The Geometric Meaning of the Notion of Joint Unpredictability of a Bivariate VAR(1) Stochastic Process

Transcription

1 Econometrics 2013, 3, ; doi: /econometrics OPEN ACCESS econometrics ISSN Article The Geometric Meaning of the Notion of Joint Unpredictability of a Bivariate VAR(1) Stochastic Process Umberto Triacca Department of Computer Engineering, Computer Science Mathematics, University of L Aquila, Via Vetoio I Coppito, L Aquila, I-67100, Italy; umberto.triacca@ec.univaq.it; Tel.: , Fax.: Received: 21 August 2013; in revised form: 5 November 2013 / Accepted: 5 November 2013 / Published: 14 November 2013 Abstract: This paper investigates, in a particular parametric framework, the geometric meaning of joint unpredictability for a bivariate discrete process. In particular, the paper provides a characterization of the joint unpredictability in terms of distance between information sets in an Hilbert space. Keywords: Hilbert spaces; predictability; stochastic process JEL Classification: C18; C32 1. Introduction Let (Ω, F, P ) be a probability space {y t y 1,t, y 2,t ; t 0, 1,...} a bivariate stochastic process defined on (Ω, F, P ). We consider the differenced process { y t y 1,t, y 2,t ; t 1,...}, where is the first-difference operator. Following Caporale Pittis 1 Hassapis et al. 2, we say that the process { y t ; t 1,...} is jointly unpredictable if E( y t+1 σ(y t,..., y 0 )) 0 t (1) where σ(y t,..., y 0 ) is the σ-field generated by past vectors y i i 0,..., t. The goal of this paper is to show that the notion of joint unpredictability, in a particular parametric framework, can be characterized by a geometric condition. This characterization is given in terms of distance between information sets in an Hilbert space. In particular, we will show that the process { y t ; t 1,...} is jointly unpredictable if only if the information contained in its past is much

2 Econometrics 2013, distant from the information contained in its future. Even if our result is not as general as might seem desirable, we think that the intuition gained from this characterization makes the notion of joint unpredictability more clear. The rest of the paper is organized as follows. Section 2 presents the utilized mathematical framework. Sections 3 presents the geometric characterization. Section 4 concludes. 2. Preliminaries Definitions, notation, preliminary results from Hilbert space theory will be presented prior to establish the main result. An excellent overviews of the applications of Hilbert space methods to time series analysis can be found in Brockwell Davis 3. We use the following notations symbols. Let (Ω, F, P ) be a probability space. We consider the Hilbert space L 2 (Ω, F, P ) of all real square integrable rom variables on (Ω, F, P ). The inner product in L 2 (Ω, F, P ) is defined by z, w E(zw) for any z, w L 2 (Ω, F, P ). The space L 2 (Ω, F, P ) is a normed space the norm is given by w E(w 2 ) 1/2. The distance between z,w L 2 (Ω, F, P ) is d(z,w) z w. A sequence {z n } L 2 (Ω, F, P ) is said to converge to a limit point z L 2 (Ω, F, P ) if d(z n, z) 0 as n. A point z L 2 (Ω, F, P ) is a limit point of a set M (subset of L 2 (Ω, F, P )) if it is a limit point of a sequence from M. In particular, M is said to be closed if it contains all its limit points. If S is a arbitrary subset of L 2 (Ω, F, P ), then the set of all α 1 z α h z h (h 1, 2,...; α 1,..., α h arbitrary real numbers; z 1,..., z h arbitrary elements of S) is called a linear manifold spanned by S is symbolized by sp(s). If we add to sp(s) all its limit points we obtain a closed set that we call the closed linear manifold or subspace spanned by S, symbolized by sp(s). Two elements z, w L 2 are called orthogonal, we write z w, if z, w 0. If S is any subset of L 2 (Ω, F, P ), then we write x S if x s for all s S; similarly, the notation S T, for two subsets S T of L 2 (Ω, F, P ), indicates that all elements of S are orthogonal to all elements of T. For a given z L 2 (Ω, F, P ) a closed subspace M of L 2 (Ω, F, P ), we define the orthogonal projection of z on M, denoted by P (z M), as the unique element of M such that z P (z M) z w for any w M. We remember that if z M, then P (z M) 0. If M N are two arbitrary subsets of L 2 (Ω, F, P ), then the quantity d (M, N) inf { m n ; m M, n N} is called distance between M N. We close this section introducing some further definitions, concerning discrete stochastic processes in L 2 (Ω, F, P ). Let {x t } be a univariate stochastic process. We say that {x t } is integrated of order one (denoted x t I(1)) if the process { x t x t x t 1 } is stationary whereas {x t } is not stationary. We say that the bivariate stochastic process { y t y 1,t, y 2,t } is integrated of order one if y 1,t I(1) y 2,t I(1). A stochastic process {y t } Granger causes another stochastic process {x t }, with respect to a given information set I t that contains at least x t j, y t j, j > 0, if x t can be better predicted by using past values of y than by not doing so, all other information in I t (including the past of x) being

