# Exchange rate forecasts and stochastic trend breaks

Filename | Description | Size | |||
---|---|---|---|---|---|

01Front.pdf | 65.66 kB | Adobe PDF | |||

02Whole.pdf | 664.14 kB | Adobe PDF |

- Recently Added
- In Progress
- Open Access

Files in This Item:

Filename | Description | Size | |||
---|---|---|---|---|---|

01Front.pdf | 65.66 kB | Adobe PDF | |||

02Whole.pdf | 664.14 kB | Adobe PDF |

Copyright Clearance Process

- Recently Added
- In Progress
- Open Access

This item is open access.

This thesis examines the forecastability of exchange rates in the presence of
trend breaks. In particular, its focus is the predictive power of the interest
rate differential for the exchange rate.
Chapter 1 is the Introduction to the thesis. In this Chapter, I briefly
review the relevant literature on exchange rate predictability, forecasting in
the presence of structural breaks and modelling trends in exchange rate time
series.
Models are often evaluated via their out-of-sample forecasts over a single
out-of-sample period. However, not all out-of-sample (OOS) periods are of
equal difficulty - poor forecast performance of a model over a certain OOS
period might actually be evidence in favour of the model if the OOS period
was particularly difficult. In Chapter 2, I develop a way to quantify the
difficulty of an OOS period affected by trend breaks. This method uses
the deficit between the mean square forecast error of the optimal univariate
forecast of the trend breaking process and the random walk forecast. This
MSFE deficit is what needs to be made up by any extra information in a
model in order to beat the random walk. In Chapter 2, I use the degree of
difficulty measure in an ex-post analysis of the forecasts of a VEqCM over
two separate periods.
Chapter 3 shows that when an out-of-sample period has trend breaks, the
forecast error densities generated from a recursive forecasting procedure can
have a spurious multimodality at various horizons. This is clearly problematic
for any statistics calculated from these densities - in particular, any forecast
evaluation statistics or tests. It would also produce misleading value-at-risk
calculations. I develop a limit theory explaining why this occurs. In the
second half of the Chapter I show how the forecast error density can be
disentangled from the trend breaks. This allows an estimate of the extent to
which breaks have affected a particular forecast statistic.
In Chapter 4 I use a general trend representation (from the work of
P.C.B.Phillips) to model inter-break exchange rate behaviour. I show how
this broken trend representation can be used to estimate the trend breaks in
an exchange rate series. I show that with the general trend representation,
only `large ' breaks are identified - i.e., small (potentially spurious) breaks
can be modelled with an unbroken trend. In an empirical application to the
AUD-GBP exchange rate, I find that the estimated breaks can be rationalized using some recent theory on the effect of monetary policy shocks on
exchange rate trends.
Chapter 5 concludes the thesis.

Please use this identifier to cite or link to this item: