No label defined (Q4940647)

Entropy methods are an important tool in the theory of empirical processes based on independent and identically distributed random variables and indexed by classes of sets or functions. The purpose of this monograph (based on the author's doctoral thesis) is to develop entropy methods for some martingales which are indexed by an additional parameter and to apply the results to the theory of statistical inference for stochastic processes. Three types of processes \((X^\psi \mid \psi\in\Psi)\) indexed by an arbitrary set \(\Psi\) are considered: (i) each coordinate process \(t\mapsto X_t^\psi\) is a stochastic integral with respect to a multivariate point process, (ii) each process \(t\mapsto X_t^\psi\) is a partial sum process of a discrete time martingale and (iii) each process \(t\mapsto X_t^\psi\) is a continuous local martingale. Chapter 1 gives a general overview of the results and introduces basic concepts and notations. Chapter 2 is devoted to maximal inequalities for expectations, that is, to bounds for \(E\sup_t \sup_{\psi,\varphi\in\Psi} |X_t^\psi - X_t^\varphi|1_B\), where the indicator \(1_B\) provides a truncation in terms of a certain ``quadratic modulus'' or an ``exponential modulus''. The bounds themselves are in terms of appropriate notions of entropy for the parameter set \(\Psi\). Chapter 3 is a central part of the book and gives weak convergence results for the three classes of martingales in \(\ell^\infty\)-spaces of bounded functions. Here, the maximal inequalities from Chapter 2 are crucial for the tightness part of the proofs. The remaining four chapters contain statistical applications: Integral estimators (Chapter 4), \(M\)-estimators (Chapter 5), Non-parametric maximum likelihood estimators (Chapter 6) and Miscellanies (Chapter 7). This monograph is highly technical and not easy to read. But it is the first step into a systematic study of entropy methods in the framework of martingales, and therefore will be of importance.