ESG Rating Dispersion and Expected Stock Return in China

Figures & data

Table 1. Summary statistics.

	N	Mean	SD	Pearson correlations
	(1)	(2)	(3)	(4)	(5)	(6)
Panel A: Summary statistics and correlation coefficients
				SynTao	Sino	CASVI
SynTao	16,670	4.04	1.38
Sino	16,670	4.33	1.69	0.25
CASVI	9,165	5.76	1.83	0.36	0.35
Wind	14,222	5.84	1.37	0.45	0.30	0.38
		P5	P25	P50	P75	P95
Panel B: Statistical distribution of key variables
SynTao		2.00	4.00	4.00	5.00	6.00
Sino		1.00	4.00	5.00	5.00	6.00
CASVI		4.00	4.75	5.50	6.00	8.80
Wind		4.00	5.00	6.00	6.00	8.00
AcrossAll_Disp		0.47	0.58	1.34	1.88	2.65
Pairwise_Disp		0.35	0.47	1.12	1.60	2.36

Panel A reports the key statistical characteristics of company-level ESG ratings released by four major ESG rating agencies in China during the sample period ranging from July 2016 to January 2022, including the number of observations, mean, standard deviation, and Pearson correlation coefficient. Panel B reports the primary quantiles of two prominent measures of ESG rating dispersion. The measure calculated based on Christensen, Serafeim, and Sikochi (2022) and Brandon, Krueger, and Schmidt (2021) is referred to as “AcrossAll_Disp,” while the measure computed according to Avramov et al. (2022) is denoted as “Pairwise_Disp.”

Table 2. Performance of portfolios sorted by ESG rating dispersion.

	SR	CAPM	FF5+UMD	FF5+IML	FF5+IML(RIV-GMM)
Panel A: 1-Month Holding Period
Low	0.341	1.197***	1.388***	1.485***	1.545***
		(2.97)	(2.90)	(4.15)	(3.05)
High	−0.031	−0.538	−0.261	−0.392*	−0.185
		(−1.39)	(−1.28)	(−1.82)	(−0.75)
Low Minus High	0.929	1.735***	1.649***	1.877***	1.731***
		(4.14)	(3.66)	(5.88)	(2.87)
Panel B: 3-Month Holding Period
Low	0.180	0.844***	0.577**	0.802***	0.553**
		(2.93)	(2.26)	(2.65)	(2.06)
High	0.001	−0.102	−0.078	−0.173	−0.078
		(−0.41)	(−0.53)	(−1.12)	(−0.53)
Low Minus High	0.325	0.945***	0.656**	0.975***	0.631**
		(2.80)	(2.47)	(2.79)	(2.29)

The monthly value-weighted return of portfolios with holding periods of one and three months are reported in Panel A and Panel B, respectively. All portfolios are formed using firm-level market capitalizations as the individual weight. The Sharpe ratio of monthly excess returns over the risk-free rate are reported in monthly units.

“CAPM-Alpha” is the risk-adjusted return calculated by adjusting exposure to the market factor (MKT), with the following regression specification:

$r_{i,t}-r_{f,t}={\alpha }_t+{\beta }_{t}MK{T}_t+{\in }_{i,t}$

“FF5+UMD-Alpha” is the risk-adjusted return calculated by adjusting exposure to the market factor (MKT), the size factor (SMB), the book-to-market factor (HML), the profitability factor (RMW), the investment factor (CMA) and the momentum factor (UMD), with the following regression specification:

$r_{i,t}-r_{f,t}={\alpha }_t+{\beta }_t^{1}MK{T}_t+{\beta }_t^{2}SM{B}_t+{\beta }_t^{3}HM{L}_t+{\beta }_t^{4}RM{W}_t+{\beta }_t^{5}CM{A}_t+{\beta }_t^{6}UM{D}_t+{\in }_{i,t}$

“FF5+IML-Alpha” is the risk-adjusted return calculated by adjusting exposure to the Fama-French five-factor model (Fama and French 2018) and the illiquidity premium factor (IML), with the following regression specification:

