Modeling and forecasting of financial time series

Abstract

This thesis is concerned with modeling and forecasting of financial time series and consists of two parts.<br />In part 1 (chapter 2-6) we present some model classes and their estimation which are used for the modeling and forecasting of stock returns.<br />In chapter 2 we consider the only univariate model used in this thesis which is the univariate ARX model.<br />Chapter 3 deals with VARX model, in which there is no restriction on parameters matrices, and the number of independent parameters is increasing with the square of dimension of the dependent variables.<br />In chapter 4 we present the static principal component model.<br />In the subsequent chapter we discuss the generalized dynamic factor model; In particular the model, presented by Stock and Watson (2002) and the model given by Forni et al. (2002) In chapter 6 we present the Reduced Rank regression, autoregression model.<br />The second part of the thesis is devoted to the estimation of large covariance matrices for portfolio optimization. In this part of the thesis we use two different techniques to shrink the covariance matrix.<br />The first is called the LASSO and the second method used here is called Nested LASSO. In chapter 8 we present the empirical analysis and discuss the results of the above techniques on real stock data.<br />