EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2058. Detecting patterns in financial data through time-frequency domain clustering

Invited abstract in session TA-63: Models for Financial Data and Risk Management, stream OR in Banking, Finance and Insurance: New Tools for Risk Management.

Tuesday, 8:30-10:00
Room: S14 (building: 101)

Authors (first author is the speaker)

1. Francesca Fortuna
Economic, Roma Tre University

Abstract

This talk introduces a strategy for detecting representative patterns in a time series of financial data. The proposal can be framed in Functional Data Analysis, where the time series is considered as a sample observed on a discrete time grid of an underlying continuous function. We are interested in studying the dynamics of this time series from the perspectives of temporal evolution and frequency content. To this end, we adopt a wavelet transformation of the raw time series to recover the underlying function. Wavelets provide a multi-resolution representation where the signal is represented as the sum of a coarse signal approximation and a set of multiscale detail coefficients providing information about the temporal data at different frequency levels. This allows highlighting trends and cycles in the time domain, also assessing variance at different frequency levels. Our contribution is the development of a clustering method for Functional Data, in which the time series, represented as a matrix of wavelet coefficients, is navigated to discover interesting patterns. We propose splitting the time domain into non-overlapping frames and clustering the data by a functional k-means. In this approach, each centroid is a representative pattern of time frames. The distance used to compare data is an optimally weighted Euclidean distance giving weights to frequency components. In this sense, we propose a new clustering objective function and an algorithm that allows optimizing it.

Keywords

Status: accepted

Back to the list of papers