$r_{i,t}-r_{f,t}={\alpha }_t+{\beta }_t^{1}MK{T}_t+{\beta }_t^{2}SM{B}_t+{\beta }_t^{3}HM{L}_t+{\beta }_t^{4}RM{W}_t+{\beta }_t^{5}CM{A}_t+{\beta }_t^{6}IM{L}_t+{\in }_{i,t}$

Due to the observation provided by Adrian et al (2017, 2018). that traditional illiquidity measures are endogenous, we employ RIV GMM to tackle this issue following Racicot, Rentz, and Théoret (2018). The alpha calculated through this method is denoted as “FF5+IML(RIV-GMM)-Alpha.”

We report t-statistics in parentheses using Newey and West (1987) standard errors with three lags, where the number of lags is based on autocorrelations in monthly stock returns. The asterisks *, **, *** denote statistical significance at the 10%, 5% and 1% level respectively.

Figure 1. Spread in alpha between bottom and top quintile portfolios.

Since July 2017, we have constructed quintile portfolios based on the 12-month average ESG rating dispersion and track their average monthly value-weighted CAPM-alpha for the subsequent three months. We rebalance portfolios at the end of each rating-release month, thereby generating a time series of excess returns from August 2017 to April 2022. Panel A plots the time series of the spread in monthly CAPM-alpha between the two extreme quintile portfolios (bottom versus top). Panel B plots the cumulative spread in CAPM-alpha between the two extreme quintile portfolios.

Figure 2. Buy-and-hold return from bottom and top quintile portfolios..

Since July 2017, we have formed quintile portfolios and calculated buy-and-hold returns over holding periods ranging from 1 to 12 months without mid-way selling. We averaged these returns across all bottom (top) portfolios for each holding period and presented them as dots with connecting lines. Panel A shows results for overlapping holding periods, and Panel B displays the results for non-overlapping holding periods. We adopt a simple technique to avoid overlapping return observations. For a given holding period is h, we only consider observations gathered at time t = 0, h, 2h, 3h, … In this way, the return from time 0 to h does not overlap with the return from time h to 2h.

Table 3. Main regression and robustness check.

	1-Month Holding Period			3-Month Holding Period
	(1)	(2)	(3)	(4)	(5)	(6)
Panel A: Main regression
AcrossAll_Disp	−0.651***	−0.796***		−0.382***	−0.526***
	(−2.74)	(−2.69)		(−2.73)	(−3.04)
Pairwise_Disp			−0.852***			−0.550***
			(−2.58)			(−2.84)
Trans		−0.349	−0.345		−0.193	−0.171
		(−1.30)	(−1.28)		(−0.57)	(−1.10)
Control Variables	Yes	Yes	Yes	Yes	Yes	Yes
Firm F.E.	Yes	Yes	Yes	Yes	Yes	Yes
Year-Month F.E.	Yes	Yes	Yes	Yes	Yes	Yes
R^2	0.274	0.279	0.279	0.251	0.250	0.250
Observations	12,258	10,991	10,991	36,894	33,207	33,207
		AcrossAll_Disp			Pairwise_Disp
		(1)	(2)		(3)	(4)
Panel B: IV−2SLS
Dispersion			−4.816***			−5.643***
			(−2.59)			(−2.62)
Media_IV		−0.026***			−0.023***
		(−3.78)			(−3.79)
Reports_IV		0.066***			0.053***
		(8.49)			(7.71)
Trans_IV		0.079***			0.071***
		(7.40)			(7.39)
Tan_IV		−0.010***			−0.009***
		(−8.35)			(−8.22)
Dispersion_IV		−0.016**			−0.015*
		(−1.99)			(−1.95)
F-statistics			38.72			41.31
Hansen-J (p-value)			0.34			0.32

We regress stock monthly returns on the average ESG rating dispersion over the past 12 months, controlling for both firm and year-month fixed effects in all regressions, and report robust standard errors. Furthermore, we include known stock-level return predictors, including average ESG rating (ESG_Rating), market capitalization (ME), book-to-market ratio (BM), leverage (Lev), return on assets (ROA), revenue growth rate (Growth), total years since being listed (Age), systematic risk (Beta), momentum (Mom), and liquidity (Liquid). Panel A examines the cases based on 1-month and 3-month holding periods, respectively. Information transparency (Trans) is included in columns (3) and (6) to further mitigate the potential influence of omitted variables. Panel B employs an instrumental variable two-stage regression (IV−2SLS) to address endogeneity concerns.

