Esta monografia é sobre uma técnica de análise de séries temporais
que é frequentemente chamada de “análise de espectro singular” ou SSA. O
algoritmo SSA básico parece simples, mas entender o que ele faz e como
ele se encaixa entre as outras técnicas de análise de séries temporais
não é nada simples.
Gastar tanto tempo em apenas uma técnica deve ser justificado de
alguma forma. A justificativa é nossa crença nas capacidades do SSA:
estamos absolutamente convencidos de que, para uma ampla gama de séries
temporais, o SSA pode ser extremamente útil. Mais do que isso,
acreditamos firmemente que, em um futuro próximo, nenhum pacote
estatístico será pensado sem incorporar as facilidades da SSA, e todos
os livros de análise de séries temporais conterão uma seção relacionada
à SSA.
Embora não seja amplamente conhecido entre estatísticos e
econometristas, o SSA tornou-se uma ferramenta padrão em meteorologia e
climatologia; é também uma técnica bem conhecida em física não linear e
processamento de sinais. Acreditamos que a falta de popularidade do SSA
entre os estatísticos se deve principalmente à tradição e à falta de
teoria do SSA. Devemos também aceitar que o principal princípio
metodológico da SSA não é realmente estatístico; SSA é mais uma técnica
de geometria multivariada do que de estatística.
Além da estatística e da geometria multivariada, a teoria de SSA
compreende os elementos de processamento de sinal, álgebra linear,
sistemas dinâmicos não lineares, a teoria das equações diferenciais
ordinárias e de diferenças finitas e análise funcional. Portanto, não é
surpreendente que tenhamos demorado muito para alcançar algum nível de
compreensão do que é SSA.
SSA é essencialmente uma técnica sem modelo; é mais uma ferramenta
exploratória de construção de modelos do que um procedimento
confirmatório. Visa uma decomposição da série original em uma soma de um
pequeno número de componentes interpretáveis, como uma tendência de
variação lenta, componentes oscilatórios e um ruído “sem estrutura”.
O principal conceito no estudo das propriedades SSA é a
“separabilidade”, que caracteriza o quão bem diferentes componentes
podem ser separados uns dos outros.
Uma característica importante do SSA é que ele pode ser usado para
analisar séries relativamente curtas. Por outro lado, a separação
assintótica desempenha um papel muito importante na teoria da SSA. Não
há contradição aqui porque as características assintóticas, que se
mantêm quando o comprimento da série \(N\) tende ao infinito, são encontradas para
\(N\) relativamente pequeno. Em
aplicações práticas, normalmente lidamos com séries de comprimento
variando de algumas dezenas a alguns milhares.
As possíveis áreas de aplicação da SSA são diversas: de matemática e
física a economia e matemática financeira, de meteorologia e oceanologia
a ciências sociais e pesquisa de mercado. Qualquer série aparentemente
complexa com uma estrutura potencial pode fornecer outro exemplo de
aplicação bem-sucedida de SSA.
Há um grande número de exemplos nesta mnografia. Muitos desses
exemplos são séries da vida real de diferentes áreas, incluindo
medicina, física, astronomia, economia e finanças. Esses exemplos não
são os exemplos mais empolgantes de aplicação do SSA; eles não foram
selecionados para impressionar o leitor. O propósito da seleção foi
outro: os exemplos servem apenas para ilustrar os aspectos metodológicos
e teóricos discutidos. Além disso, cada exemplo ilustra uma
característica diferente do método, de modo que o número de exemplos
dificilmente pode ser reduzido.
O suporte computacional é a libraria R Rssa “A
Collection of Methods for Singular Spectrum Analysis”, versão 1.0.5 de
2022-08-22 14:12:59 UTC no repositório CRAN e autores Anton
Korobeynikov, Alex Shlemov, Konstantin Usevich, Nina Golyandina e
maintainer: Anton Korobeynikov.
Em resumo, o SSA ou Singular Spectrum Analysis, é um método não
paramétrico moderno para a análise de séries temporais e imagens
digitais. O pacote Rssa fornece um conjunto de
implementações rápidas e confiáveis de várias rotinas para realizar
decomposição, reconstrução e previsão de séries temporais. Uma descrição
abrangente dos métodos e funções do Rssa pode ser
encontrada em Golyandina et al (2018).
