BAND SPECTRAL REGRESSION AND LASSO: CAN IT IMPROVE FORECASTING PERFORMANCE OF EXCHANGE RATE FUNDAMENTAL MODELS?
LIAW WYI WYING
LIAW WYI WYING
Citations
Altmetric:
Alternative Title
Abstract
The literature on exchange rate forecasting has provided conflicting empirical evidence on the predictive ability of fundamentals. Commonly used model specifications to forecast exchange rates are largely limited to linear time series and factor models estimated in the time domain, yet frequency domain approaches are rarely considered and used in this context. Wada (2022) devised a novel approach to exchange rate forecasting using fundamentals by combining the Least Absolute Shrinkage and Selection Operator (LASSO) with frequency domain techniques. This thesis employs (1) a modification to LASSO by allowing the tuning parameter of the penalty function to vary across rolling windows; and (2) extending the data to December 2019. This thesis concludes that forecasts
constructed by the LASSO model perform significantly better than the random walk model using Purchasing Power Parity and Monetary fundamentals. However, the predictive performance of Taylor Rule fundamentals is lacking in the frequency domain.
Keywords
Band Spectral Regression, Exchange Rate Models, Frequency Domain, Penalised Regression
Source Title
Publisher
Series/Report No.
Collections
Rights
Date
2023-04-03
DOI
Type
Thesis