1: \begin{abstract}

2: Classical least squares estimators are well-known to be robust with respect to moment assumptions 

3: concerning the error distribution in a wide variety of finite-dimensional statistical problems;  

4: generally only a second moment assumption is required for least squares estimators to maintain

5: the same rate of convergence that they would satisfy if the errors were assumed to be Gaussian.

6: In this paper, we give a geometric characterization of the robustness of shape-restricted 

7: least squares estimators (LSEs)  to error distributions with an 

8: $L_{2,1}$ moment, in terms of the `localized envelopes' of the model. 

9: 	

10: 

11: This envelope perspective gives a systematic approach to proving oracle inequalities 

12: for the LSEs in shape-restricted regression problems in the random design setting, 

13: under a minimal $L_{2,1}$ moment assumption on the errors. The canonical isotonic 

14: and convex regression models, and a more challenging additive regression model 

15: with shape constraints are studied in detail. Strikingly enough, in the additive model 

16: both the adaptation and robustness properties of the LSE can be preserved, up to 

17: error distributions with an $L_{2,1}$ moment, for estimating the shape-constrained 

18: proxy of the marginal $L_2$ projection of the true regression function. This holds 

19: essentially regardless of whether or not the additive model structure is correctly specified.

20: 

21: 

22: The new envelope perspective goes beyond shape constrained models. 

23: Indeed, at a general level, the localized envelopes give a sharp characterization 

24: of the convergence rate of the $L_2$ loss of the LSE between the worst-case 

25: rate as suggested by the recent work of the authors \cite{han2017sharp}, and the 

26: best possible parametric rate. 

27: 

28: \end{abstract}

29: