1: \begin{abstract}

2: \noindent This paper presents uniform convergence rates for kernel

3: regression estimators, in the setting of a structural nonlinear cointegrating

4: regression model. We generalise the existing literature in three ways.

5: First, the domain to which these rates apply is much wider than has

6: been previously considered, and can be chosen so as to contain as

7: large a fraction of the sample as desired in the limit. Second, our

8: results allow the regression disturbance to be serially correlated,

9: and cross-correlated with the regressor; previous work on this problem

10: (of obtaining uniform rates) having been confined entirely to the

11: setting of an exogenous regressor. Third, we permit the bandwidth

12: to be data-dependent, requiring only that it satisfy certain weak

13: asymptotic shrinkage conditions. Our assumptions on the regressor

14: process are consistent with a very broad range of departures from

15: the standard unit root autoregressive model, allowing the regressor

16: to be fractionally integrated, and to have an infinite variance (and

17: even infinite lower-order moments).

18: \end{abstract}

19: