I have been turning the idea of taxing bad stuff (where the costs of producing or consuming a good has adverse consequences that are not adequately compensated for). The normal economic idea is to assume that if a cost will be borne by third parties, there should be a tax to discourage that by an optimal amount commensurate with the costs it imposes. I think this basic idea is correct, but that there are cases where direct parties to a transaction (the producer or consumer) will bear a cost, but they do not respond appropriately.
Think of this: smoking imposes a future cost on the smoker. This means that a rational smoker should save more than a non-smoker because they can expect higher medical costs in the future. My instincts tell me this is not true or, even if it is, smokers probably don't set aside an actuarially appropriate amount.
I have another thought about taxing bads, which I think a lot of people fall into. Quite often, one will find that an individual will estimate the cost of producing or consuming a bad by multiplying the cost for outcomes that are associated with the bad by the number of incidences to come up with the total costs of the bad in question. While this would be acceptable if there was strict causation, such that if you don't do X, Y will not result, but if you do X, there is a probability that Y will result, this is not a general case. In my own estimation, there is usually a probability that Y will result, even without consuming or producing X. Thus, the appropriate cost is actually determined by the conditional probability of Y, given X, minus the conditional probability of Y, given not X. In a concrete example, there is a correlation which is most cause-effect between the abuse of alcohol and the incidence of family violence (just to make this a more pure externality), but even without alcohol abuse, there still exists a risk of family violence. Thus, the cost of alcoholism with regard to increased risk of family violence is determined by the costs associated with family violence times the probability of family violence, given alcoholism, minus the probability of family violence without alcoholism.
My final point is about delayed effects of the production or consumption of a bad. There is always a temptation to use the current costs being incurred and dividing by the production or consumption of the bad to determine a proper tax, but while in some cases this may work out well, it only works if the effects are immediate. If, however, all or part of the effects are delayed, then the proper method is to have a tax based upon the present discounted value of the costs imposed by producing or consuming the bad today. Let us say that we want to tax coal emissions because of negative health effects on people who live near a coal power plant. While some health effects may occur today, much of them would probably be things like the development of emphysema or lung cancer. Most of the costs would be delayed for many years, so the appropriate measurement of the costs is the present discounted value of the future costs of emphysema and cancer, etc.
I hope this may have been instructive. I have wanted to discuss these issues for quite a while, as I am generally a believer in pigouvian taxes and want to develop my own thoughts on the intricacies of such policies.