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.

L<strong>in</strong>k<strong>in</strong>g Granger <strong>Causality</strong> and the Pearl Causal Model with Settable SystemsTo def<strong>in</strong>e recursive settable systems, for b ≥ 0 def<strong>in</strong>e Π [0:b] := Π 0 ∪ ... ∪ Π b−1 ∪ Π b .Def<strong>in</strong>ition 3.3 (Recursive Partitioned Settable System) Let S be a partitioned settablesystem. For b = 0,1,..., B, let Z[0:b] Π denote the vector conta<strong>in</strong><strong>in</strong>g the sett<strong>in</strong>gs ZΠ ifor i ∈ Π [0:b] and tak<strong>in</strong>g values <strong>in</strong> S [0:b] ⊆ Ω 0 × i∈Π[1:b] S i , S [0:b] ∅. For b = 1,..., B andi ∈ Π b , suppose that r Π := {ri Π} is such that the responses YΠ i= X Π i(1,·) are determ<strong>in</strong>edas:= ri Π (ZΠ [0:b−1] ;a).Y Π iThen we say that Π is a recursive partition, that r Π is recursive, and that S :={(A,a),(Ω,F ), (Π,X Π )} is a recursive partitioned settable system or simply that S isrecursive.Example 3.2 is a recursive settable system, as the responses of block 1 depend onthe sett<strong>in</strong>gs of block 0, and the responses of block 2 depend on the sett<strong>in</strong>gs of block 1.Canonical settable systems are recursive settable systems <strong>in</strong> which the sett<strong>in</strong>gs fora given block equal the responses for that block, i.e.,Z[b] Π = YΠ [b] := rΠ [b] (ZΠ [0:b−1];a), b = 1,..., B.Without loss of generality, we can represent canonical responses and sett<strong>in</strong>gs solely asa function of ω 0 , so thatZ Π [b] (ω 0) = Y Π [b] (ω 0) := r Π [b] (ZΠ [0:b−1] (ω 0);a), b = 1,..., B.The canonical representation drops the dist<strong>in</strong>ction between sett<strong>in</strong>gs and responses; wewriteY[b] Π = rΠ [b] (YΠ [0:b−1];a), b = 1,..., B.It is easy to see that the structural VAR of Example 3.3 corresponds to the canonicalrepresentation of a canonical settable system. The canonical responses y 0 and {u t }belong to the first block, and canonical responses y t = (y 1,t ,y 2,t ) belong to block t + 1,t = 1,2,... Example 3.3 implements the time partition, where jo<strong>in</strong>t responses for a giventime period depend on previous sett<strong>in</strong>gs.4. <strong>Causality</strong> <strong>in</strong> Settable Systems and <strong>in</strong> the PCMIn this section we exam<strong>in</strong>e the relations between concepts of direct causality <strong>in</strong> settablesystems and <strong>in</strong> the PCM, specifically the PCM notions of direct cause and controlleddirect effect (Pearl, J. (2000, p. 222); Pearl, J. (2001, def<strong>in</strong>ition 1)). The close correspondencebetween these notions for the recursive systems relevant to Granger causalityenables us to take the first step <strong>in</strong> l<strong>in</strong>k<strong>in</strong>g Granger causality and causal notions <strong>in</strong> thePCM. Section 5 completes the cha<strong>in</strong> by l<strong>in</strong>k<strong>in</strong>g direct structural causality and Grangercausality.13

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

Saved successfully!

Ooh no, something went wrong!