3 Econometrics 2013, used in either case. More formally, we say that {y t } is Granger causal for {x t } with respect to H xy (t) sp {x t, y t, x t 1, y t 1,...} if x t+1 P (x t+1 H xy (t)) 2 < x t+1 P (x t+1 H x (t)) 2 where H x (t) sp {x t, x t 1,...}. Two stochastic processes, {x t }, {y t }, both of which are individually I(1), are said to be cointegrated if there exists a non-zero constant β such that {z t x t βy t } is a stationary (I(0)) process. It is important to note that cointegration between two variables implies the existence of causality (in the Granger sense) between them in at least one direction (see Granger 4). 3. A Geometric Characterization In this section we assume that { y t y 1,t, y 2,t ; t 0, 1,... } be a bivariate stochastic process defined on (Ω, F, P ), integrated of order one, with y 1,0 y 2,0 0, that has a VAR(1) representation y t Ay t 1 + u t (2) where A a 11 a 12 a 21 a 22 is a fixed (2 2) coefficient matrix u t, is i.i.d. with E (u t ) 0 E(u t u t) Σ σ1 2 0 for all t E(u 0 σ2 2 t u s) 0 for s t. In this framework we have that {y 2,t } does not Granger cause {y 1,t } if only if a Similarly, {y 1,t } does not Granger cause {y 2,t } if only if a We observe that the VAR residuals are usually correlated hence the covariance matrix Σ is seldom a diagonal matrix. However, because the main aim of this study is pedagogical, we assume that Σ is diagonal for analytical convenience. We consider the following information sets: I y1 (t+) { y 1,t+1, y 1,t+2,...}, I y2 (t+) { y 2,t+1, y 2,t+2,...}, H y1 (t) sp { y 1,t, y 1,t 1,...} H y2 (t) sp { y 2,t, y 2,t 1,...}. Theorem 3.1. Let y t be a VAR(1) process defined as in (2). The differenced process { y t ; t 1,...} is jointly unpredictable if only if d (I y1 (t+), H y2 (t)) σ y1 d (I y2 (t+), H y1 (t)) σ y2 Theorem 1 provides a geometric characterization of the notion of joint unpredictability of a bivariate process in term of distance between information sets. It is important to note that d (I y1 (t+), H y2 (t)) σ y1 d (I y2 (t+), H y1 (t)) σ y2 Thus we have that the process { y t y 1,t, y 2,t ; t 1,...} is jointly unpredictable if only if the distances d (I y1 (t+), H y2 (t)) d (I y2 (t+), H y1 (t)) achieve their maximum value, respectively.

