The work discusses different non-standard problems under different types of short and long range dependence.
In the first part we introduce new point-wise confidence interval estimates for monotone functions observed with additive and dependent noise. Existence of such monotone trend is quite common in time series data. We study both short- and long-range dependence regimes for the errors. The interval estimates are obtained via the method of inversion of certain discrepancy statistics. This approach avoids the estimation of nuisance parameters such as the derivative of the unknown function, which other methods are forced to deal with. The resulting estimates are therefore more accurate, stable, and widely applicable in practice under mild assumptions on the trend and error structure. While motivated by earlier work in the independent context, the dependence of the errors, especially long-range dependence leads to new phenomena and new universal limits based on convex minorant functionals of drifted fractional Brownian motion.
In the second part we investigate the problem of M-estimation, the technique of extracting a parameter estimate by minimizing a loss function is used in almost every statistical problems. We focus on the general theory of such estimators in the presence of dependence in data, a very common feature in time series or econometric applications. Unlike the case of independent and identically distributed observations, there is a lack of an overarching asymptotic theory for M-estimation under dependence. In order to develop a general theory, we have proved a new triangular version of functional central limit theorem for dependent observations, which is useful for broader applications beyond our current paper. We use this general CLT along with standard empirical process techniques to provide the rate and asymptotic distribution of minimizer of a general empirical process. We have used our theory to make inferences for many important problems like change point problems, excess-mass-baseline-inverse problem, different regression settings including maximum score estimator, least absolute deviation regression and censored regression among others.