Multivariate out-of-sample tests for Granger causality

A time series is said to Granger cause another series if it has incremental predictive power when forecasting it. While Granger causality tests have been studied extensively in the univariate setting, much less is known for the multivariate case. Multivariate out-of-sample tests for Granger causality are proposed and their performance is measured by a simulation study. The results are graphically represented by size-power plots. It emerges that the multivariate regression test is the most powerful among the considered possibilities. As a real data application, it is investigated whether the consumer confidence index Granger causes retail sales in Germany, France, the Netherlands and Belgium.     

Causality analysis of neural connectivity: New tool and limitations of spectral Granger causality

Granger causality (GC) is one of the most popular measures to reveal causality influence of time series based on the estimated linear regression model and has been widely applied in economics and neuroscience due to its simplicity, understandability and easy implementation. Especially, its counterpart in frequency domain, spectral GC, has recently received growing attention to study causal interactions of neurophysiological data in different frequency ranges. In this paper, on the one hand, for one equality in the linear regression model (frequency domain) we point out that all items at the right-hand side of the equality make contributions (thus have causal influence) to the unique item at the left-hand side of the equality, and thus a reasonable definition for causality from one variable to another variable (i.e., the unique item) should be able to describe what percentage the variable occupies among all these contributions. Along this line, we propose a new spectral causality definition. On the other hand, we point out that spectral GC has its inherent limitations because of the use of the transfer function of the linear regression model and as a result may not reveal real causality at all and lead to misinterpretation result. By one example we demonstrate that the results of spectral GC analysis are misleading but the results from our definition are much reasonable. So, our new tool may have wide potential applications in neuroscience.     

矢量格兰杰因果关系及其在复杂网络中的应用

自从格兰杰1969年提出因果关系的概念之后,格兰杰因果关系在构造生物网络(基因网络、蛋白质网络、神经网络)的结构方面的应用越来越广泛,但是它只能用于研究单个节点和单个节点之间的内在联系。而实际的生物网络由于基因、神经元等的相互作用,往往呈现出非常复杂的网络结构,需要研究网络节点构成的组与组之间的内在联系。将格兰杰因果关系进行推广,得到矢量格兰杰因果关系的研究方法,并通过两个模拟的例子验证了方法的有效性。     

运用偏相关因果分析推断神经网络的真实结构

自从格兰杰提出因果关系的概念之后,格兰杰因果关系在构造神经网络的结构方面的应用越来越广泛,因为它可以得到神经网络的一个有向图。对于只有两个神经元的神经元网络,可以用通常的格兰杰因果关系去分析它们谁是因,谁是果。对于三个神经元以上的神经网络,由于神经元之间存在间接的作用,就不能象对两个神经元直接运用格兰杰因果关系去研究它们之间的结构了,而要用偏相关因果关系进行分析。论文介绍了偏相关因果关系的基本概念,并对一个模拟的三个神经元的网络比较了格兰杰因果关系和偏相关因果关系的区别。     

频域格兰杰因果关系及其在信号处理中的应用

自从格兰杰1969年提出因果关系的概念之后,格兰杰因果关系在信号处理、计算神经科学等许多领域的应用越来越广。人们可以利用格兰杰因果关系来分析多个变量之间的直接的相互作用,从而进一步研究各类变量之间的内在联系。以往都是在时域空间进行分析的,也就是说分析的对象都是时间序列数据,研究这些变量之间随着时间的变化是如何联系的。在时域空间的基础上,进一步从频域空间上对变量进行研究,分析在哪个频率段上变量之间存在相互作用,所得到的结论当然更具有意义。     

Predicting software defects with causality tests

In this paper, we propose a defect prediction approach centered on more robust evidences towards causality between source code metrics (as predictors) and the occurrence of defects. More specifically, we rely on the Granger causality test to evaluate whether past variations in source code metrics values can be used to forecast changes in time series of defects. Our approach triggers alarms when changes made to the source code of a target system have a high chance of producing defects. We evaluated our approach in several life stages of four Java-based systems. We reached an average precision greater than 50% in three out of the four systems we evaluated. Moreover, by comparing our approach with baselines that are not based on causality tests, it achieved a better precision.     

