Essays in economic and financial time series analysis

Abstract

The research presented in this thesis contributes to four areas in the Economic and Financial Time Series Analysis literature. These include the topics of (i) Selection of Long Memory Time Series Models, (ii) Bootstrapping Strongly Dependent Data, (iii) Forecasting Key Macroeconomic Variables and (iv) Portfolio Optimisation. The first part focuses on strongly dependent series. It aims to establish an asymptotically consistent information criterion for long memory processes when the long memory parameter is semi parametrically estimated. A set of Monte Carlo experiments and the analysis of monthly inflation time series show the validity of the new methodology. Next, we are concerned with the issue of bootstrap in strongly dependent data. We introduce a fractional differencing bootstrap methodology that allows the implementation of any resampling method in such series. Evidence of robustness is given by Monte Carlo experiments using various block and residuals resampling schemes. The second part of the thesis investigates the issue of forecasting macroeconomic variables. Heuristic methods for the optimisation of information criteria are employed and their forecasting performance is compared to the standard choices in the literature. The empirical application in Euro Area dataset suggests that the non-standard methods should be taken into consideration as they provide better forecasts on average. The last part of the thesis investigates the applied performance of covariance shrinkage in the portfolio optimisation problem when the universe of assets is large. Our approach suggests the use of a shrinkage coefficient that optimises functions with financial interpretation. Empirical results provide evidence that the shrinkage portfolios obtained using the suggested approach are characterised by higher Sharpe Ratios, cumulative returns and profit/loss ratio.