Accounts such as revenues and expenses are nonnegative by convention, making their distributions inherently skewed. Because of this asymmetry, changing the means of these accounts can affect their higher-order moments as well, which in turn alters their informativeness as performance measures. In a parametric multitasking setting, I show that the principal incentivizes not just more effort toward tasks that improve expected outcomes at a lower cost, but toward tasks that naturally improve informativeness. I further show that for a large class of positively-skewed distributions, mean-increasing actions harm informativeness whereas mean-decreasing actions improve it. For example, cost cutting makes expenses more informative, whereas sales growth makes revenues less informative. As a result, the principal incentivizes more cost cutting than revenue growth, even when the marginal cost and productivity of these tasks are the same.

We revisit the optimal use of information under moral hazard by assuming that the agent chooses distributions nonparametrically at a cost given by an f-divergence. Under this assumption, the optimal contract behaves as if the principal were making inferences about outcomes she values. Consequently, Holmström's (1979) informativeness principle does not apply. A performance measure is useful for contracting if and only if it is informative about value, not the agent's action.

Using a principal-agent framework in which the agent chooses the joint distribution over all contractible and non-contractible signals, we provide a theoretical justification for contracting on aggregated accounting estimates. The optimal contracting process can be decomposed into three stages: estimating individual items that the principal values, aggregating those estimates using the weights in the principal's objective (as opposed to weights driven by sensitivity or precision), and writing a one-dimensional contract on the aggregated estimate. Using a highly tractable specification of our model in which optimal contracts are linear and normal distributions arise endogenously, we show that optimal measurement rules are conservative yet produce unbiased estimates, and we rationalize the immediate expensing of R&D, the capitalization of PP&E, and the accrual of credit sales.

We develop a novel multitasking framework in which an agent alters the means, variances and correlations of normally-distributed variables in response to linear and nonlinear incentives. Using this model, we study how contractual and regulatory incentives shape financial and ESG activities. Most notably, we show that ESG performance shares – which apply stronger financial incentives for better ESG performance – incentivize managers to 1) engage in green innovation, increasing the correlation between financial and ESG outcomes, 2) amplify risk in green firms and reduce risk in brown firms, 3) strengthen average financial and ESG performance, and 4) improve measurement quality, increasing the correlation between outcomes and their measures. Moreover, ESG-contingent income tax rates -- which apply lower income tax rates for better ESG performance -- motivate firms to offer managers ESG performance shares, thereby inducing these activities across all taxable corporations.