非线性格兰杰因果关系在计算神经科学的应用

自从格兰杰1969年提出因果关系的概念之后,格兰杰因果关系的应用越来越广泛,但都是用来分析线性时间序列数据之间的内在联系。将线性格兰杰因果关系推广到非线性的情形,首先利用核函数的方法建立非线性时间序列模型,再按照线性格兰杰因果关系的基本思想定义非线性格兰杰因果关系,最后通过一个模拟的例子验证该方法的有效性。模拟数据的结果表明,该方法能有效地分析非线性数据之间的内在联系。     

Unbalance estimation using linear and nonlinear regression

This paper considers the problem of unbalance estimation of rotating machinery. It is formulated as a parameter estimation problem, where the unknowns enter nonlinearly in a regression model. By use of a certain method, the problem can be reformulated as a linear estimation procedure with a closed form solution. This procedure is sometimes known as the influence coefficient method. In its derivation, no special treatment is devoted to disturbing terms and imperfections in the model. Therefore, a novel method is derived which takes disturbances into account, leading to a nonlinear estimator.The two procedures are compared and analyzed with respect to their statistical accuracy. Using the example of unbalance estimation of a separator, the nonlinear approach is shown to give superior performance.     

Observability functions for linear and nonlinear systems

A generalization of the zero-state observability function is considered for nonlinear systems. The linear time-invariant case is considered as an application in model reduction problems.     

Spurious Granger causality between a broken-trend stationary process and a stochastic trend process

This paper examines spurious Granger causality between a trend stationary process with structural breaks and a stochastic trend process. Monte Carlo simulations show that whether or not there are deterministic variables in the testing models, the sample size and the parameter values of the data generation process can affect the empirical frequencies of spurious Granger causality relations in different degrees. The analysis also points out that an alternative rank-based causality test method can avoid the risk of spurious causality to some extent by adopting an intercept and deterministic trend term in the testing regressions.     

On system gains for linear and nonlinear systems

System gains, and bounds for system gains, are determined for stable linear and nonlinear systems in different signal setups which include ℓ_p signal setups and certain persistent signal generalizations of the ℓ₂ and ℓ₁ signal setups. These results show that robust H_∞ and ℓ₁ control generalize to very versatile persistent signal settings. Relationships between different system gains are also derived. Finally, an application of nonlinear system gain bounds is given by establishing induced ℓ_∞ modelling error bounds for a class of (generalized) piecewise linear systems approximated with simpler linear time-invariant models.     

Determination of ECoG information flow activity based on Granger causality and Hilbert transformation

Analysis of directional information flow patterns among different regions of the brain is important for investigating the relation between ECoG (electrocorticographic) and mental activity. The objective is to study and evaluate the information flow activity at different frequencies in the primary motor cortex. We employed Granger causality for capturing the future state of the propagation path and direction between recording electrode sites on the cerebral cortex. A grid covered the right motor cortex completely due to its size (approx. 8 cm × 8 cm) but grid area extends to the surrounding cortex areas. During the experiment, a subject was asked to imagine performing two activities: movement of the left small finger and/or movement of the tongue. The time series of the electrical brain activity was recorded during these trials using an 8 × 8 (0.016–300 Hz band with) ECoG platinum electrode grid, which was placed on the contralateral (right) motor cortex. For detection of information flow activity and communication frequencies among the electrodes, we have proposed a method based on following steps: (i) calculation of analytical time series such as amplitude and phase difference acquired from Hilbert transformation, (ii) selection of frequency having highest interdependence for the electrode pairs for the concerned time series over a sliding window in which we assumed time series were stationary, (iii) calculation of Granger causality values for each pair with selected frequency. The information flow (causal influence) activity and communication frequencies between the electrodes in grid were determined and shown successfully. It is supposed that information flow activity and communication frequencies between the electrodes in the grid are approximately the same for the same pattern. The successful employment of Granger causality and Hilbert transformation for the detection of the propagation path and direction of each component of ECoG among different sub-cortex areas were capable of determining the information flow (causal influence) activity and communication frequencies between the populations of neurons successfully.     

