Bond University Research Repository Does systematic ... · Systematic sampling induces spurious...
Transcript of Bond University Research Repository Does systematic ... · Systematic sampling induces spurious...
-
Bond UniversityResearch Repository
Does systematic sampling preserve granger causality with an application to high frequencyfinancial data?
Rajaguru, Gulasekaran; Abeysinghe, Tilak; O'Neill, Michael
Published: 01/01/2017
Document Version:Other version
Licence:CC BY-NC-ND
Link to publication in Bond University research repository.
Recommended citation(APA):Rajaguru, G., Abeysinghe, T., & O'Neill, M. (2017). Does systematic sampling preserve granger causality with anapplication to high frequency financial data?. Australian Conference of Economists: Economics for Better Lives,Sydney, Australia. http://ace2017.org.au/wp-content/uploads/2017/07/Gulasekaran-Rajaguru-Full-Paper.pdf
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
For more information, or if you believe that this document breaches copyright, please contact the Bond University research repositorycoordinator.
Download date: 19 Jun 2021
https://research.bond.edu.au/en/publications/16f99087-c3c8-4057-b62d-156a935280b0http://ace2017.org.au/wp-content/uploads/2017/07/Gulasekaran-Rajaguru-Full-Paper.pdf
-
2017 Australian Conference of Economists, Sydney, 19-21 July
2017
Does Systematic Sampling Preserve Granger Causality with an Application
to High Frequency Financial Data?
Tilak Abeysinghe, Michael O'Neill andGulasekaran Rajaguru
-
Motivation
There was a lot of high powered analysis of this topic, but I came away from a reading of it with the feeling that it was one of the most unfortunate turnings for econometricsin the last two decades, and it has probably generated more nonsense results than anything else during that time.
-Pagan (1989)
-
Motivation Financial Development (FD) and Economic
Growth (EG) Controversial causal results
FD EG: McKinnon (1973), King and Levine (1993a,b), Neusser and Kugler (1998), and Levine et al. (2000), Christopoulos and Tsionas (2004), Dimitris and Efthymios (2004), Shan (2006), and Zhang, Wang and Wang (2012)
EG FD: Gurley and Shaw (1967), Goldsmith (1969), Jung (1986) and Liang and Teng (2006)
Bidirectional: Shan, Morris, and Sun (2001) and Demetriades and Hussein (1996), Calderón and Liu (2003), Hassan, Fung (2009), Sanchez and Yu – (2011)and Kar et al. (2011)
-
Literature Review
Univariate ARIMA models: Wei(1990)
Dynamic relationships: Temporal Aggregation weakens the distributed lag relationships, Telser(1967), Zellner and Montmarquette(1971), Sims(1971), Wei and Tiao (1975), Tiao and Wei(1976,1978), Wei and Metha (1980)
Forecasting: Lutkepohl (1987), Marcellino (1999)
Misspecification involved in cross- country regression: Ericsson et al,(2000)
Causality: Wei (1982), Mamingi (1996), Marcellino (1999), Rajaguru (2004), Rajaguru and Abeysinghe ( 2008), Abeysinghe and Rajaguru (2012) and Ashley and Tsang(2014)
Temporal aggregation turns one-way causality into feedback system
Systematic sampling preserves the causal directions
Temporal Aggregation / Systematic Sampling
Unit Roots: Pierce and Snell(1995) Co-integration: Phillips(1991)
Literature ReviewTemporal Aggregation /
Systematic Sampling
Forecasting:
Lutkepohl (1987),
Marcellino (1999)
Causality:
Wei (1982),
Mamingi (1996),
Marcellino (1999),
Rajaguru (2004),
Rajaguru and Abeysinghe ( 2008),
Abeysinghe and Rajaguru (2012)
and Ashley and Tsang(2014)
Misspecification involved in cross- country regression: Ericsson et al,(2000)
Dynamic relationships:
Temporal Aggregation weakens the distributed lag relationships,
Telser(1967), Zellner and Montmarquette(1971), Sims(1971), Wei and Tiao (1975), Tiao and Wei(1976,1978), Wei and Metha (1980)
Univariate ARIMA models: Wei(1990)
Unit Roots:
Pierce and Snell(1995)
Co-integration:
Phillips(1991)
Systematic sampling preserves the causal directions
Temporal aggregation turns one-way causality into feedback system
-
Systematic Sampling• Let ),...,,( 21 ntttt zzzz = , t=1,2,…,T be an
equally spaced n-variate basic disaggregated
series.
• Systematic sampling : ττ mzZ = (τ =1,2,…,N
and T=mN) - sampling from zt at every mth
interval (m is a positive integer).
·
Let , t=1,2,…,T be an equally spaced n-variate basic disaggregated series.
·
Systematic sampling : (=1,2,…,N and T=mN) - sampling from zt at every mth interval (m is a positive integer).
)
,...,
,
(
2
1
nt
t
t
t
z
z
z
z
=
t
t
m
z
Z
=
t
-
Systematic Sampling• Let wt = ),...,,( 21 nttt www , jt
djt zLw j)1( −= , be a weakly
stationary process.
