On Greg Mankiw's 'Do No Harm'

A rebuttal to Greg Mankiw's claim that the government should not interfere in voluntary exchanges. This is from Rakesh Vohra at Theory of the Leisure Class:

Do No Harm & Minimum Wage: In the March 23rd edition of the NY Times Mankiw proposes a 'do no harm' test for policy makers:

…when people have voluntarily agreed upon an economic arrangement to their mutual benefit, that arrangement should be respected.

There is a qualifier for negative externalities, and he goes on to say:

As a result, when a policy is complex , hard to evaluate and disruptive of private transactions, there is good reason to be skeptical of it.

Minimum wage legislation is offered as an example of a policy that fails the do no harm test. ...

There is an immediate 'heart strings' argument against the test, because indentured servitude passes the 'do no harm' test. ... I want to focus instead on two other aspects of the 'do no harm' principle contained in the words 'voluntarily'and 'benefit'. What is voluntary and benefit compared to what? ...

When parties negotiate to their mutual benefit, it is to their benefit relative to the status quo. When the status quo presents one agent an outside option that is untenable, say starvation, is bargaining voluntary, even if the other agent is not directly threatening starvation? The difficulty with the `do no harm’ principle in policy matters is the assumption that the status quo does less harm than a change in it would. This is not clear to me at all. Let me illustrate this...

Assuming a perfectly competitive market, imposing a minimum wage constraint above the equilibrium wage would reduce total welfare. What if the labor market were not perfectly competitive? In particular, suppose it was a monopsony employer constrained to offer the same wage to everyone employed. Then, imposing a minimum wage above the monopsonist’s optimal wage would increase total welfare.

[There is also an example based upon differences in patience that I left out.]