By Amanda Michaud and Davic Wiczer
http://d.repec.org/n?u=RePEc:nys:sunysb:17-11&r=dge
Using retrospective data, we introduce evidence that occupational exposure significantly affects disability risk. Incorporating this into a general equilibrium model, social disability insurance (SDI) affects welfare through (i) the classic, risk-sharing channel and (ii) a new channel of occupational reallocation. Both channels can increase welfare, but at the optimal SDI they are at odds. Welfare gains from additional risk-sharing are reduced by overly incentivizing workers to choose risky occupations. In a calibration, optimal SDI increases welfare by 2.3% relative to actuarially fair insurance, mostly due to risk sharing.
This is the classic insurance conundrum: it enhances welfare through risk-sharing, but also reduces it through encouraging risk-taking. Here it is with disability insurance that influences occupational choice. In the paper, the disability is expressed as not being able to general labor income. I wonder if accounting for the physical hurt of a disability would significantly affect the outcome. I can understand that this is difficult to model and calibrate, though.
[a previous version of this post had the wrong paper title]
NEP-DGE Blog 
You hit exactly the key mechanism: “Here it is with disability insurance that influences occupational choice.” and by choosing occupations folks are also choosing their risk level: so the old moral hazard thing also has a distortion on labor allocation and therefore productivity.
We show that the distortion can actually be a good thing though: with incomplete markets people take inefficiently little risk / allocate themselves to safe occupations.
To make a quantitative, you’re right, it’s really a challenge. In the baseline quantitative portion we added a health disutility calibrated to Low & Pistaferri (AER 2015) who identify the change in marginal utility of consumption from health using PSID linked to CEX.
That said, we think one (of many) really fruitful angle would be to find quantitative gains from what the model points would be better policy: risk-weighted DI premia.
And thanks for noticing the paper!