This thesis comprises three main chapters focusing on a number of issues related to forecasting economic and nancial time series. Chapter 2 contains a detailed empirical study comparing forecast perfor- mance of a number of popular term structure models in predicting the UK yield curve. Several questions are addressed and investigated, such as whether macroeconomic information helps in forecasting yields and whether predict- ing performance of models change over time. We nd evidence of signi cant time-variation in forecast accuracy of competing models, particularly during the recent nancial crisis period. Chapter 3 explores density forecasts of the yield curve which, unlike the point forecasts, provide a full account of possible uncertainties surrounding the forecasts. We contribute by evaluating predictive performance of the recently developed stochastic-volatility arbitrage-free Nelson-Siegel models of Chris- tensen et al. (2010). The one-month-ahead predictive densities of the models appear to be inferior compared to those of their constant-volatility counter- parts. The advantage of modelling time-varying volatilities becomes evident only when forecasting interest rates at longer horizons. Chapter 3 deals with a more general problem of forecasting time series under structural change and long memory noise. Presence of long memory in the data is often easily confused with structural change. Wrongly account- ing for one when the other is present may lead to serious forecast failure. In our search for a forecast method that can perform reliably in presence of both features we extend the recent work of Giraitis et al. (2013). A forecast strategy with data-dependent discounting is adopted and typical robust-to- structural-change methods such as rolling window regression, forecast averag- ing and exponentially weighted moving average methods are exploited. We provide detailed theoretical analyses of forecast optimality by considering cer- tain types of structural changes and various degrees of long range dependence in noise. An extensive Monte Carlo study and empirical application to many UK time series ensure usefulness of adaptive forecast methods.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:667263 |
Date | January 2014 |
Creators | Mansur, Mohaimen |
Publisher | Queen Mary, University of London |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://qmro.qmul.ac.uk/xmlui/handle/123456789/8576 |
Page generated in 0.0016 seconds