Three in four rights issues here leave the old share dearer than a share assembled from rights, even after a two-point allowance for costs. That gap appears in 193 of 258 offerings. It is the paper's most compelling finding. The ESG argument that Verdú Henares, Carchano Alcina and Belza González attach to it carries much less weight.
Buying the same share by different routes
A rights issue gives the buyer a choice. Buy the stock outright, or acquire enough rights and pay the subscription cash for a new share. The authors calculate a daily "theoretical" share price for the latter route. They add the cost of the rights needed to subscribe N new shares to N times the issue price, then divide by N.
The strategy buys this synthetic share and shorts the old share, substituting a CFD when borrow is unavailable. It permits only this direction because the reverse leg cannot be constructed. The daily net return equals the old share's premium over the theoretical price, less a flat 2 percentage points. This allowance covers commissions, exchange fees, spread and borrow. The authors average daily returns over the rights-trading window. Their definition is explicit: "an arbitrage opportunity is deemed present if the strategy's average net return is positive."
Using LSEG Refinitiv daily closes, the authors assemble 258 rights issues across 41 countries from 2000 to 2023. The international sample comprises Europe 129, Asia and Oceania 88, Africa and Middle East 30, Americas 11. Their starting hypothesis is that stronger sustainability indicators should accompany more efficient pricing. They find the reverse. They use logit and probit to regress the arbitrage label on the issuer's offering-year ESG score, and OLS for the average return. Controls comprise the issue-to-ex-right price ratio, the new-to-old share ratio, a dummy for dilution of 50% or more, book-to-market, Amihud illiquidity and a developed-country dummy.
ESG's headline logit coefficient is 0.0130 (SE 0.0060). Arbitrage appears in 82.19% of offerings from B-rated issuers, against 70.42% from D-rated issuers. The authors interpret the difference as ESG-driven share overvaluation. They qualify that interpretation as "preliminary evidence" and describe their estimates as associations rather than causal estimates.
Does ESG help call the trade?
Only a little.
The implied t-stats come to 2.17 for logit, 1.85 for probit and 2.27 for OLS on 167 offerings. Table 5 places three stars alongside most standard errors and R-squared values. It also stars the logit Environmental coefficient of 0.0013, despite its SE of 0.0230. Those stars convey nothing. Table 6 goes as far as starring a Granger p-value of 0.8801 (ESG into the return, lag 1). The estimates warrant 5% significance for OLS and logit and 10% for probit; the stars claim 1%.
The individual pillars fail to hold up. Environmental's logit coefficient is 0.0013 (SE 0.0230), Social's is 0.0107 (SE 0.0160), and Governance's is -0.0019 (SE 0.0110). The authors acknowledge that significance belongs only to the aggregate score and describe its effect as modest. Yet their abstract advertises "a positive and statistically significant association", while Table 5 awards it stars at 1%. Nor do the rating grades form a monotone sequence: the 76.47% for A issuers falls below B.
With the base rate near 0.75, the logit slope translates to roughly 0.0130 x 0.75 x 0.25, or about 0.0024 per ESG point. Moving ESG by one standard deviation (24.38 points) changes the arbitrage probability by about 6 percentage points (0.0024 x 24.38).
The classification figures deserve the same base-rate check. The ESG logit correctly labels 77.91% of offerings; calling arbitrage on every offering gets 74.81%. Its advantage is 3.1 points, roughly eight additional correct calls among 258 (0.031 x 258). The pillar logit's 76.74% lies closer to the base rate still. The authors offer accuracy, an S-statistic p-value of 0.6884 and pseudo-R2 as support for calling the models "statistically sound". Failure to reject the S test addresses calibration. The 3.1-point gain in accuracy answers the question about separation.
The authors also explain that they seek to establish whether ESG helps explain arbitrage, rather than maximize R², and that modest R² values are normal in ESG research. The t-stats of 2.17 (logit) and 1.85 (probit) measure precisely that contribution. The comparison of 77.91% with 74.81% uses the authors' own evidence of fit. We are simply setting it beside the base rate.
Year, country and sector fixed effects are absent from the specification. Its sole country-level control is the high-development dummy. Of the 258 offerings, 157 occur in 2016 to 2023; Sweden and Australia supply 34 apiece. If ESG coverage and arbitrage frequency change together over time, the score could be acting as a proxy for the period. The authors explain that they excluded firm betas, market volatility measures and crisis dummies to concentrate on rights-offering mechanics and conserve degrees of freedom. That choice is defensible with 258 observations. It also prevents reading the coefficient as an ESG effect.
The Granger results run against the levels results. For prediction of the arbitrage label at lag 1, p-values are 0.0278 for ESG, 0.0059 for Environmental and 0.0110 for Social. Environmental is thus the strongest Granger predictor, despite being the weakest pillar in the regressions. Every p-value for prediction of the return exceeds 0.62. We did not find an account of how the authors order 258 independent offerings from 41 countries into a series. A "lag" has no clear interpretation without that ordering.
Clearing 2pp of costs still leaves execution open
The positive gap establishes that the old share costs more than the rights route. Either an expensive share or a cheap right fits that observation equally well. The authors favour ESG overvaluation of the shares. They also acknowledge that the new shares may be inappropriately valued and that clientele explanations cannot be excluded. The trade alone does not identify which leg moved.
Across all six models, the developed-country dummy is negative. Its t-stats of about 1.9 (OLS) and 1.6 (logit) fit an explanation based on frictions in either leg. The OLS estimate is -3.3751 with an SE of 1.7410. Model 5 reports a logit coefficient of -0.7533 against 0.4750; Model 6 reports -0.6647 against 0.9870, giving a t of 0.67. Each of the three falls below the ESG t of 2.17 that we have already called thin.
The same 2-point cost allowance applies across 41 countries and 24 years. Only brokerage and exchange fees are country-specific, according to the paper. This treats rights-market spreads and borrow fees as approximately uniform. The authors defend the flat allowance as conservative: a high all-in proxy reduces false positives and, in their argument, absorbs differences in trading frictions across countries. That defence requires 2 points to exceed realised all-in costs in every market and year. The paper supplies no country-level cost or borrow figures to establish that condition.
The paper says CFDs produce "comparable results", though we did not find figures demonstrating this. We also did not find historical records of borrow availability or recalls. A gap remaining positive in three quarters of offerings is consistent with the short leg often being unavailable. The distance between a quoted gap and an executable fill is the same issue we raised in our Polymarket review.
The return series undergoes substantial trimming: a median-absolute-deviation filter discards 91 of 258 observations. The remaining mean is still 3.42, compared with a median of 1.38 and a standard deviation of 5.71. Units are left unstated. The minimum, exactly -1.0000, is suspiciously round, and its interpretation depends on those units.
The 193 positive gaps deserve a research ticket; the ESG link does not
Sample attrition is severe. Requiring ESG data reduces 27,348 offerings to 760; requiring rights and share price series leaves 258. The authors maintain that these losses are unrelated to arbitrage, drawing on Verdú et al. (2025). They nevertheless concede that residual selection "cannot be completely dismissed." We agree with that concession.
We could not test these results ourselves. Our data contains neither subscription-rights price histories nor offering terms, specifically the rights-per-share ratio and subscription price. We therefore cannot construct the synthetic long leg. Historical Refinitiv ESG scores are also unavailable to us.
We would change our view if the ESG coefficient survived year and country effects, and if the gap persisted on days with documented borrow. For now, the useful finding for a desk is the 193 of 258 positive average gaps. A desk can compare those with its own rights-issue fills. The ESG coefficient remains a footnote to that count.