Local Gaussian CrossSpectrum Analysis
Abstract
:1. Introduction
2. Definitions
2.1. The Local Gaussian Correlations
2.2. The Local Gaussian CrossSpectrum
 1.
 With ${G}_{k}^{}$ and ${G}_{\ell}^{}$ as the univariate marginal cumulative distributions of, respectively, ${\left(\right)}_{{Y}_{k,t}^{}}^{}$ and ${\left(\right)}_{{Y}_{\ell ,t}^{}}^{}$, and Φ as the cumulative distribution of the univariate standard normal distribution, define normalized versions ${\left(\right)}_{{Z}_{k,t}^{}}^{}$ and ${\left(\right)}_{{Z}_{\ell ,t}^{}}^{}$ by$$\begin{array}{c}\hfill {\left(\right)}_{{Z}_{k,t}^{}}^{:}t\in \mathbb{Z},\phantom{\rule{2.em}{0ex}}{\left(\right)}_{{Z}_{\ell ,t}^{}}^{:}t\in \mathbb{Z}& .\end{array}$$
 2.
 For a given point $\mathit{v}=({v}_{1}^{},{v}_{2}^{})$ and for each bivariate lag h pair ${\mathit{Z}}_{k\ell :h:t}^{}:=({Z}_{k:t+h}^{},{Z}_{\ell :t}^{})$, a local Gaussian crosscorrelation ${\rho}_{k\ell :\mathit{v}}\left(h\right)$ can be computed based on a fiveparameter local Gaussian approximation of the bivariate density of ${\mathit{Z}}_{k\ell :h:t}^{}$ at $({v}_{1}^{},{v}_{2}^{})$.
 3.
 When ${\sum}_{h\in \mathbb{Z}}^{}\left{\rho}_{k\ell :\mathit{v}}\left(h\right)\right<\infty $, the local Gaussian crossspectrum at the point $\mathit{v}$ is defined as$$\begin{array}{c}\hfill {f}_{k\ell :\mathit{v}}\left(\omega \right):=\sum _{h=\infty}^{\infty}{\rho}_{k\ell :\mathit{v}}\left(h\right)\xb7{e}^{2\pi i\omega h}.\end{array}$$
 1.
 ${f}_{k\ell :\mathit{v}}\left(\omega \right)$ coincides with ${f}_{k\ell}\left(\omega \right)$ for all $\mathit{v}\in {\mathbb{R}}^{2}$ when ${\left(\right)}_{{\mathit{Y}}_{t}^{}}^{}$ is a multivariate Gaussian time series.
 2.
 The following holds when $\stackrel{\u02d8}{\mathit{v}}:=({v}_{2}^{},{v}_{1}^{})$ is the diagonal reflection of $\mathit{v}=({v}_{1}^{},{v}_{2}^{})$:$$\begin{array}{cc}\hfill {f}_{k\ell :\mathit{v}}\left(\omega \right)& =\overline{{f}_{\ell k:\stackrel{\u02d8}{\mathit{v}}}\left(\omega \right)},\hfill \end{array}$$$$\begin{array}{cc}\hfill {f}_{k\ell :\mathit{v}}\left(\omega \right)& ={\rho}_{k\ell :\mathit{v}}\left(0\right)+\sum _{h=1}^{\infty}{\rho}_{\ell k:\stackrel{\u02d8}{\mathit{v}}}\left(h\right)\xb7{e}^{+2\pi i\omega h}+\sum _{h=1}^{\infty}{\rho}_{k\ell :\mathit{v}}\left(h\right)\xb7{e}^{2\pi i\omega h}.\hfill \end{array}$$
2.3. Related Local Gaussian Entities
Algorithm 1 For a sample ${\left(\right)}_{{\mathit{y}}_{t}^{}}^{=}$ of size n from a multivariate time series, an mtruncated estimate ${\widehat{f}}_{k\ell :\mathit{v}}^{m}\left(\omega \right)$ of ${f}_{k\ell :\mathit{v}}\left(\omega \right)$ is constructed by means of the following procedure. 