• where γwii(k) is the autocovariance of the i-th component, wit at
lag k.
• γwij(k) is the cross covariance between i-th and j-th components
• γwii(0) is the variance of the i-th series
·
Let wt =, , be a weakly stationary process.
· where wii(k) is the autocovariance of the i-th component, wit at lag k.
· wij(k) is the cross covariance between i-th and j-th components
· wii(0) is the variance of the i-th series
)
,...,
,
(
2
1
nt
t
t
w
w
w
jt
d
jt
z
L
w
j
)
1
(
-
=
-
Relationship between disaggregated and Systematic
Sampled Series
ττττ jmdm
mjdm
jd wLLzLZL jjj ).....1( )1()'1( W 1j
−+++=−=−= . • The dj-th difference of the systematically sampled
series (j-th component) is simply the weighted sum of the dj-th difference of the basic series.
.
· The dj-th difference of the systematically sampled series (j-th component) is simply the weighted sum of the dj-th difference of the basic series.
t
t
t
t
jm
d
m
m
j
d
m
j
d
w
L
L
z
L
Z
L
j
j
j
)
.....
1
(
)
1
(
)
'
1
(
W
1
j
-
+
+
+
=
-
=
-
=
-
Relationship between cross-covariances of disaggregated and
Systematic Sampled SeriesProposition 1
The cross covariance between i-th and j-th components of the
systematically sampled series Wiτ and Wjτ-k can be expressed in terms of cross
covariances of the i-th and j-th components of the basic disaggregated series itw
and jtw , that is,
))1(()......1( ),( )( 12 −+++++== +−− mdmkLLLWWCovk jwij
ddmkji
Wij
ji γγ ττ (1)
))1(()......1( ),( )( 12 −+++++== +−− mdmkLLLWWCovk iwji
ddmkij
Wji
ji γγ ττ (2)
Proposition 1
The cross covariance between i-th and j-th components of the systematically sampled series Wi and Wj-k can be expressed in terms of cross covariances of the i-th and j-th components of the basic disaggregated series and , that is,
(1)
(2)
))
1
(
(
)
......
1
(
)
,
(
)
(
1
2
-
+
+
+
+
+
=
=
+
-
-
m
d
mk
L
L
L
W
W
Cov
k
j
w
ij
d
d
m
k
j
i
W
ij
j
i
g
g
t
t
))
1
(
(
)
......
1
(
)
,
(
)
(
1
2
-
+
+
+
+
+
=
=
+
-
-
m
d
mk
L
L
L
W
W
Cov
k
i
w
ji
d
d
m
k
i
j
W
ji
j
i
g
g
t
t
it
w
jt
w
-
Systematic Sampling and Granger Causality
consider the following bivariate VAR(1) system with )(~ 11 dIz t
and )(~ 22 dIz t such that itdit zLw i)1( −= for i=1,2:
+
=
−
−
t
t
t
t
t
t
ee
ww
ww
2
1
12
11
2221
1211
2
1
ϕϕϕϕ
,
22
21
2
1
00
,00
~σ
σN
ee
t
t ,
• 012 ≠ϕ implying Granger causality from 2w to 1w
• 021 ≠ϕ implying Granger causality from 1w to 2w
consider the following bivariate VAR(1) system with and such that for i=1,2:
, ,
·
implying Granger causality from to
·
implying Granger causality from to
it
d
it
z
L
w
i
)
1
(
-
=
÷
÷
ø
ö
ç
ç
è
æ
+
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
=
÷
÷
ø
ö
ç
ç
è
æ
-
-
t
t
t
t
t
t
e
e
w
w
w
w
2
1
1
2
1
1
22
21
12
11
2
1
j
j
j
j
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
2
2
2
1
2
1
0
0
,
0
0
~
s
s
N
e
e
t
t
0
12
¹
j
2
w
1
w
0
21
¹
j
2
w
)
(
~
1
1
d
I
z
t
)
(
~
2
2
d
I
z
t
-
Systematic Sampling and Granger Causality
consider the following bivariate VAR(1) system based on
systematically sampled series:
+
=
−
−
τ
τ
τ
τ
τ
τ
ϕϕϕϕ
2
1
12
11*22
*21
*12
*11
2
1
EE
WW
WW
,
( ) ( )* *11 22 12 12 12 11 11 12
11 122 2
11 22 12 11 22 12
(1) (0) (1) (0) (1) (0) (1) (0)ˆ ˆlim , lim(0) (0) (0) (0) (0) (0)
W W W W W W W W
W W W W W Wp pγ γ γ γ γ γ γ γφ φ
γ γ γ γ γ γ
− −= =
− − ,
( )212221112222221*
21)0()0()0(
)0()1()0()1(ˆlimWWW
WWWW
pγγγ
γγγγϕ−
−= , ( )2122211
12211122*22
)0()0()0(
)0()1()0()1(ˆlimWWW
WWWW
pγγγ
γγγγϕ
−
−=
consider the following bivariate VAR(1) system based on systematically sampled series:
,
,
,
(
)
2
12
22
11
12
22
22
21
*
21
)
0
(
)
0
(
)
0
(
)
0
(
)
1
(
)
0
(
)
1
(
ˆ
lim
W
W
W
W
W
W
W
p
g
g
g
g
g
g
g
j
-
-
=
(
)
2
12
22
11
12
21
11