Table 4. Examination of the mechanism based on institutional investor demand.

	CAPM	FF5+UMD	FF5+IML	FF5+IML(RIV-GMM)
Panel A: Portfolio sorts using subsample
Low Ownership	0.829	0.636	0.969	0.534
	(1.46)	(1.37)	(1.41)	(1.12)
High Ownership	1.536**	1.010**	1.405***	0.991**
	(2.30)	(2.16)	(2.84)	(2.11)
	Low Ownership		High Ownership
	(1)	(2)	(3)	(4)
Panel B: Regression using subsample
AcrossAll_Disp	−0.555		−0.642***
	(−1.46)		(−2.62)
Pairwise_Disp		−0.500		−0.722***
		(−1.55)		(−2.82)
Control Variables	Yes	Yes	Yes	Yes
Firm F.E.	Yes	Yes	Yes	Yes
Year-Month F.E.	Yes	Yes	Yes	Yes
R^2	0.287	0.286	0.264	0.264
Observations	11,389	11,389	10,290	10,290

In Panel A, the sample is initially split into two subgroups - “Low Ownership” (below the median) and “High Ownership” (above the median) - based on institutional ownership at the end of each rating-release month since July 2017. These subgroups are then independently sorted into 3x3 portfolios based on the mean of the stock-level ESG scores assigned by the four raters for the current month and the ESG rating dispersion over the past twelve months. This 2x3x3 grouping approach is employed to avoid insufficient observations within any individual portfolio. We track the value-weighted returns over the next subsequent three months, and employ the same risk-factor models as Table 2 to implement risk-adjustment. For brevity, we only report the alpha spread between the low- and high-dispersion portfolios within high-score stocks, which is our primary focus. The t-statistics displayed in parentheses are corrected using Newey and West (1987) with three lags. Numbers with “*,” “**,” and “***” are significant at the 10%, 5%, and 1% levels, respectively. Panel B divides the sample into three subgroups, namely “Low,” “Medium,” and “High,” based on institutional ownership at the 30th and 70th percentiles. The subgroups are then regressed separately using the same specifications as the main regression in Table 3. For simplicity, we only report the regression results for the extreme subgroups.

Table 5. Examination of the mechanism based on belief dispersion.

	CAPM	FF5+UMD	FF5+IML	FF5+IML(RIV-GMM)
Panel A: Portfolio sorts using subsample
Pessimistic	0.155	0.197	0.377	0.216
	(0.33)	(0.43)	(0.70)	(0.46)
Optimistic	1.972***	1.312***	1.969***	1.306***
	(3.06)	(2.60)	(3.65)	(2.66)
Panel B: Portfolio sorts using subsample
Low Mispricing	0.282	0.101	0.952*	0.093
	(0.40)	(0.16)	(1.84)	(0.15)
High Mispricing	1.841**	1.428**	1.628**	1.412**
	(2.31)	(2.02)	(2.37)	(2.01)
	PessimisticLow Mispricing Low Mispricing	Pessimistic High Mispricing	Optimistic Low Mispricing	Optimistic High Mispricing
	(1)	(2)	(3)	(4)
Panel C: Regression using subsample
AcrossAll_Disp	0.167	0.004	−1.232	−4.184***
	(0.22)	(0.00)	(−1.26)	(−2.86)
Control Variables	Yes	Yes	Yes	Yes
Firm F.E.	Yes	Yes	Yes	Yes
Year-Month F.E.	Yes	Yes	Yes	Yes
R^2	0.505	0.397	0.494	0.399
Observations	1,920	2,068	1,752	1,791