2.4. Estimation
The Input Parameters and Some Other Technical Details
2.5. Asymptotic Theory for ${\widehat{f}}_{k\ell :\mathit{v}}^{m}\left(\omega \right)$
2.5.1. A Brief Sketch of the Requirements for ${\mathit{Y}}_{t}^{}=({Y}_{1,t}^{},\cdots ,{Y}_{d,t}^{})$
2.5.2. Convergence Theorems for ${\widehat{f}}_{k\ell :\mathit{v}}^{m}\left(\omega \right)$, ${\widehat{\alpha}}_{k\ell :\mathit{v}}^{m}\left(\omega \right)$, and ${\widehat{\varphi}}_{k\ell :\mathit{v}}^{m}\left(\omega \right)$
3. Visualizations and Interpretations
3.1. Sanity Testing the Implemented Estimation Algorithm
3.1.1. Bivariate Gaussian White Noise
3.1.2. Bivariate Local Trigonometric Examples
 Select $r\ge 2$ bivariate time series ${\left(\right)}_{(}^{{C}_{1,i}^{}}$.
 Select a random variable I with values in the set $\{1,\cdots ,r\}$, and use this to sample a collection of indices ${\left(\right)}_{{I}_{t}^{}}^{}$ (that is, for each t, an independent realization of I is taken). Let ${p}_{i}^{}:=\mathrm{P}({I}_{i}^{}=i)$ denote the probabilities for the different outcomes.
 Define ${Y}_{t}^{}$ by means of the equation$${Y}_{1,t}^{}:=\sum _{i=1}^{r}{C}_{1,i}^{}\left(t\right)\xb7\U0001d7d9\{{I}_{t}^{}=i\},$$$${Y}_{2,t}^{}:=\sum _{i=1}^{r}{C}_{2,i}^{}\left(t\right)\xb7\U0001d7d9\{{I}_{t}^{}=i\}.$$The indicator function 𝟙{·} ensures that only one of the bivariate $({C}_{1,i}^{}\left(t\right),{C}_{2,i}^{}\left(t\right))$components contributes for a given value t, that is, it is also possible to write $({Y}_{1,t}^{},{Y}_{2,t}^{})=({C}_{1,{I}_{t}^{}}^{}\left(t\right),{C}_{2,{I}_{t}^{}}^{}\left(t\right))$.
3.2. A Real Multivariate Time Series and a Poorly Fitted GARCHType Model
3.2.1. The DAX–CAC Subset of the EuStockMarketsLogReturns
3.2.2. A Simple CopulaGARCHModel Fitted to the EuStockMarkets LogReturns
 Fit the selected model to the data.
 Perform a local Gaussian spectrum investigation based on simulated samples from the fitted model. The parameters should match those used in the investigation of the original data.
 Compare the plots based on the original data with corresponding plots based on the simulated data from the model. It can be of interest to not only compare the Co, Quad, and Phaseplots, but also include plots that show the traces and the estimated local Gaussian auto and crossspectra.
4. Conclusions
Supplementary Materials
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Notes
1  This multivariate approach was initiated in the first author’s Ph.D. thesis, available at https://bora.uib.no/handle/1956/16950. The version in this paper has been extended with new methods and visualizations that were developed due to review comments related to the univariate theory published in JT22. 
2  This is due to the way the local Gaussian correlation is defined; see Tjøstheim and Hufthammer (2013) for details. 
3  The corresponding coordinates are $(1.28,1.28)$, $(0,0)$, and $(1.28,1.28)$. 
4  The Amplitudeplots are not included here since the interesting details (in most cases) would already have been detected by the other plots. 
5  If you have a black and white copy of this paper, then read “blue” as “light” and “red” as “dark”. 
6  The dotted lines represent the means of the estimated values, whereas the 90% pointwise confidence intervals are based on the 5% and 95% quantiles of these samples. 
7  In this respect, the situation is similar to the detection of a pure sinusoidal for the global spectrum. 
8  The corresponding script in the Rpackage localgaussSpec enables an investigation of all the combinations between DAX, SMI, CAC, and FTSE, but only the DAX–CAC subset will be discussed here. 
