Use this URL to cite or link to this record in EThOS:
Title: Spectural analysis of ultra-high-frequency foreign exchange market data
Author: Giampaoli, Iacopo
ISNI:       0000 0004 2738 0744
Awarding Body: University of Essex
Current Institution: University of Essex
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
This thesis presents a novel application of advanced methods from Fourier analysis to the study of ultra-high-frequency (UHF) financial data, both in a univariate and a multivariate setting. The use of the Lomb-Scargle Fourier Transform (L8FT) provides a robust and natural framework to take into account the irregular spacing in the time domain, whilst avoiding restrictive assumptions, complex model specifications, or resorting to traditional imputation methods, such as interpolation and regular resampling, which not only cause artefacts in the data and loss of information, but also the generation of spurious information. An event-based approach (intrinsic time), which is, by its nature, inhomogeneous in physical time, is then employed using different directional-change event thresholds to filter the foreign exchange (FX) tick-data, leading to a power-law relationship. The calculated spectral density demonstrates that the price process in intrinsic time contains different periodic components, implying the existence of new stylised facts of UHF data in the frequency domain. The intrinsic time scale allows, in addition, to model the behaviour of heterogeneous agents that differ in their perception of the market, in their risk profiles, and hence in their trading frequencies. The inhomogeneous nature of UHF data makes the analysis of lead-lag relationships between different financial variables or markets partic- ularly problematic. Most estimators in the literature require either regular sampling of the unevenly-spaced data, or interpolation onto an equally-spaced grid, which in turn create spurious data and impu- tation bias. Therefore, this thesis suggests the use of cross-spectral analysis to study lead-lag relationships between different FX rates. The LSFT and Welch-Overlapped-Segment-Averaging (WOSA) procedure, in combination with an intrinsic time scale, permit fast pro- cessing of all transaction data at arbitrarily high frequencies, whilst allowing to draw inferences on the behaviour of heterogeneous market agents. The spectral analysis reveals the existence of various lead-lag patterns between different exchange rates, that could potentially be employed to develop new algorithms, or to optimise existing strategies in automated trading. The findings herein suggest that this framework, built upon the LFST, is an efficient method to identify periodic patterns in individual time series, as well as to analyse lead-lag relationships between multiple series, and hence, would be a valuable tool for researchers and prac- titioners alike.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available