Thesis

On September 13, 2001, I defended my thesis on
Time Varying Parameter Models for Inflation and Exchange Rates
at the Erasmus University Rotterdam.

Below are some details, the table of contents is also available.
Title: Time Varying Parameter Models for Inflation and Exchange Rates
Summary: Without time variation, predicting a time series would be easy, but of little interest. However, economic time series change over time, in value and sometimes also in the underlying data generating process.

The thesis analyses variation in two sets of time series in detail, allowing for time variation wherever necessary for the goal at hand. In the first chapters, inflation series are analysed. A structural break in the mean of the process is found to influence the estimated correlation structure of the series strongly. More specifically, a change in average inflation during the period of the oil crises changes the estimate for a parameter governing the so-called long-memory behaviour strongly. Accounting for a change in the mean improves the inference on the prediction intervals of inflation at horizons of more than 6 months.

A second part of the thesis describes Bayesian sampling methods and applies these to models for exchange rates. With exchange rates, the changes are more subtle than with inflation, almost continuous. The trending of the exchange rate seems to change gradually, and the variation of returns on the exchange rate may vary between quiet periods and highly volatile months. With a structural time series model, these changes are tracked on a day-to-day basis. The results of the model estimation are used for inference, to make a decision about whether or not the risk concerning a foreign investment should be hedged. An elaborate analysis of the local changes in exchange rate series proves to provide sufficient basis for such a decision, resulting in a better risk/return ratio.

Supervisors: Herman K. van Dijk (promotor),
Philip Hans Franses (promotor),
Marius Ooms,
Peter Boswijk
Keywords: Long memory, time-varying parameter modelling, prediction, decision, MCMC sampling methods.
Graduate School: Tinbergen Institute
Date of defense: September 13, 2001
Published as: Tinbergen Institute Research Series, TI 256 (PDF, 1.5MB)
The thesis is available through the EUR repository of the Erasmus University as a PDF document (1.5MB).

The acknowledgements, table of contents, introduction, bibliography and summary in Dutch are viewable in HTML, after a rough translation using the TTH program.


[Main page] [Top] [Next: Front matter]
Last change: 7/12/2001, 12/5/2008