Tipicamente, o uso do pacote Rssa começa com a
decomposição da série temporal usando ssa. Depois
disso, é necessário um agrupamento adequado das séries temporais
elementares. Isso pode ser feito, por exemplo, olhando para o gráfico da
decomposição, utilizando o comando plot.
Alternativamente, pode-se examinar a chamada matriz de correlação \(\omega\), utilizando wcor.
O agrupamento automático pode ser realizado por meio de
grouping.auto. Além disso, os métodos Oblique SSA podem
ser usados para melhorar a separabilidade da série,
iossa, fossa.
A próxima etapa inclui a reconstrução da série temporal usando o
agrupamento selecionado reconstruct. Termina-se com
estimativas de frequências parestimate, diversar
previsões de séries forecast,
rforecast, vforecast e, se houver,
preenchimento de lacunas gapfill,
igapfill.
Abarbanel, H.D.I. (1996). Analysis of Observed Chaotic Data.
Springer, New York.
Allen, M.R. and Smith, L.A. (1996). Monte-Carlo SSA: Detecting
irregular oscillations in the presence of coloured noise. Journal
of Climate, 9:3373–3404.
Anderson, T.W. (1994). Statistical Analysis of Time Series.
Wiley, New York.
Berge, P., Pomeau, Y. and Vidal, Ch. (1986). Order within Chaos:
Towards a Deterministic Approach to Turbulence. Wiley, New York.
Bouvet, M. and Clergeot, H. (1988). Eigen and singular value
decomposition technique for the solution of harmonic retrival
problems. In E.F. Deprettere, editor, SVD and Signal Processing:
Algorithms, Applications and Architechtures, pages 93–114.
North-Holland, Amsterdam.
D. Brillinger, D. (1975). Time Series. Data Analysis and
Theory. Holt, Rinehart and Winston, Inc., New York.
Broomhead, D.S. and King, G.P. (1986b). On the qualitative analysis
of experimental dynamical systems. In S. Sarkar, editor, Nonlinear
Phenomena and Chaos, pages 113–144. Adam Hilger, Bristol.
Broomhead, D.S., Jones, R., King, G.P. and Pike, E.R. (1987).
Singular system analysis with application to dynamical systems.
In E.R. Pike and L.A. Lugaito, editors, Chaos, Noise and Fractals, pages
15–27. IOP Publishing, Bristol.
Broomhead, D.S. and King, G.P. (1986a). Extracting qualitative
dynamics from experimental data. Physica D, 20:217–236.
Buchstaber, V.M. (1994). Time series analysis and
grassmannians. In S. Gindikin, editor, Applied Problems of Radon
Transform, volume 162 of AMS Transactions — Series 2, pages 1–17. AMS,
Providence.
Clemens, J.C. (1994). Whole earth telescope observation of the white
dwarf star PG1159-035. In A.S. Weigend and N.A. Gershenfeld,
editors, Time Series Prediction: Forecasting the Future and
Understanding the Past. Addison-Wesley, Reading.
Corduneanu, C. (1968). Almost Periodic Functions. Interscience
Publishers, New York.
Cutler, C.D. and Kaplan, D.T. (1997) (eds). Nonlinear Dynamics and
Time Series: Building a Bridge between the Natural and Statistical
Sciences. American Mathematical Society, Providence, Rhode Island.
Danilov, D. (1997a). Principal components in time series
forecast. Journal of Computational and Graphical Statistics,
6:112–121.
Danilov, D. (1997b). The ‘Caterpillar’ method for time series
forecasting. In D. Danilov and A. Zhigljavsky, editors, Principal
Components of Time Series: the ‘Caterpillar’ method, pages 73–104.
University of St.Petersburg, St.Petersburg. (In Russian).
Danilov, D. and Zhigljavsky, A. (1997) (eds). Principal Components
of Time Series: the ‘Caterpillar’ method. University of
St.Petersburg, St.Petersburg. (In Russian).