4 Econometrics 2013, It is intuitive to think that if these distances achieve their maximum value, then σ(y t,..., y 0 ) does not contain any valuable information about the future of the differenced series, y t y 1,t, y 2,t hence these are jointly unpredictable with respect to the information set σ(y t,..., y 0 ), that is E( y t+1 σ(y t,..., y 0 )) 0. We recall that Theorem 1 holds only in a bivariate setting Lemmas In order to prove Theorem 1, we need the following lemmas. Lemma 3.2. Let V be a closed subspace of L 2 (Ω, F, P ) G a subset of L 2 (Ω, F, P ) such that g η R, g G. G V if only if d (G, V ) η. Proof. Focker Triacca (5, p. 767). Lemma 1 establishes a relationship between the orthogonality of sets/spaces in the Hilbert space L 2 (Ω, F, P) their distance. We note that the orthogonality between G V holds if only if the distance d (G, V ) achieves the maximum value. In fact, d (G, V ) can not be greater than η since 0 V. Lemma 3.3. The processes {y 1,t } {y 2,t } are not cointegrated if only if A I. Proof. By (2) we have y 1,t y 1,t a 11 1 a 12 a 21 a 22 1 y 1,t 1 y 2,t 1 + These equations must be balanced, that is the order of integration of (a 11 1)y 1,t 1 + a 12 y 2,t 1 a 21 y 1,t 1 + (a 22 1)y 2,t 1 must be zero. ( ) If A I, since (a 11 1)y 1,t 1 + a 12 y 2,t 1 I(0) a 21 y 1,t 1 + (a 22 1)y 2,t 1 I(0), we can have three cases. Case (1) A a ij, with a ij 0 i, j 1, 2, i j a ii 1 i 1, 2. Case (2) a 11 a 12 A 0 1 with a 11 1 a Case (3) with a 21 0 a A a 21 a 22 In all three cases, there exists at least a not trivial linear combination of the processes {y 1,t } {y 2,t } that is stationary. Thus we can conclude that {y 1,t } {y 2,t } are cointegrated. ( ) If A I, then a 12 a 21 0 so {y 1,t } does not Granger cause {y 2,t } {y 2,t } does not Granger cause {y 1,t }. It follows that {y 1,t } {y 2,t } are not cointegrated.

5 Econometrics 2013, Lemma 3.4. If {y 1,t } {y 2,t } are cointegrated, then a 11 + a 22 1 < 1. Proof. We subtract y 1,t 1, y 2,t 1 from both sides of Equation (2) by obtaining y 1,t y 2,t y 2,t a 11 1 a 12 a 21 a 22 1 α 2 y 1,t 1 y 2,t 1 If {y 1,t } {y 2,t } are cointegrated, we have y 1,t α 1 y 1,t 1 β 1 β 2 + β 1 ϑ 1 ϑ 2 ϑ 1 ϑ 2 α 1 α 2 y 2,t 1 1 β 2 /β 1 y 1,t 1 1 β y 1,t 1 y 2,t 1 (y 1,t 1 βy 2,t 1 ) + y 2,t 1 where β (β 2 /β 1 ) is the cointegration coefficient ϑ 1 β 1 α 1 ϑ 2 β 1 α 2 are the speed of adjustment coefficients. We observe that y 1,t β y 2,t ϑ 1 y 1,t 1 βϑ 2 y 1,t 1 βϑ 1 y 2,t 1 + β 2 ϑ 2 y 2,t 1 + β (3) By rearranging Equation (3) we obtain an AR(1) model for y 1,t βy 2,t : y 1,t βy 2,t δ(y 1,t 1 βy 2,t 1 ) + β where δ 1 + ϑ 1 βϑ 2 a 11 + a Since {y 1,t } {y 2,t } are cointegrated, {y 1,t βy 2,t } is a stationary process so a 11 + a 22 1 < 1 Lemma 3.5. The process { y t y 1,t, y 2,t ; t 0, 1,... } is jointly unpredictable if only if A I Proof. ( ) process { y t y 1,t, y 2,t ; t 0, 1,... } is jointly unpredictable, then E(y t σ(y t 1,..., y 1 )) y t 1 On the other h, since y t Ay t 1 + u t with E (u t ) 0 E(u t u t) Σ for all t E(u t u s) 0 for s t, we have that E(y t σ(y t 1,..., y 1 )) Ay t 1

