Does market sentiment drive stock returns? – Underground Bank

gerardo martinez

In 1936, John Maynard Keynes famously coined the term ‘Animal Spirits’ to illustrate how people make decisions based on impulse, overlooking the benefits and drawbacks of their actions. To what extent are environmental, social and governance (ESG) asset prices driven by market participant sentiment, as opposed to economic fundamentals? To answer this question, I make use of natural language processing (NLP) and an original corpus of tweets to capture market sentiment around climate change. When estimating a factor model, I find that sentiment is associated with immediate returns of stock indices related to climate change. These results are strongest for the days with the most extreme yields. Market sentiment could be particularly helpful in explaining large movements in ESG asset prices.

Upcoming and Upcoming: ESG Assets

ESG assets are portfolios of stocks and bonds whose underlying companies satisfy environmental, social, and governance factors. They represent a rapidly growing part of asset management portfolios: according to Bloomberg IntelligenceESG exchange traded funds (ETFs) cumulative assets exceeded $360 billion in 2021, and that number is expected to reach $1.3 trillion by 2025.

The growing importance of these assets makes ESG returns and volatility an important subject of study. Firstly, we would like to measure to what extent market sentiment around ESG can drive asset prices. And if the effect is significant, ESG assets could act as triggers or amplifiers of stress in financial markets if there were to be a significant adverse turn in sentiment.

To the best of my knowledge, this post is the first to use a climate change sentiment gauge, built using NLP tools and an original sample of tweets, as input to models explaining asset performance. I look at three stock market indices designed to measure the performance of companies in clean energy-related businesses globally and in the UK:

• The S&P Global Clean Energy Index (GCEI).
• The FTSE Environmental Opportunity Index for Alternative and Renewable Energy (EORE).
• The FTSE Environmental Opportunities UK (EOUK) Index.

Graph 1 plots the performance of the indices, which move closely with political events related to climate change policy.

Chart 1: Climate Change Related Stock Indices and General Benchmarks (01/01/2016 = 100)

Sources: Bloomberg and author’s calculations.

All About Attitude: Measuring Market Sentiment

To construct a measure of market sentiment around climate change, I draw from the Twitter APIs an original sample of more than 700,000 tweets filtered by keywords closely related to climate change. I’m restricting my search to English-language tweets posted in the US and UK. I follow a standard pipe to remove duplicates, clean up, and pre-process the text of each tweet.

I apply two existing pre-trained natural language processing tools (INSTINCT Y VADER) to the resulting data set. Chart 2 shows the resulting tweet counts, divided into positive and negative sentiment according to VADER. It also shows the average FLAIR score for each day in the sample. All three metrics are normalized using the Z score.

There is a strong correlation between the three indicators. The spikes in the count of negative and positive tweets closely track the extreme values ​​of the mean FLAIR score. These extremes are often linked to political developments around climate change.

Chart 2: Measures of market sentiment around climate risk

Source: Author’s calculations.

Linking Market Sentiment and Overnight Yields

To assess the extent to which market sentiment influences ESG asset returns, I estimate a factor regression linking ESG indices returns to VADER and FLAIR scores, controlling for additional factors. These factors include the price-earnings ratio of each index, the difference between 20-year and 30-day government bonds (time horizon risk), investment-grade corporate bond spreads (confidence risk), and yields. of a reference index. index (the S&P 500 in the case of the S&P GCEI, and the FTSE 100 for the FTSE EORE and FTSE EOUK indices).

Table A shows that the effect of market sentiment on returns is statistically significant, but modest. The effect is especially clear for the FTSE EORE index. A 1 standard deviation increase in the count of positive tweets is associated with a increase in daily EORE returns of 10 basis points. Conversely, a 1 standard deviation increase in the count of negative tweets is associated with a decrease in daily returns of 14 basis points. For comparison, the unqualified standard deviation of EORE returns in the sample is 76 basis points.