Diggle, P. (1998). Time series. A Biostatistical Introduction.
Clanderon Press, Oxford.
Draper, N.R. and Smith, H. (1998). Applied Regression Analysis.
Wiley, New York, third edition.
Edwards, R.E. (1979). Fourier Series. A Modern Introduction, volume
1. Springer-Verlag, New York, second edition.
Efron, B. and Tibshirani, R. (1986). Bootstrap methods for standard
errors, confidence intervals and other measures of statistical
accuracy. Statistical Science, 1(1):54–75.
Elsner, J.B. and Tsonis, A.A. (1996). Singular Spectrum Analysis. A
New Tool in Time Series Analysis. Plenum Press, New York.
Fowler, A.C. and Kember, G. (1998). Singular systems analysis as a
mowing-window spectral method. European Journal of Applied
Mathematics, 9:55–79.
Gantmacher, F.R. (1998). The Theory of Matrices. Chelsea,
Providence, RI.
Gelfond, A.O. (1967). Finite Difference calculus. Nauka,
Moscow. (In Russian).
Ghil, M. and Taricco, C. (1997). Advanced spectral analysis
methods. In G.C. Castagnoli and A. Provenzale, editors, Past and
Present Variability of the Solar-Terrestrial System: Measurement, Data
Analysis and Theoretical Models, pages 137–159. IOS Press, Amsterdam.
Ghil, M. and Vautard, R. (1991). Interdecadal oscillations and the
warming trend in global temperature time series. Nature,
350:324–327.
Golyandina, N., Nekrutkin, V. and Zhigljavsky, A. (2018), “Analysis of
Time Series Structure: SSA and Related Techniques”. Monographs on
Statistics and Applied Probability, Champman & Hall/CRC.
Grenander, U. and Szeg¨o, G. (1984). Toeplitz Forms and their
Applications. Chelsea, New York.
Hipel, K.W. and McLeod, A.I. (1994). Time Series Modelling of Water
Resources and Environmental Systems. Elsevier Science, Amsterdam.
Janowitz, M.F. and Schweizer, B. (1989). Ordinal and percentile
clustering. Mathematical Social Sciences, 18:135–186.
Jolliffe, I.T. (1986). Principal Component Analysis. Springer
Series in Statistics. Springer-Verlag, New York.
Kantz, H. and Schreiber, T. (1997). Nonlinear Time Series
Analysis. Cambridge University Press, Cambridge.
Kendall, M.G. and Stuart, A. (1976). Design and Analysis, and Time
Series, volume 3 of The Advanced Theory of Statistics. Charles
Griffin, London, third edition.
Kuipers, L. and Niederreiter, H. (1974). Uniform Distribution of
Sequences. Wiley, New York.
Ledemann, W. and Lloyd, E. (1984) (eds). Statistics, Part B, volume
VI of Handbook of Applied Mathematics. Wiley, Chichester.
Loève, M. (1963). Probability Theory. D. Van Nostrand Company
inc., Princeton, New Jersey, third edition.
Madisetti, V.K. and Williams, D.B. (1998) (eds). The Digital Signal
Processing Handbook. CRC Press, Boca Raton.
Makridakis, S., Wheelwright, S. and Hyndman, R.J. (1998).
Forecasting: Methods and Applications. Wiley, New York, third
edition.
Marple-Jr, S.L. (1987). Digital Spectral Analysis. Prentice
Hall, New Jersey.
McCleary, R. and Hay, R.A (1980). Applied Time Series Analysis for
the Social Sciences. Sage Publication, London.
Montgomery, D.C. and Johnson, L.A. (1976). Forecasting and Time
Series Analysis. McGraw-Hill, New York.
Moskvina, V.G. and Zhigljavsky, A.A. (2000). Application of the
singular-spectrum analysis for change-point detection in time
series. Journal of Time Series Analysis.
Nekrutkin, V.V. (1997). Decomposition of time series. In D.
Danilov and A. Zhigljavsky, editors, Principal Components of Time
Series: the ‘Caterpillar’ method, pages 194–227. University of
St.Petersburg, St.Petersburg. (In Russian).