6 Econometrics 2013, Hence we have so Ay t 1 y t 1 A I ( ) If A I then y t y t 1 + u t with E (u t ) 0 E(u t u t) Σ for all t E(u t u s) 0 for s t, hence we have E(y t σ(y t 1,..., y 0 )) y t 1 Thus we can conclude that the process { y t y 1,t, y 2,t ; t 1,... } is jointly unpredictable. Before to conclude this subsection we observe that Equation (2) can be written in lag operator notation. The lag operator L is defined such that Ly t y t 1. We have that (I AL)y t u t or 1 a 11 L a 12 L a 21 L 1 a 22 L y 1,t y 2,t 3.2. Proof of Theorem 1 Sufficiency. If then, by Lemma 1, we have d (I y1 (t+), H y2 (t)) σ y1 d (I y2 (t+), H y1 (t)) σ y2 I y1 (t+) H y2 (t) I y2 (t+) H y1 (t) Now we assume that a 12 a 21 are not both equal to zero. We can have three cases. Case (1) a 12 0 a This implies that r 1,t (a 11 1)y 1,t + a 12 y 2,t + Thus y 2,t < y 1,t+1, y 2,t > E( y 1,t+1 y 2,t ) (a 11 1)E(y 1,t ) + a 12 E(y 2,t ) + E(+1 ) t (a 11 1)E(y 1,t ) + a 12 E( u 2,s ) (a 11 1)E(y 1,t ) + a 12 σ 2 2 s1

7 Econometrics 2013, Now, we note that Thus but this is absurd since E(y 1,t ) a t 11E(y 1,0 ) 0 < y 1,t+1, y 2,t > a 12 σ I y1 (t+) H y2 (t) Case (2) a 12 0 a In this case we have < y 2,t+1, y 1,t > a 21 σ Again this is absurd since I y2 (t+) H y1 (t) Case (3) a 12 0 a We note that y 1,t (1 a 22 L)γ(L) a 12 Lγ(L) a 21 Lγ(L) (1 a 11 L)γ(L) y 2,t where γ(l) 1 L (1 a 11 L)(1 a 22 L) a 12 a 21 L 2 By Lemma 2, we have that {y 1t } {y 2t } are cointegrated hence the matrix a 11 1 a 12 A I a 21 a 22 1 has rank 1. It follows that a 12 a 21 (1 a 11 )(1 a 22 ) Thus γ(l) 1 L (1 a 11 L)(1 a 22 L) (1 a 11 )(1 a 22 )L 2 1 L (1 L)(1 + L) (1 L)(a 11 + a 22 )L L (a 11 + a 22 )L 1 1 (a 11 + a 22 1)L 1 1 δl where δ a 11 + a Since {y 1t } {y 2t } are cointegrated, by Lemma 3 we have that δ < 1 hence Now, we can have two cases. γ(l) 1 + δl + δ 2 L

8 Econometrics 2013, Case (a) δ 0. In this case we have y 1,t a a 12 1 y 2,t a a 11 1 Thus but this is absurd since < y 1,t+1, y 2,t > a 12 σ2 2 0 < y 2,t+1, y 1,t > a 21 σ1 2 0 I y1 (t+) H y2 (t) I y2 (t+) H y1 (t) Case (b) δ 0. In this case we have Thus y 1,t + (a 11 1) 1 + δ(a 11 1) a a (a 11 1) δ i 1 i + a 1 i 1 i i0 y 2,t + (a 22 1) 1 + δ(a 22 1) a a 21 δ (a 22 1) δ i 1 i + a 21 δ i 1 i i0 < y 1,t+1, y 2,t > (a 11 1)σ1 1 δ a δ (a 22 1) σ 1 δ 2a < y 2,t+1, y 1,t > (a 22 1)σ2 1 δ a δ (a 11 1) σ 1 δ 1a Now, we consider the system 1 + (a22 1) δ 1 δ σ (a 11 1)σ1 1 δ 2 (a 22 1)σ2 1 δ (a11 1) δ 1 δ σ i0 i0 a 12 a The determinant of the matrix 1 + (a22 1) δ 1 δ σ (a 11 1)σ1 1 δ 2 (a 22 1)σ2 1 δ (a11 1) δ 1 δ σ is σ 2 1σ 2 2 ( 1 δ ) 1 + δ