9  
10  Use remotes::install_github("LAJordanger/localgaussSpec") to install the package. See Section S6.1 in the online Supplementary Material for further details. 
References
 AïtSahalia, Yacine, Julio CachoDiaz, and Roger J. A. Laeven. 2015. Modeling financial contagion using mutually exciting jump processes. Journal of Financial Economics 117: 585–606. [Google Scholar] [CrossRef]
 Baruník, Jozef, and Tobias Kley. 2019. Quantile coherency: A general measure for dependence between cyclical economic variables. The Econometrics Journal 22: 131–52. [Google Scholar] [CrossRef]
 Berentsen, Geir Drage, Tore Selland Kleppe, and Dag Bjarne Tjøstheim. 2014. Introducing localgauss, an R Package for Estimating and Visualizing Local Gaussian Correlation. jJSTATSOFT 56: 1–18. [Google Scholar] [CrossRef]
 Birr, Stefan, Tobias Kley, and Stanislav Volgushev. 2019. Model assessment for time series dynamics using copula spectral densities: A graphical tool. Journal of Multivariate Analysis 172: 122–46, Dependence Models. [Google Scholar] [CrossRef]
 Bollerslev, Tim, and Eric Ghysels. 1996. Periodic Autoregressive Conditional Heteroscedasticity. Journal of Business & Economic Statistics 14: 139–51. [Google Scholar] [CrossRef]
 Brillinger, David R. 1965. An Introduction to Polyspectra. The Annals of Mathematical Statistics 36: 1351–74. [Google Scholar] [CrossRef]
 Brillinger, David R. 1991. Some history of the study of higherorder moments and spectra. Statistica Sinica 1: 24J. [Google Scholar]
 Brillinger, David R., ed. 1984. The Collected Works of John W. Tukey. Volume I. Time Series: 1949–1964. Introductory Material by William S. Cleveland and Frederick Mosteller. Wadsworth Statistics/Probability Series; Pacific Grove: Wadsworth. [Google Scholar]
 Brockwell, Peter J., and Richard A. Davis. 1986. Time Series: Theory and Methods. New York: Springer, Inc. [Google Scholar]
 Chen, Tianbo, Ying Sun, and TaHsin Li. 2021. A semiparametric estimation method for the quantile spectrum with an application to earthquake classification using convolutional neural network. Computational Statistics & Data Analysis 154: 107069. [Google Scholar] [CrossRef]
 Chung, Jaehun, and Yongmiao Hong. 2007. Modelfree evaluation of directional predictability in foreign exchange markets. Journal of Applied Econometrics 22: 855–89. [Google Scholar] [CrossRef]
 Ciaburro, Giuseppe, and Gino Iannace. 2021. Machine LearningBased Algorithms to Knowledge Extraction from Time Series Data: A Review. Data 6: 55. [Google Scholar] [CrossRef]
 Eyjolfsson, Heidar, and Dag Tjøstheim. 2023. Multivariate selfexciting jump processes with applications to financial data. Bernoulli, in press. [Google Scholar]
 Ghalanos, Alexios. 2022a. rmgarch: Multivariate GARCH Models. R Package Version 1.39. Available online: https://cran.rproject.org/package=rmgarch (accessed on 10 December 2022).
 Ghalanos, Alexios. 2022b. rugarch: Univariate GARCH Models. R Package Version 1.49. Available online: https://cran.rproject.org/package=rugarch (accessed on 10 December 2022).
