Finance and accounting are two research fields in management that have witnessed very important methodological developments in recent years. For example in Finance, the sophistication of the financial industry requires advanced models to solve problems of portfolio and risk management. Despite the progress, practioneers still face many challenges. For instance, it is well known that financial and economic variables present upward and downward trends. A simple but actual illustration is to think about recessions and expansions or in bull and bear markets. An econometric tool that has been developed to address market regimes is the regime switching model (RSM). It has been applied successfully to many time series data such as interest rates and stock returns. Regime switching models have nevertheless shortcomes. First, because it requires complex maximum likelihood optimization procedures, when applied to the joint estimation of many variables the number of estimated parameters easily reaches dozens, becoming unfeasible to estimate. Moreover, when applied to panel data (for example, more than one stock market indices) RSM assumes by construction the same regime at a given time point to all observed values, not allowing differentiating diverse market behaviour. Recently, the RSM have been extended to allow for different regime switching of variables (recognizing heterogeneity in variables) to panel data (Dias et al. (2008, 2009)). However, when applying this methodology to finance and accounting variables, it does not account some well known patterns in financial data like persistence and volatility clustering. This project aims to extend previous developments by incorporating a Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) structure in modeling the conditional variances with clustering and regime switching, simultaneously. Both types of models allow capturing general nonlinear structures in data by incorporating unobserved heterogeneity (and struc...

Using cutting-edge methodology and, when required, developing new methods, this project aims to improve the skills of those involved in longitudinal research, in particular methodologists and researchers in the social sciences. The proposed work addresses both important methodological questions and substantive issues. In particular, data from the Consortium of Household panels for European socio-economic Research (CHER) will be used to illustrate methodological challenges when modelling socio-economic change. Using exemplars from the substantive research this project will present strategies for choosing the most appropriate statistical methods for analysing data with a longitudinal structure, taking into account measurement errors and complex survey designs.Survey data is the main source of information when regarding demographic and social characteristics of the population, economic activity, lifestyle patterns, and public opinion. Longitudinal survey data allow for the periodic measurement of individual’s demographic and socio-economic changes in their conditions. In panel studies the same and (or) different variables are measured on the same units at least at two time points. Panel data is particularly adequate for investigating changes at the individual level. Longitudinal studies also allow us to distinguish the degree of variation in the response variable across time for one person from the variation among subjects and, in principle, also to make stronger causal interpretations mainly regarding inferences about changes, by determining the direction and magnitude of causal relationships. Furthermore, panel data is capable, for example, of providing measures before and after important social and economic policy events. Several statistical approaches have been used to analyse and model panel survey data. These include random effects models, transition models (as is the case of graphical chain models), structural equation models and latent curve growth models. Long...