9 Econometrics 2013, Since σ 2 1σ 2 2 > 0 δ 1+δ 1, we have that σ 2 1σ 2 2 ( 1 δ ) δ Thus a 12 0, a 21 0 implies that < y 1,t+1, y 2,t > 0 or < y 2,t+1, y 1,t > 0, but this is absurd since I y1 (t+) H y2 (t) I y2 (t+) H y1 (t) In all Cases (1 3) we obtain an absurd conclusion, thus we can state that Now, we prove that a 11 a We have that a 12 0, a 21 0 y i,t (a ii 1)y i,t + u it i 1, 2 Since the error term u t, is stationary these equations must be balanced, that is the order of integration of y i,t (a ii 1)y i,t must be the same. By the hypothesis that y i,t I(1), it follows that y it I(0) (i.e., stationary) (a ii 1)y i,t is I(1), hence y i,t (a ii 1)y i,t + u i,t i 1, 2 implies that a 11 a Thus A I hence, by Lemma 4, it follows that the process { y t ; t 1,...} is jointly unpredictable. Necessity. If the process { y t ; t 1,...} is is jointly unpredictable, then by Lemma 4 it follows that A I hence y 1,t y 2,t t. This implies that P ( y 1,t+h H y2 (t)) 0 P ( y 2,t+h H y1 (t)) 0 h > 0. Therefore we have that y 1,t+h H y2 (t) y 2,t+h H y1 (t) h > 0. Thus, by Lemma 1, it follows that Theorem 1 is proved. d (I y1 (t+), H y2 (t)) σ y1 d (I y2 (t+), H y1 (t)) σ y2 4. Conclusions In this paper we have considered the following geometric condition concerning the distance between information sets d (I y1 (t+), H y2 (t)) σ y1 d (I y2 (t+), H y1 (t)) σ y2 (4) It says that the distances d (I y1 (t+), H y2 (t)) d (I y2 (t+), H y1 (t)) achieve their maximum value, respectively. Theorem 1 tells us that, under the hypothesis that the process y t follows a bivariate VAR(1) model, the condition Equation (4) represents a geometric characterization of the notion of joint unpredictability. If this condition holds, the processes y 1 y 2 are jointly unpredictable since the past of the bivariate process y t does not contain any valuable information about the future of the differenced series. The information in the past is too far from the future information. Even if the bivariate VAR(1) assumption is far from general, we think that this geometric characterization is useful in order to throw light on the concept of joint unpredictability of a stochastic process.

10 Econometrics 2013, Acknowledgments We thank two anonymous referees for their helpful constructive comments, Fulvia Focker for helpful comments. Conflicts of Interest The author declares no conict of interest. References 1. Caporale, G.M.; Pittis, N. Cointegration predictability of asset prices. J. Int. Money Financ. 1998, 17, Hassapis C.; Kalyvitis, S.; Pittis, N. Cointegration joint efficiency of international commodity markets. Q. Rev. Econ. Financ. 1999, 39, Brockwell, P.J.; Davis, R.A. Time Series: Theory Methods, 2nd ed.; Springer-Verlag: New York, NY, USA, Granger, C.W.J. Developments in the study of cointegrated economic variables. Oxf. Bull. Econ. Stat. 1986, 48, Focker, F.; Triacca, U. Interpreting the concept of joint unpredictability of asset returns: A distance approach. Physica A 2006, 2, c 2013 by the author; licensee MDPI, Basel, Switzerl. This article is an open access article distributed under the terms conditions of the Creative Commons Attribution license (