Impact on equities

Based on our findings related to currencies, we investigate the effects of our climate news index on equity returns denoted in USD. Specifically, we examine the behavior of average cumulative returns following days marked by a positive spike in our global index (CAI), taking into account both stock-specific exposure and country-specific conditions.

We use emission intensity as a proxy for stock-specific exposure, defined as emissions in tons of CO2 divided by sales. We source annual data on emission intensity from the Trucost dataset. For country-specific conditions, we use sensitivity coefficients denoted by β, as reported in table 2. Our dataset includes 16,774 firms and totals 23,305,563 firm-day observations.

We estimate the following regression:

where c denotes the country in which the headquarter of the firm is located, i refersto one of our firms, βc represents our equally weighted sensitivity, E is the emission intensity of firm i assessed the year prior to day t, and Top is a dummy variable that equals 1 if the day falls in the set of extreme climate news days. The left-hand side variable is defined as follows:

that is, it is averaged over the time horizon k, it is in percentage, and it is multiplied by the average number of top-days in a year.

Since emission intensity is very persistent at the annual frequency, the composite coefficient is almost time-invariant in our dataset. Consequently, this composite coefficient primarily reflects heterogeneity across different countries and stocks. The literature has proxied the exposure of a firm to climate risk either by examining its emissions or considering geographical factors such as distance from the ocean or latitude.

As in the model, we hypothesize that a firm’s equity returns’ sensitivity to climate news reflects the interaction of its brownness and the degree of investor sensitivity to climate news within a specific country. This analysis is agnostic on whether the valuation adjustment is driven by uncertainty about physical damages or transition risk. Given the low correlation between our βis and vulnerability, we interpret these valuation adjustments as driven mostly by heterogeneous sensitivity to global climate sentiment (discount rate channel).

According to our model, global climate news should play an important role mostly among firms with high emissions in highly sensitive countries (γ2,k < 0). Our estimates confirm this prediction. In figure 11, we depict the estimates of our composite parameter over different daily horizons, k.

In the top portion of the figure, we first sort countries according to their exposure β. The top-decile of the distribution of βs is 2.27 (Saudi Arabia). The bottom-decile is 0.197 (South Africa). We then select the values of carbon emission intensity for a representative green (brown) firm using the 5th (95th) percentile of the emission intensity distribution, which corresponds to the median value within the top-decile (bottom-decile).13 Given these numbers, we can depict the implied estimated composite coefficient and its associated standard error across different daily horizons, k.

The bottom two panels convey the same information as the top panels but from a different perspective. Specifically, the bottom left (right) panel presents the composite coefficients for a representative green (brown) firm in a low-β country and in a high-β country.

Our analysis yields several significant findings. Firstly, firms situated in countries with moderate sensitivity to climate sentiment news experience a more moderate loss of value. This effect holds for both high- and low-emission firms. Conversely, firms in countries with high β values suffer a more severe loss of value, an effect that is particularly pronounced for high-emission firms. Across all these scenarios, the negative impact on cumulative returns is very persistent.

Type of news. Following our analysis of currency markets, we examine whether these results are primarily driven by top-attention days in which news focuses on either physical or transition risk. Specifically, we estimate the following specification: