1: \begin{abstract}

2: We study the connection between multicalibration and boosting for squared error regression. First we prove a useful characterization of multicalibration in terms of a ``swap regret'' like condition on squared error. Using this characterization, we give an exceedingly simple algorithm that can be analyzed both as a boosting algorithm for regression and as a multicalibration algorithm for a class $\cH$ that makes use only of a standard squared error regression oracle for $\cH$. We give a weak learning assumption on $\cH$ that ensures convergence to Bayes optimality without the need to make any realizability assumptions --- giving us an agnostic boosting algorithm for regression. We then show that our weak learning assumption on $\cH$ is both necessary and sufficient for multicalibration with respect to $\cH$ to imply Bayes optimality. We also show that if $\cH$ satisfies our weak learning condition relative to another class $\cC$ then multicalibration with respect to $\cH$ implies multicalibration with respect to $\cC$. Finally we investigate the empirical performance of our algorithm experimentally using an open source implementation that we make available on GitHub\footnote{Our code repository can be found at \url{https://github.com/Declancharrison/Level-Set-Boosting}}. 

3: \end{abstract}

4: