The information content and redistribution effects of state and municipal rating changes in Mexico

The fiscal and financial reforms carried out in Mexico in 2000 have encouraged a widespread presence of rating agencies and have allowed several States and Municipalities to raise funds through bond offerings in the capital market. Any local government in Mexico intending to access credit and capital markets must count with at least one credit rating from one of the three main agencies: FitchRatings, Moody’s and Standard & Poor’s. This paper investigates the impact of rating changes to State and Municipal governments on bond returns in Mexico. By employing a Capital Asset Pricing Model (CAPM) structure for the mean equation that allows conditional volatility, we find strong support for the Information Content Signaling Hypothesis (ICSH), i.e., rating upgrades (downgrades) are followed by greater (lower) bond returns. We also find some support for the Wealth Redistribution Hypothesis (WRH) indicating that rating upgrades (downgrades) are followed by lower (greater) bond returns. In addition to this, we find high volatility persistence, significant asymmetric responses of volatility to bad and good news, a negative association between market volatility and the level of bond returns and significant effects of volatility in response to rating changes. Finally, the estimations show the market anticipates and responds to rating changes within five-day momentum windows. There is a comparatively stronger reaction of returns on the event day favoring the hypothesis of market inefficiency. JEL: C22, G14, G20, H74, H77


Results
Estimating the main equation for bond returns they find a negative (and highly significant 1 ) coefficient for the variance term. They interpret this as a negative correlation between risk premium and conditional variance (which is not how I read the coefficient). They also confuse me by claiming that their finding is (i) "in line with the seminal paper by Nelson (1991)", but (ii) still "counterintuitive", yet (iii) not necessarily in contradiction to Backus and Gregory (1993) as well as somes other literature (p. 16-17).
Further they find low Betas (correlation of state and municipial bond returns with market risk).
Finally, on their main issue, they find that rating changes mostly confirm the signaling hypothesis (ratings convey some information), but that often enough, the contrary redistribution effect overrides the signaling effect. They also find an impact on higher moments of the return distributions.

Comments
First of all, the topic and setup of the paper is interesting. I did also benefit from the descriptive part. There is an interesting observation on the "trust fund effect" in Mexico, which I would expect to cause some negative trend to ratings, but a positive trend to bond returns in the period under examination. I do not understand why the effect is introduced but not used in the empirical part of the paper.
Discussiong the bank regulation motive for having a rating the authors might mention the likely influence of the discussions on Basel 2 capital standards in the late nineties.
One problem I had was with the presentation of the theoretical background. I do understand the signaling hypothesis (good borrowers are keen to get ratings in order to signal their quality). But I do not understand the asset wealth redistribution hypothesis. I read the explanation on pp. 8 and 19 several times. I am also familiar with the elements, like shareholders' incentives to issue debt in order to redistribute wealth from existing debtholders to shareholders. Yet, with all best intentions, I fail to understand the redistribution story as it is explained in the paper.
The empirical approach is basically plausible if one thinks that fat tails and conditional volatility are an important phenomena in the present context. (A bit more discussion of tail shape of bond returns and empirical alternatives might have been helpful, though) .
But let me focus on the main critical issues: Using equation (1), the authors perform some initial analysis of the data. They find that bond returns on four state offerings (among a total of 40 issues, see p. 5) "converge satisfactorily and do not exhibit correlation in the residuals nor squared residuals" (p. 13). Rather than rejecting the model, they reject the other data and restrict their further empirical analysis to the four bond issues that behaved well. I tend to consider this the crucial issue in the paper.
Further, I cannot judge the pros and cons of explaining the stochastic process for bond returns and the impact of rating events in one model. However, as far as rating changes may affect not only returns (the dependent variable) but also the volatility term (an explanatory variable) the method may call for some further explanation for the benefit of non-specialists like myself.
At one point, the authors start to refer to expected returns. They do not explain how they observe expected returns, though. Does expected refer to estimated returns from the model? Or does the term mean average returns (the authors refer to table 2)? On this issue, the reader would need some clarification.
The results are to a large part surprising (positively speaking) or implausible (negative view). Estimated bond returns are close to zero, the variance being 30 times higher than the returns. Would this suggest that investors in these bonds are not only satisfied with a zero (nominal) return but would also be risk-neutral? Or, given that the volatility parameter has a negative sign, do investors happily forego return, because they get some risk instead??? Note that the negative volatility parameter is all the more puzzling since the authors assume that the distribution of returns may have fat tails.

Final remark
While I am not a specialist regarding the particular empirical methods used, I do think it is the authors' job to explain their paper in a way that a typical economist can understand the more tricky issues it raises.