A large-scale analysis of the heterogeneity of markets' reactions to the disclosure of nonfinancial information

Figures & data

Table 1. Number of announcements.

	Earnings announcement	M&A, Offerings, ESG ratings	Nonfinancial disclosure
Region
Asia-Pacific	7509	12,886	7509
Europe	7922	15,800	7922
Latin America and Caribbean	941	1222	941
United States and Canada	5531	15,531	5531
Sector
Communication Services	1204	2908	1204
Consumer Discretionary	2609	4207	2609
Consumer Staple	1756	2980	1756
Energy	1506	2636	1506
Financial	2047	9203	2047
Health Care	1318	2508	1318
Industrial	4388	7210	4388
Information Technology	1554	2803	1554
Material	2798	4396	2798
Real Estate	1341	3957	1341
Utilities	1382	2631	1382
Year of announcement
2009	821	460	821
2010	1312	3093	1312
2011	1799	3885	1799
2012	1762	4318	1762
2013	1789	4296	1789
2014	1924	4386	1924
2015	2003	4049	2003
2016	2220	4766	2220
2017	2388	5336	2388
2018	2693	5193	2693
2019	3192	5657	3192
Total	21903	45,439	21,903

The table shows the count of the number of announcements by geographies, sectors and years. Table S1 shows the count by country. The first and third columns are identical because we have matched the earnings and nonfinancial disclosure announcements so that a firm is in the universe in a particular year if we have data on both.

Table 2. Statistics of the nonfinancial disclosure.

	Integrated report (%)	External audit (%)	GRI guidelines (%)	Global activities (%)
Region
Asia-Pacific	33	84	97	79
Europe	43	81	96	94
Latin America and Caribbean	34	70	99	94
United States and Canada	12	55	97	95
Sector
Communication Services	36	83	96	92
Consumer Discretionary	33	77	96	85
Consumer Staple	31	73	97	91
Energy	26	71	97	96
Financial	35	81	97	90
Health Care	34	71	95	92
Industrial	33	76	95	86
Information Technology	24	75	98	85
Material	27	72	98	92
Real Estate	37	72	95	92
Utilities	26	80	99	94
Year of announcement
2009	27	61	94	93
2010	40	56	80	94
2011	39	66	84	92
2012	32	83	99	92
2013	32	83	99	91
2014	30	84	99	90
2015	28	82	100	90
2016	31	79	99	89
2017	30	80	100	88
2018	31	79	99	88
2019	31	79	100	88

The table shows the composition by geographies, sectors, and years, of the nonfinancial disclosures in terms of fraction of integrated reports, audited reports, reports that follow GRI guidelines, and reports that disclose on global activities. The fraction is calculated with respect to the total numbers of reports shown in Table 1 (third column) after removing missing values of the variables as explained in Section 2. Table S2 shows the composition by country.

Table 3. Summary statistics of the sample.

	25%	50%	75%
Size	7.30	8.24	9.23
Invested capital over Revenue	0.73	1.18	2.01
Tangibility	0.07	0.21	0.42
Profitability	0.06	0.10	0.16
Emissions Intensity	6.67	19.22	56.07
Idiosyncratic volatility	0.05	0.06	0.08
Environmental Score	3.50	5.10	6.70
Social Score	3.40	4.60	5.70
Governance Score	4.00	5.39	6.81

Size is the log of revenue (SALE) adjusted for inflation. Tangibility is property plant and equipment (PPENT) over total book asset (AT). Profitability is Earnings before interests, taxes, depreciation and amortization (EBITDA) over the previous year book assets. Emission intensity is direct plus first-tier indirect emissions (in tCO${}_2$e) over revenue. Idiosyncratic volatility is the standard deviation of the residuals of the factor model presented in Section 2. The ESG scores are from MSCI, S&P and Refinitiv.

Figure 1. Overview of the study design. The figure shows the basic structure of our study design. For each firm i and nonfinancial disclosure (red in (a)) that does not overlap with other announcements (cyan in (a)) we compute the average change in analysts' forecast of earnings prior and after the event during a 30 days event window ($\mathcal{R}$). We then compute the same statistics over dates with no significant event (gray) and we create a null distribution (gray in (b)). Then, we compute the probability that the changes in earnings' forecasts in the null model were greater than changes occurred in the event ($\mathcal{S}$). Finally, we plot the distribution of this statistics over the whole population. (c,d) show two hypothetical shapes of this distribution. We conclude that an effect exists in the population if the distribution of this statistics is not uniform, as shown in (c).