Nekrutkin, V.V. (1999). Approximation spaces and continuation of
time series. In S.M. Ermakov and Yu.N. Kashtanov, editors,
Statistical Models with Applications in Econometrics and Neibouring
Fields, pages 3–32. University of St.Petersburg, St.Petersburg. (In
Russian).
Nicolis, G. and Prigogine, I. (1989). Exploring Complexity. An
Introduction. Freeman, New York.
O’Donovan, T.M. (1983). Short Term Forecasting: An Introduction to
the Box-Jenkins Approach. Wiley, New York.
Osipov, V.F. (1988). Borh-Frenel analysis on local-compact
comutative groups. Leningrad State University, Leningrad. (In
Russian).
Otnes, R.K. and Enochson, L. (1978). Applied Time Series Analysis,
volume 1 of Basic Techniques. Wiley, New York.
Palus, M. and Novotna, D. (1998). Detecting modes with nontrivial
dynamics embedded in colored noise: enhanced monte carlo ssa and the
case of climate oscillations. Physics Letters A, 248:191–202.
Plaut, G. and Vautard, R. (1994). Spells of low-frequency
oscillations and weather regimes in the northern hemisphere.
Journal of the Atmospheric Sciences, 51:210–236.
Priestley, M.B. (1991). Spectral Analysis and Time Series.
Academic Press, London.
Prothero, D.L. and Wallis, K.F. (1976). Modelling macroeconomic time
series (with discussion). Journal of the Royal Statistical Society,
A, 139(Part 4):468–500.
Rao, T.S. (1976). Canonical factor analysis and stationary time
series models. Sankhya: The Indian Journal of Statistics,
38B:256–271.
Rao, T.S. and Gabr, M.M. (1984). An Introduction to Bispectral
Analysis and Bilinear Time Series Models. Springer-Verlag.
Riesz, R. and Sz.-Nagy, B. (1990). Functional analysis. Dover
Publications, New York.
Rozanov, Yu.A. (1967). Stationary random processes. Holden-Day,
San-Francisco.
Sauer, Y., Yorke, J.A. and Casdagli, M. (1991). Embedology.
Journal of Statistical Physics, 65:579–616, 1991.
Schuster, H.G. (1995). Deterministic Chaos. An introduction.
Physik-Verlag, Weinheim, third edition.
Takens, F. (1981). Detecting strange attractors in turbulence.
In D.A.Rand and L.-S.Young, editors, Dynamical Systems and Turbulence,
volume 898 of Lecture Notes in Mathematics, pages 366–381.
Springer-Verlag, Berlin.
Tong, H. (1993). Non-linear Time Series: A Dynamical System
Approach. Oxford University Press, Oxford.
Tukey, J.W. (1977). Exploratory Data Analysis. Addison-Wesley,
Reading.
Varadi, F. Pap, J.M., Ulrich, R.K., Bertello, L. and Henney, C.J.
(1999). Searching for signal in noise by random-lag singular
spectrum analysis. The Astrophysical Journal, 526:1052–1061.
Vautard, R. and Ghil, M. (1989). Singular-spectrum analysis in
nonlinear dynamics, with applications to paleoclimatic time series.
Physica D, 35:395–424.
Vautard, R., Yiou, P. and Ghil, M. (1992). Singular-spectrum
analysis: A toolkit for short, noisy chaotic signals. Physica D,
58:95–126.
Wei, W.W.S. (1990). Time Series Analysis: Univariate and
Multivariate Methods. Addison-Wesley, New York.
Weigend, A.S. and Gershenfeld, N.A. (1993) (eds). Time Series
Prediction: Forecasting the Futiurte and Understanding the Past.
Addison-Wesley, Reading.
Yiou, P., Baert, E. and Loutre, M.F. (1996). Spectral analysis of
climate data. Surveys in Geophysics, 17:619–663.
Yiou, P., Sornette, D. and Ghil, M. (2000). Data-adaptive wavelets
and multi-scale singularspectrum analysis. Physica D, 142:254–290.