In Panel A, we use the average analyst recommendation ratings of individual stocks (using the most recent recommendation ratings before rebalancing portfolios) to proxy for investor optimism, and the sample is initially split into two subgroups - “Pessimistic” (below the median) and “Optimistic” (above the median).In Panel B, we divide the sample into two subgroups - “Low Mispricing” (below the median) and “High Mispricing” (above the median) - based on stock-level R-squared obtained from running CAPM on stock returns. Subsequently, we perform a 3 × 3 independent double sorts based on ESG scores and dispersion, following the methodology in Table 4 Panel A. This generates 18 (2 × 3 × 3) portfolios in Panel A and Panel B. For brevity, we only display the alpha of the LMH portfolio within high-score stocks. The t-statistics displayed in parentheses are adjusted using Newey and West (1987) with three lags. Numbers with “*,” “**,” and “***” are significant at the 10%, 5%, and 1% levels, respectively. In Panel C, we initially partition the sample into three subgroups based on investor optimism at the 30th and 70th percentiles - “Pessimistic,” “Neutral,” and “Optimistic.” Subsequently, within each subgroup, we further segregate the sample into another three subsets based on the 30th and 70th percentiles of the mispricing proxy - “Low Mispricing,” “Middle Mispricing,” and “High Mispricing.” This leads to the creation of nine (3x3) subsamples. To be concise, we solely present the regression outcomes for the extreme subgroups, while maintaining the same specifications as the main regression shown in Table 3.

Table

Variables	Descriptions	Details
AcrossAll_Disp	Dispersion Proxy1	Sample standard deviation of all firm-level available ratings
Pairwise_Disp	Dispersion Proxy2	Average pairwise standard deviation of ESG ratings provided by each rater-rater pair for a given firm.
ESG_Rating	Average ESG Rating	Arithmetic mean of all available firm-level ESG ratings.
ME	Market Capitalization	The logarithm of stock price times shares outstanding.
BM	Book-to-Market	Book equity divided by market capitalizations.
Lev	Leverage	Total debt divided by total assets.
ROA	Return on Assets	Revenues minus costs of goods sold divided by total assets.
Growth	Revenue Growth	Growth rate of revenue.
Age	Firm Age	Total years since firm gets listed.
Beta	Systematic Risk	Computed from the most recent 250 daily return observations.
Mom	Momentum	Cumulative returns of the most recent 6 months.
Liquid	Liquidity	The logarithm of Amihud illiquidity proxy.
Trans	Transparency	Firm-level transparency ratings issued by the Shanghai Stock Exchange and the Shenzhen Stock Exchange are converted to scores using the following rules: “A” = 4, “B” = 3, “C” = 2, and “D” = 1.
Media	Media Attention	Total company mentions in news headlines.
Reports	Reports Attention	Number of analyst reports covering a given firm.
Tan	Tangible Asset Ratio	Tangible assets divided by total assets.
Mispricing	Mispricing Degree	The R-squared obtained from running CAPM on stock returns.
Optimism	Investor Optimism	Stock-level analyst recommendation ratings are converted to scores using the following rules: “Buy” = 5, “Outperform” = 4, “Hold” = 3, “Underperform” = 2, and “Sell” = 1.
Risk factors
MKT	Market factor	Excess return on market portfolio over risk-free rate.
SMB	Size factor	The small minus big premium (Fama and French 1993).
HML	Value factor	The high book-to-market minus the low book-to-market premium (Fama and French 1993).
UMD	Momentum factor	The winner minus loser premium (Carhart 1997).
RMW	Profitability factor	The robust minus weak premium (Fama and French 2018).
CMA	Investment factor	The conservative minus aggressive premium (Fama and French 2018).
IML	Illiquidity Factor	The illiquid-minus-liquid premium (Amihud et al. 2015).