 Hjort, Nils Lied, and M. C. Jones. 1996. Locally parametric nonparametric density estimation. Annals of Statistics 24: 1619–47. [Google Scholar] [CrossRef]
 Hong, Yongmiao. 1999. Hypothesis Testing in Time Series via the Empirical Characteristic Function: A Generalized Spectral Density Approach. Journal of the American Statistical Association 94: 1201–20. [Google Scholar] [CrossRef]
 Hong, Yongmiao, Jun Tu, and Guofu Zhou. 2007. Asymmetries in Stock Returns: Statistical Tests and Economic Evaluation. The Review of Financial Studies 20: 1547–81. [Google Scholar] [CrossRef]
 Jordanger, Lars Arne, and Dag Tjøstheim. 2022. Nonlinear Spectral Analysis: A Local Gaussian Approach. Journal of the American Statistical Association 117: 1010–27. [Google Scholar] [CrossRef]
 Klimko, Lawrence A., and Paul I. Nelson. 1978. On Conditional Least Squares Estimation for Stochastic Processes. Annals of Statistics 6: 629–42. [Google Scholar] [CrossRef]
 Künsch, Hans R. 1989. The Jackknife and the Bootstrap for General Stationary Observations. The Annals of Statistics 17: 1217–41. [Google Scholar] [CrossRef]
 Li, TaHsin. 2020. From zero crossings to quantilefrequency analysis of time series with an application to nondestructive evaluation. Applied Stochastic Models in Business and Industry 36: 1111–30. [Google Scholar] [CrossRef]
 Li, TaHsin. 2021. Quantilefrequency analysis and spectral measures for diagnostic checks of time series with nonlinear dynamics. Journal of the Royal Statistical Society: Series C (Applied Statistics) 2: 270–90. [Google Scholar] [CrossRef]
 Li, TaHsin. 2022a. Quantile Fourier Transform, Quantile Series, and Nonparametric Estimation of Quantile Spectra. Preprint. [Google Scholar] [CrossRef]
 Li, TaHsin. 2022b. QuantileFrequency Analysis and Deep Learning for Signal Classification. Preprint. [Google Scholar] [CrossRef]
 Li, Zeda. 2023. Robust conditional spectral analysis of replicated time series. Statistics and Its Interface 16: 81–96. [Google Scholar] [CrossRef]
 Otneim, Håkon, and Dag Tjøstheim. 2017. The locally Gaussian density estimator for multivariate data. Statistics and Computing 27: 1595–616. [Google Scholar] [CrossRef]
 Politis, Dimitris N., and Joseph P. Romano. 1992. A General Resampling Scheme for Triangular Arrays of αMixing Random Variables with Application to the Problem of Spectral Density Estimation. The Annals of Statistics 20: 1985–2007. [Google Scholar] [CrossRef]
 R Core Team. 2020. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. [Google Scholar]
 Teräsvirta, Timo, Dag Tjøstheim, and Clive William John Granger. 2010. Modelling Nonlinear Economic Time Series. Oxford: Oxford University Press. [Google Scholar]
 Tjøstheim, Dag, and Karl Ove Hufthammer. 2013. Local Gaussian correlation: A new measure of dependence. Journal of Econometrics 172: 33–48. [Google Scholar] [CrossRef]
 Tjøstheim, Dag, Håkon Otneim, and Bård Støve. 2021. Statistical Modeling Using Local Gaussian Approximation. Cambridge: Academic Press. [Google Scholar]
 Tukey, John W. 1959. An introduction to the measurement of spectra. In Probability and Statistics, The Harald Cramér Volume. Edited by Ulf Grenander. Stockholm: Almqvist and Wiksell, pp. 300–30. [Google Scholar]
 Zhao, Zifeng, Peng Shi, and Zhengjun Zhang. 2022. Modeling multivariate time series with copulalinked univariate dvines. Journal of Business & Economic Statistics 40: 690–704. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. 
© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Jordanger, L.A.; Tjøstheim, D. Local Gaussian CrossSpectrum Analysis. Econometrics 2023, 11, 12. https://doi.org/10.3390/econometrics11020012
Jordanger LA, Tjøstheim D. Local Gaussian CrossSpectrum Analysis. Econometrics. 2023; 11(2):12. https://doi.org/10.3390/econometrics11020012
Chicago/Turabian StyleJordanger, Lars Arne, and Dag Tjøstheim. 2023. "Local Gaussian CrossSpectrum Analysis" Econometrics 11, no. 2: 12. https://doi.org/10.3390/econometrics11020012