RICERCA ITALIANA     

Numerical and graphical methods for the analysis of time series data

Università degli Studi di Udine

Abstract

The ultimate objective of the present research proposal is to bring together a number of methods and techniques for the analysis of time series data deriving from nonlinear dynamical systems theory (in the following referred to as simply DST). The resulting software, a single-platform set of algorithms, will be distributed on the internet under an open-source license which preserves the intellectual paternity while making the program and code available gratis for use and further development. The website will also host a set of instructions, a set of exercises along with relevant data sets, and links to the sites of the platform environment and other pertinent links. Side products of the project would be: training of programmers in the chosen open-source platform; training of young researchers in the use of DST techniques; applications of these methods by researchers in the fields of economics and finance, geophysics, biology and ecology; proceedings of a workshop.

To meet our objectives we have organized ourselves and our work as follows. All internal members (as well as interested external members) will be involved in reaching a consensus regarding: the economic, financial, biological and geophysical phenomena to be studied; the relevant data sets; the characteristics of these databases important for software design; the already existing algorithms for time series analysis to be included; desirable functions not yet available in a usable or transformable >>>

Principal Investigator

Marji LINES Università degli Studi di UDINE

Research Objectives

The ultimate objective of the present research proposal is to make an open-source, single-platform set of algorithms for time series analysis deriving from nonlinear dynamical systems theory (DST) available on a website which also hosts a set of instructions, a set of exercises along with relevant data sets, and links to the sites of the platform environment and other pertinent links. Side products of the project would be: training of programmers in the chosen open-source platform; training of young researchers in the use of DST techniques; an international, interdisciplinary workshop following the project.

The motivating factors are various and can be summarized as follows.

1. While these methods have been discussed in the scientific literature over the past two or three decades, generally each researcher or research group has developed the algorithms for personal use and with little regard for portability, documentation, user-friendliness. A primary goal of the project is to encourage the use of such methods by producing an integrated, user-friendly, computationally efficient set of algorithms to be distributed under a GNU open source license on the web. It is our opinion that the "open source" character of the package: (i) promotes a continual evolution of the set to include new techniques and methods as they develop; (ii) contributes to the quality control of work within the scientific community by permitting researchers to validate >>>

First Results

At the end of the 16 months we expect to have a testable software available to members of the research group (and others who were not able to join this group due to previous commitments), downloadable from a preliminary website.SOFTWARE AND WEBSITE
At the end of the project we expect the site to be fairly sophisticated and include software plus documentation, along with working papers of applications and their associated data sets and, where possible, sets of exercises for teaching purposes in the various fields represented in the project.

CONFERENCE/WORKSHOP
After the conclusion of the project we intend to host an international, interdisciplinary conference/workshop on applications of the methods, remaining issues and open questions, open to all interested researchers and students, in order to publicize the program, present the results of the analysis and elicit comments and critiques.

Timescale

24 months

National and international background

The field of time series analysis is relatively new to statistical science, some would date its modern development to the seminal work of Box and Jenkins (1970). It experienced a phenomenal growth in the last quarter century in both the theoretical and applied research programs. This is no doubt due, in part, to the exponentially expanding availability of CPU time, in tandem with its exponentially declining price, which makes it possible to perform computationally heavy algorithms on large empirical data sets. Progress followed a typical path of establishing theoretical foundations for linear processes, eventually extending theory, where possible, to monotonic processes and processes with known nonlinearities. Evidence of nonlinearity in real data (i.e., time series of state variables in economics, financial markets, biological and geophysical data) attracted a growing attention on the part of time series analysts, but only lately have statistical models with unspecified functional forms made their way to mainstream statistics. Parametric and semiparametric models are still the mainstay of modern statistics, whereas nonparametric models occupy a small, though rapidly growing niche.

Recently, a different body of techniques for the analysis of time series and experimental data has been developed, inspired by dynamical systems theory (often nicknamed "chaos theory" and here referred to as DST). Broadly speaking, the DST approach to time series analysis can >>>