

As described above the methodology in the paper certainly does not prove causality. There are credible omitted variables that have not been explained away. Indeed, casuality isn’t mentioned at all in the paper’s abstract. In the conclusion to the paper they only go so far as saying:

Overall we are confident that a cautious causal interpretation of the findings is warranted.

Cautious causality isn’t a technical term I’m familiar with, but the sceptic in me interprets it as “if you don’t look too closely we might convince you it’s causal”. 

By the time you get to the YouTube video summarising the results the authors’ confidence in causality has grown. They conclude:

In summary we do find that there is causality between divestment and the reduction of carbon emissions. Investors can therefore have real impact against climate change by divesting.

Hey presto, the Responsible Investor article strapline refers to a “causal link”. 

This paper falls as long way short of proving a causal link between divestment and reduced carbon emissions notwithstanding its publicity. If there is a causal link underlying the data, the facts are most consistent with it being due to tilting, best in class, or firm specific divestment rather than a blanket sector divestment approach. 

But overall this paper doesn’t significantly tilt the scales of evidence in favour of divestment, and certainly not blanket sector divestment. The reporting about the paper is also a useful case study in why we should be sceptical of brief headlines that describe complex research.

The debate continues.