Note that the effect of positive and negative tweet counts is similar, but opposite in sign. This is encouraging, as it is natural to interpret market sentiment as the difference between positive and negative individual sentiment.

The estimated effects on the S&P GCEI are of similar magnitude and direction, although the coefficient on positive tweet count is not significant at the 10% significance level. However, I find no significant effect of positive and negative tweet counts on FTSE EOUK returns. One possible explanation is that the FTSE EOUK index captures UK companies. By contrast, most of the tweets in the sample were located in the US and therefore may not capture climate change sentiment specific to local UK factors.

Focusing our analysis on the most extreme 10% of performers (highest and lowest) yields larger coefficients on VADER sentiment metrics. For example, on US election day 2016, I estimate that market sentiment lowered the returns of the FTSE EORE and S&P GCEI by around 30 basis points, based on the difference between the negative and positive tweet counts. The regression on the most extreme sample estimates that this effect is 300 basis points, which would explain 60% and 85% of the negative returns observed, respectively.

While FLAIR and VADER scores do react to important events, they are likely to contain a significant amount of noise on a day-to-day basis. Adding periods with smaller returns to the sample is likely to add noisy FLAIR and VADER observations, reducing the regression estimates to zero.

The opposite is true for FLAIR sentiment scores. Taking the regression results at face value, days with negative market sentiment are associated with higher returns. But on days with extreme returns, the sentiment effect measured by FLAIR scores disappears. Given the strong correlation between FLAIR and VADER scores, sentiment is likely to be captured through VADER scores, with FLAIR estimates driven primarily by noise.

Table A: Effect of Market Sentiment on ESG Returns

***p < 0.01: significant coefficient at the 1% level **p < 0.05 *** p<0.10

Table A also shows the same set of coefficients, estimated in returns five periods ahead. Neither coefficient is statistically significant. This is encouraging: we would expect changes in market sentiment to be quickly factored into investors’ information set and market prices to adjust accordingly.

Market sentiment over time

In order to shed light on the dynamic relationship of ESG returns and market sentiment (as well as the other factors), I ran a vector autoregression (WAS). I am particularly interested in the feed-through of shocks in market confidence indicators to ESG returns. To this end, Graph 3 presents the breakdown of the variance of the estimated model for each of the three ESG indices. The variance decomposition is computed over the forecast errors over a 20-day horizon and then averaged for ease of exposure.

The three indicators of market sentiment together explain a very small fraction of the variance of the forecast error. Combined with the results of the regressions for one-day and five-day returns, these findings suggest that market sentiment shocks do not explain returns beyond a one-day time horizon. One interpretation is that market sentiment shocks typically occur around major political events (see Chart 3) and that market participants can quickly discount their effects, so they have little effect on returns over a longer horizon. long.

Graph 3: Decomposition of the variance, average over the forecast horizon of 20 days

Source: Author’s calculations.

conclusions

The results of this analysis suggest that market sentiment on climate change is associated with the overnight returns of ESG stock indices. The estimated effect is modest in magnitude, but it is especially clear and strong when the analysis is limited to the periods with the most extreme returns. However, it is not universal across all indices and sentiment indicators. And a dynamic analysis shows that exogenous shocks to market sentiment do not explain returns beyond a one-day horizon.

However, these results have several implications for financial market regulators. First, they open the door to enriching asset price forecasting models by including additional inputs such as fundamentals or market sentiment and new tools such as machine learning models. Second, regulators will be able to take advantage of the new dataset on market sentiment and asset prices to study patterns of market reaction to changes in sentiment, such as procyclical asset purchases or asset reallocations.

Gerardo Martínez works in the Bank’s Capital Markets Division.

If you would like to get in touch with us, please email bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator and will only be posted when a full name is provided. Bank Underground is a blog for Bank of England staff to share views that challenge, or support, prevailing orthodox policies. The views expressed here are those of the authors and not necessarily those of the Bank of England or its policy committees.