Table

SynTao Green Finance		Sino-Securities Index		CASVI		Wind
A	10	AAA	10	AA+	10	AAA	10
A-	8	AA	9	AA	9.6	AA	9
B+	6	A	8	AA-	9.2	A	8
B	5	BBB	6	A+	8.8	BBB	6
B-	4	BB	5	A	8.4	BB	5
C+	2	B	4	A-	8	B	4
C	0	CCC	2	BBB+	6	CCC	0
		CC	1	BBB	5.75
		C	0	BBB-	5.5
				BB+	5.25
				BB	5
				BB-	4.75
				B+	4.5
				B	4.25
				B-	4
				CCC	2
				CC	1
				C	0

We utilize ESG rating data issued by the four major Chinese ESG rating agencies available on Wind Financial Terminal. All these rating agencies adopt a letter-grade rating system. After conducting statistical analysis on the overall sample, we found that these four rating agencies actually use three major grade levels, namely A, B, and C, with further subgrades such as “BBB” or “B+” under each major grade. Considering that investors are primarily concerned about which major grade (i.e., A, B or C) a letter-type rating belongs to and then subsequently focus on the specific subgrades under the major grades, we establish the following rules to convert letter-type ratings to numerical scores: First, for each rating agency’s subgrades of major grade A, the highest and lowest subgrades are assigned scores of 10 and 8, respectively. The remaining subgrades of major grade A are evenly distributed into the score range of 8–10. Second, for each rating agency’s subgrades of major grade B, the highest and lowest subgrades are assigned scores of 6 and 4, respectively. The remaining subgrades of major grade B are evenly distributed into the score range of 4–6. Finally, for subgrades of major grade C, set the upper limit of the score to 2 and the lower limit to 0, and repeat the same process.

Christensen, D. M., G. Serafeim, and A. Sikochi. 2022. Why is corporate virtue in the eye of the beholder? The case of ESG ratings. The Accounting Review 97 (1):147–75. doi:10.2308/TAR-2019-0506.

 Web of Science ®Google Scholar

Brandon, R. G., P. Krueger, and P. S. Schmidt. 2021. ESG rating disagreement and stock returns. Financial Analysts Journal 77 (4):104–27. doi:10.1080/0015198X.2021.1963186.

 Web of Science ®Google Scholar

Avramov, D., S. Cheng, A. Lioui, and A. Tarelli. 2022. Sustainable investing with ESG rating uncertainty. Journal of Financial Economics 145 (2):642–64. doi:10.1016/j.jfineco.2021.09.009.

 Web of Science ®Google Scholar

Fama, E. F., and K. R. French. 2018. Choosing factors. Journal of Financial Economics 128 (2):234–52. doi:10.1016/j.jfineco.2018.02.012.

 Web of Science ®Google Scholar

Adrian, T., M. Fleming, O. Shachar, and E. Vogt. 2017. Market liquidity after the financial crisis. Annual Review of Financial Economics 9 (1):43–83. doi:10.1146/annurev-financial-110716-032325.

 Google Scholar

Adrian, T., J. Kiff, and H. S. Shin. 2018. Liquidity, leverage, and regulation 10 years after the global financial crisis. Annual Review of Financial Economics 10 (1):1–24. doi:10.1146/annurev-financial-110217-023113.

 Google Scholar

Racicot, F., W. F. Rentz, and R. Théoret. 2018. Testing the new Fama and French factors with illiquidity: A panel data investigation. Finance, Presses universitaires de Grenoble 39 (3):45–102. doi:10.3917/fina.393.0045.

 Google Scholar

Newey, W. K., and K. D. West. 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55 (3):703–08. doi:10.2307/1913610.

 Web of Science ®Google Scholar

Fama, E. F., and K. R. French. 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33 (1):3–56. doi:10.1016/0304-405X(93)90023-5.

 Web of Science ®Google Scholar

Carhart, M. M. 1997. On persistence in mutual fund performance. The Journal of Finance 52 (1):57–82. doi:10.1111/j.1540-6261.1997.tb03808.x.

 Web of Science ®Google Scholar

Amihud, Y., A. Hameed, W. Kang, and H. Zhang. 2015. The illiquidity premium: International evidence. Journal of Financial Economics 117 (2):350–68. doi:10.1016/j.jfineco.2015.04.005.

 Web of Science ®Google Scholar