On linear models for nonlinear systems

Best linear time-invariant (LTI) approximations are analysed for several interesting classes of discrete nonlinear time-invariant systems. These include nonlinear finite impulse response systems and a class of nonsmooth systems called bi-gain systems. The Fréchet derivative of a smooth nonlinear system is studied as a potential good LTI model candidate. The Fréchet derivative is determined for nonlinear finite memory systems and for a class of Wiener systems. Most of the concrete results are derived in an ?_∞ signal setting. Applications to linear controller design, to identification of linear models and to estimation of the size of the unmodelled dynamics are discussed.     

Ultimate boundedness control of linear systems with band-bounded nonlinear actuators and additive measurement noise

For linear systems driven by band-bounded nonlinear actuators, a set of linear matrix inequality (LMI) based sufficient conditions are derived for finding state feedback controllers which assure ultimate boundedness of state trajectories. Besides actuator nonlinearity, it is assumed that additive noise exists when state variables are measured for feedback. The purpose is to minimize the ultimate boundedness region while tolerating noise of the largest magnitude. When a state feedback controller is determined for a given system by solving the LMI conditions or by any other means, a less conservative LMI condition is given for further examination of the resultant ultimate boundedness region and tolerable noise magnitude.     

Output feedback nonlinear control for a linear motor in suspension mode

This work extends an existing model of a permanent magnet linear synchronous motor and proposes a nonlinear controller for it. The motor is assumed to operate without mechanical bearings, so that the airgap length varies and requires regulation; accordingly, the extended model captures both longitudinal and normal dynamics. The output feedback control law mainly relies on the recently proposed power integrator method but also uses the standard techniques of exact feedback linearization and Lyapunov redesign. Uniformly bounded stability of the closed-loop system is achieved in spite of cogging forces and end-effects.     

Output feedback pole placement for linear time-varying systems with application to the control of nonlinear systems

The output feedback pole placement problem is solved in an input-output algebraic formalism for linear time-varying (LTV) systems. The recent extensions of the notions of transfer matrices and poles of the system to the case of LTV systems are exploited here to provide constructive solutions based, as in the linear time-invariant (LTI) case, on the solutions of diophantine equations. Also, differences with the results known in the LTI case are pointed out, especially concerning the possibilities to assign specific dynamics to the closed-loop system and the conditions for tracking and disturbance rejection. This approach is applied to the control of nonlinear systems by linearization around a given trajectory. Several examples are treated in detail to show the computation and implementation issues.     

Inference of disjoint linear and nonlinear sub-domains of a nonlinear mapping

This paper investigates new ways of inferring nonlinear dependence from measured data. The existence of unique linear and nonlinear sub-spaces which are structural invariants of general nonlinear mappings is established and necessary and sufficient conditions determining these sub-spaces are derived. The importance of these invariants in an identification context is that they provide a tractable framework for minimising the dimensionality of the nonlinear modelling task. Specifically, once the linear/nonlinear sub-spaces are known, by definition the explanatory variables may be transformed to form two disjoint sub-sets spanning, respectively, the linear and nonlinear sub-spaces. The nonlinear modelling task is confined to the latter sub-set, which will typically have a smaller number of elements than the original set of explanatory variables. Constructive algorithms are proposed for inferring the linear and nonlinear sub-spaces from noisy data.     

Fast approximate identification of nonlinear systems

In this paper, a method is presented to extend the classical identification methods for linear systems towards nonlinear modelling of linear systems that suffer from nonlinear distortions. A well chosen, general nonlinear model structure is proposed that is identified in a two-step procedure. First, a best linear approximation is identified using the classical linear identification methods. In the second step, the nonlinear extensions are identified with a linear least-squares method. The proposed model not only includes Wiener and Hammerstein systems, it is also suitable to model nonlinear feedback systems. The stability of the nonlinear model can be easily verified. The method is illustrated on experimental data.