Figure 2. The impact of nonfinancial disclosure on analysts' revisions of earnings. (a) shows the distribution of the statistics $\mathcal{S}$ as in Figure 1(c). The panel illustrates that there is a significant effect across the population. In (b–d), we look at the event-specific effects. Specifically, in (b), we show the distribution of the magnitude of the statistically significant and not significant effects, compared to the null model. (c) shows the evolution of the magnitude of the effects in time. Finally, (d) shows the fraction of integrated reports in the population. Overall, the figure shows that analysts revise their earnings' estimates following the disclosure of nonfinancial information. The effect of the revision is concentrated in a small fraction of the population and integrated reporting is associated with a higher likelihood of observing a statistically significant effect.

Figure 3. Analysts' reactions and information asymmetry. The figure shows the ratio (a) and the level (b) of idiosyncratic volatility in the effect and no effect population. Overall, the figure shows that the difference between the volatility in the two populations is statistically significant and persists outside the estimation window (2020).

Table 4. Probit model.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Environmental score	$-{0.07}^{\ast }$	$-0.03$	$-0.05$	$-0.04$	$-0.04$	$-0.04$	$-0.03$	$-0.02$
Social score	0.05	0.01	0.02	0.02	0.02	0.02	0.02	0.02
Governance score	0.02	$-0.01$	$-0.01$	0.0	0.0	0.0	0.01	0.0
Size		$-{0.32}^{\ast \ast \ast }$	$-{0.33}^{\ast \ast \ast }$	$-{0.33}^{\ast \ast \ast }$	$-{0.34}^{\ast \ast \ast }$	$-{0.34}^{\ast \ast \ast }$	$-{0.33}^{\ast \ast \ast }$	$-{0.25}^{\ast \ast \ast }$
Tangibility			$-{0.09}^{\ast \ast }$	$-{0.09}^{\ast \ast }$	$-{0.09}^{\ast \ast }$	$-{0.09}^{\ast \ast }$	$-{0.1}^{\ast \ast }$	$-{0.1}^{\ast \ast }$
Profitability				$-{0.11}^{\ast \ast }$	$-{0.12}^{\ast \ast \ast }$	$-{0.12}^{\ast \ast }$	$-{0.11}^{\ast \ast }$	$-{0.1}^{\ast \ast }$
Invested capital over revenue					$-0.16$	$-0.16$	$-0.13$	$-0.11$
Emission intensity						0.0	0.01	0.04
Idiosyncratic volatility							${0.1}^{\ast \ast }$	${0.11}^{\ast \ast }$
Integrated report								${0.23}^{\ast \ast \ast }$

The table shows the coefficients of the Probit model for the probability of belonging to the effect population (Equation (4(4) $e_{i,t}=\overrightarrow{\alpha }{\overrightarrow{X}}_{i,t}+\beta {\mathbb{1}}_{\mathrm{IR},i,t}+\overrightarrow{\gamma }{\overrightarrow{\mathcal{F}}}_i+{ϵ}_{i,t}(4)$(4) )). All the control variables are standardized before running the regressions so that the magnitude of the coefficients are comparable among each others. ${}^{\ast }$, ${}^{\ast \ast }$, ${}^{\ast \ast \ast }$ denote statistical significance at 10%, 5% and 1%, respectively. In each model, we control for year, geography and sector fixed effects and we adjust for self-selectivity with the Heckman correction.

Table 5. Robustness tests.

