A FREQUENCY DECOMPOSITION APPROACH TO EXCHANGE RATE FORECASTING.
Item
-
Title
-
A FREQUENCY DECOMPOSITION APPROACH TO EXCHANGE RATE FORECASTING.
-
Identifier
-
AAI8601697
-
identifier
-
8601697
-
Creator
-
TABAKMAN, BRUCE.
-
Contributor
-
Salih Neftci
-
Date
-
1985
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Economics, General
-
Abstract
-
Using the tools of spectral analysis, the exchange rate between the Deutshe mark and dollar is decomposed into a high frequency component and a high frequency filtered component. It is hypothesized that an econometric model, utilizing exchange rate "fundamentals", for the filtered component combined with a time series or econometric model for the high frequency component can out-predict a similar economic model for the undecomposed series.;After a review of the economic fundamentals that determine exchange rates, several candidate models to explain the frequency decomposed time series are explored. Three variants of the monetary version of the purchasing power parity model, popular in the late seventies, are estimated and rejected as candidates for the high frequency filtered series. Additionally, two different specifications of a Vector Autoregressive system for the filtered data combined with first an ARIMA model and second a random walk specification for the high frequency component are likewise rejected. A single equation model incorporating "long run" fundamentals is finally adopted for the filtered series. For the high frequency series, an ordinary least squares model is fit using a "short run" fundamental and the lagged dependent variable. Comparing ex-post forecasting ability of these two equations, utilizing actual right hand side data, with a fundamentals-based equation for the original exchange rate reveals that the frequency decomposition based models do marginally better at horizons of two to three years.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
-
Program
-
Economics
