11.07.2015 Views

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Popescuvalued vector. A Data Generat<strong>in</strong>g Process is a qu<strong>in</strong>tuple {s a ,p w ,T a ,T w } where T a , T ware f<strong>in</strong>ite time Tur<strong>in</strong>g mach<strong>in</strong>es which perform the follow<strong>in</strong>g operations: Given an <strong>in</strong>putof the <strong>in</strong>compressible str<strong>in</strong>g p w the mach<strong>in</strong>e T w calculates a rational valued matrix w.The mach<strong>in</strong>e T a when given matrices y, a, u, t, w and a positive rational ∆t outputs avector y i+1 which is assigned for future operations to the time t i+1 = max(t) + ∆tThe def<strong>in</strong>ition is somewhat unusual <strong>in</strong> terms of the def<strong>in</strong>ition of stochastic systemsas embodiments of Tur<strong>in</strong>g mach<strong>in</strong>es, but it is quite standard <strong>in</strong> terms of def<strong>in</strong><strong>in</strong>g an<strong>in</strong>novations term w, a probability distribution thereof p w , a state y, a generat<strong>in</strong>g functionp a with parameters a and an exogenous <strong>in</strong>put u. The motivation for us<strong>in</strong>g theterm<strong>in</strong>ology of algorithmic <strong>in</strong>formation theory is to analyse causality assignment as acomputational problem. For reasons of f<strong>in</strong>ite description and computability our variablesare rational, rather than real valued. Notice that there is no real restriction onhow the time series is to be generated, recursively or otherwise. The <strong>in</strong>itial condition<strong>in</strong> case of recursion is implicit, and time is specified as dist<strong>in</strong>ct and <strong>in</strong>creas<strong>in</strong>g but otherwisearbitrarily distributed - it does not necessarily grow <strong>in</strong> constant <strong>in</strong>crements (itis asynchronous). The slight paradox about describ<strong>in</strong>g stochastic dynamical systems<strong>in</strong> algorithmic terms is the necessity of postulat<strong>in</strong>g a random number generator (an oracle)which <strong>in</strong> some ways is our ma<strong>in</strong> tool for abstract<strong>in</strong>g the complexity of the realworld, but yet is a physical impossibility (s<strong>in</strong>ce such an oracle would require <strong>in</strong>f<strong>in</strong>itecomputational time see Li and Vitanyi (1997) for overview). Also, the Tur<strong>in</strong>g mach<strong>in</strong>eswe consider have f<strong>in</strong>ite memory and are time restricted (they implement a predef<strong>in</strong>edmaximum number of operations before yield<strong>in</strong>g a default output). Otherwise the rulesof algebra (s<strong>in</strong>ce they perform algebraic operations) apply normally. The cover of aDGP can be def<strong>in</strong>ed as:Def<strong>in</strong>ition 2 The cover of a Data Generat<strong>in</strong>g Process (DGP) class is the cover of theset of all outputs y that a DGP calculates for each member of the set of admissibleparameters a,u,t,w and for each <strong>in</strong>itial condition y 1 . Two DGPs are stochasticallyequivalent if the cover of the set of their possible outputs (for fixed parameters) is thesame.Let us now attempt to def<strong>in</strong>e a Granger <strong>Causality</strong> statistic <strong>in</strong> algorithmic terms.Allow<strong>in</strong>g for the notation j..k = { j − 1, j − 2..,k + 1,k} if j > k and <strong>in</strong> reverse order ifj < k1i∑︁K(y 1, j | y 1, j−1..1 ,u j−1..1 ) − K(y 1, j | y 2, j−1..1 ,y 1, j−1..1 ,u j−1..1 ) (4)ij=1This differs from Equation (3) <strong>in</strong> two elemental ways: it is not a statement of <strong>in</strong>dependencebut a number (statistic), namely the average difference (rate) of conditional(or prefix) Kolmogorov complexity of each po<strong>in</strong>t <strong>in</strong> the presumed effect vector whengiven both vector histories or just one, and given the exogenous <strong>in</strong>put history. It is ageneralized conditional entropy rate, and may be reasonably be normalized as such:42

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!