	R.T. 1	R.T. 2	R.T. 3	R.T. 4	R.T. 5	R.T. 6	R.T. 7	R.T. 8
Environmental score	$-0.02$	$-0.01$	$-{0.09}^{\ast }$	0.01	$-0.15$	$-0.02$	0.0	0.06
Social score	0.03	0.02	$-0.02$	${0.09}^{\ast \ast }$	0.06	0.02	0.05	$-0.05$
Governance score	0.01	0.01	0.01	0.03	0.08	$-0.0$	0.04	$-0.01$
Size	$-{0.18}^{\ast \ast \ast }$	$-{0.23}^{\ast \ast \ast }$	$-{0.25}^{\ast \ast \ast }$	$-{0.22}^{\ast \ast \ast }$	$-{0.17}^{\ast \ast \ast }$	$-{0.25}^{\ast \ast \ast }$	$-{0.15}^{\ast \ast \ast }$	$-{0.32}^{\ast \ast \ast }$
Tangibility	$-{0.08}^{\ast }$	$-{0.1}^{\ast \ast }$	$-{0.11}^{\ast \ast }$	$-{0.13}^{\ast \ast }$	$-{0.13}^{\ast \ast }$	$-{0.11}^{\ast \ast }$	$-0.06$	$-{0.09}^{\ast \ast }$
Profitability	$-{0.11}^{\ast \ast }$	$-{0.08}^{\ast \ast }$	$-{0.08}^{\ast }$	$-{0.08}^{\ast }$	$-0.09$	$-{0.1}^{\ast \ast }$	$-{0.09}^{\ast \ast }$	$-{0.11}^{\ast \ast }$
Invested capital over revenue	$-0.12$	$-0.06$	$-0.1$	$-0.06$	$-0.05$	$-0.1$	$-0.05$	$-{0.13}^{\ast }$
Emission intensity	0.03	0.03	0.06	0.05	$-0.02$	0.04	0.02	${0.14}^{\ast \ast \ast }$
Idiosyncratic volatility	${0.11}^{\ast \ast }$	${0.08}^{\ast }$	${0.1}^{\ast \ast }$	0.08	${0.1}^{\ast }$	${0.1}^{\ast \ast }$	${0.09}^{\ast }$	${0.09}^{\ast \ast }$
Integrated report	${0.24}^{\ast \ast \ast }$	${0.21}^{\ast \ast \ast }$	${0.19}^{\ast \ast \ast }$	${0.22}^{\ast \ast \ast }$	${0.17}^{\ast \ast \ast }$	${0.23}^{\ast \ast \ast }$	${0.24}^{\ast \ast \ast }$	${0.22}^{\ast \ast \ast }$

The table shows the coefficients of the Probit model (Equation (4(4) $e_{i,t}=\overrightarrow{\alpha }{\overrightarrow{X}}_{i,t}+\beta {\mathbb{1}}_{\mathrm{IR},i,t}+\overrightarrow{\gamma }{\overrightarrow{\mathcal{F}}}_i+{ϵ}_{i,t}(4)$(4) )) under different specifications as discussed in Section 3.1.2. Specifically in R.T. 1, we do not correct for self-selectivity; in R.T. 2, we include data from 2020; in R.T. 3-5, we use ESG scores from different data providers; in R.T. 6 we control for country fixed effects, in R.T. 7, we control for historical values of the control variables and in R.T. 8 we use a different proxy for the dependent variable. ${}^{\ast }$, ${}^{\ast \ast }$, ${}^{\ast \ast \ast }$ denote statistical significance at 10%, 5% and 1%, respectively. In each model, we control for year, geography and sector fixed effects (except R.T. 6 where we control for country fixed effects) and we adjust for self-selectivity with the Heckman correction (except for R.T. 1).

Figure 4. Effect on equity value. (a) shows the effect of disclosure of nonfinancial information on equity values at the population level. The x-axis is the number of days prior and after the event. The y-axis is the cumulative abnormal returns (in basis points). The table on the right shows the results of the test to assess the statistical significance of the effects (at multiple time windows). (b) shows the effect of disclosure after publications with positive, negative and no effect on the revision of earnings. Overall, the table and the figure in (b) show that the disclosure of non financial information has a significant impact on firms' market value of equity but the effect is not homogeneous across the population. (c,d) shows the effect after the publication of standalone and integrated reports, respectively. The panels show that the statistically significant impact of CSR reporting on equity values is strongly driven by the publication of integrated reports.

Figure 5. No effect on cost of capital in our sample. (a) shows the distribution of the statistics ${\mathcal{S}}_{WACC}$, which, differently from Figure 2(a), is approximately uniform in [0,0.5]. (b) shows the distribution of the effect of disclosure on cost of capital for the events when analysts submit a statistically significant revision (blue) and a not statistically significant revision (green). The panel shows that none of the two distributions is significantly different from the null model. Overall, the figure shows that the disclosure of nonfinancial information has no measurable effect on cost of capital in our sample.