22
*
22
)
0
(
)
0
(
)
0
(
)
0
(
)
1
(
)
0
(
)
1
(
ˆ
lim
W
W
W
W
W
W
W
p
g
g
g
g
g
g
g
j
-
-
=
÷
÷
ø
ö
ç
ç
è
æ
+
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
=
÷
÷
ø
ö
ç
ç
è
æ
-
-
t
t
t
t
t
t
j
j
j
j
2
1
1
2
1
1
*
22
*
21
*
12
*
11
2
1
E
E
W
W
W
W
(
)
(
)
**
1122121212111112
1112
22
112212112212
(1)(0)(1)(0)(1)(0)(1)(0)
ˆˆ
lim,lim
(0)(0)(0)(0)(0)(0)
WWWWWWWW
WWWWWW
pp
gggggggg
ff
gggggg
--
==
--
-
Systematic Sampling and Granger Causality
Parameters of the Systematically Sampled Series
Cross covariance of the systematically sample series
Cross covariance of the basic series
Parameters of the basic series
Parameters of the Systematically Sampled Series
Cross covariance of the systematically sample series
Cross covariance of the basic series
Parameters of the basic series
-
Case 1: No Granger causality between the variables in the disaggregated form
+
=
−
−
t
t
t
t
t
t
ee
ww
ww
2
1
12
11
2221
1211
2
1
ϕϕϕϕ
Here 02112 == ϕϕ and with 012 =σ Proposition 2
If there does not exist Granger causality between the basic series
then the Granger causality between the systematically sampled
series is also absent.
Here and with
Proposition 2
If there does not exist Granger causality between the basic series then the Granger causality between the systematically sampled series is also absent.
0
12
=
s
÷
÷
ø
ö
ç
ç
è
æ
+
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
=
÷
÷
ø
ö
ç
ç
è
æ
-
-
t
t
t
t
t
t
e
e
w
w
w
w
2
1
1
2
1
1
22
21
12
11
2
1
j
j
j
j
0
21
12
=
=
j
j
-
Case 2: Causality between the disaggregated series is one-sided
+
=
−
−
t
t
t
t
t
t
ee
ww
ww
2
1
12
11
2221
1211
2
1
ϕϕϕϕ
Here 12 210 and 0φ φ= ≠ and with 012 =σ Theorem 1
Systematic sampling induces spurious bi-directional Granger causality among
the variables if the uni-directional causality runs from a non-stationary series to
either a stationary or a non-stationary series.
Equivalently, systematic sampling induces spurious bi-directional Granger
causality among the variables if 01 >d .
Here and with
Theorem 1
Systematic sampling induces spurious bi-directional Granger causality among the variables if the uni-directional causality runs from a non-stationary series to either a stationary or a non-stationary series.
Equivalently, systematic sampling induces spurious bi-directional Granger causality among the variables if .
0
12
=
s
0
1
>
d
÷
÷
ø
ö
ç
ç
è
æ
+
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
=
÷
÷
ø
ö
ç
ç
è
æ
-
-
t
t
t
t
t
t
e
e
w
w
w
w
2
1
1
2
1
1
22
21
12
11
2
1
j
j
j
j
1221
0 and 0
ff
=¹
-
Case 3: Causality between the disaggregated series is bi-directional
+
=
−
−
t
t
t
t
t
t
ee
ww
ww
2
1
12
11
2221
1211
2
1
ϕϕϕϕ
Here 12 210 and 0φ φ≠ ≠ and with 012 =σ • Bi-directional causal system becomes uni-directional at the lower level of
aggregation (systematic sampling) and subsequently becomes no-causality
among the variables of interest
• All causal informations concentrate on contemporaneous relationships at
the higher level of aggregation
Here and with
· Bi-directional causal system becomes uni-directional at the lower level of aggregation (systematic sampling) and subsequently becomes no-causality among the variables of interest
· All causal informations concentrate on contemporaneous relationships at the higher level of aggregation
0
12
=
s
÷
÷
ø
ö
ç
ç
è
æ
+
÷
÷
ø
ö
ç
ç
è
æ
÷
÷
ø
ö
ç
ç
è
æ
=
÷
÷
ø
ö
ç
ç
è
æ
-
-
t
t
t
t
t
t
e
e
w
w
w
w
2
1
1
2
1
1
22
21
12
11
2
1
j
j
j
j
1221
0 and 0
ff
¹¹
-
Case 3: Monte Carlo Simulation*12ˆlimϕp from a feedback system when 02211 == ϕϕ , m=3 and 021 == dd
-1-0.5
00.5
1
-1-0.5
0
0.51
-20
-10
0
10
20
12ϕ 21ϕ
)ˆ( *12ϕt
-1-0.5
00.5
1
-1-0.5
0
0.51
-1.5
-1
-0.5
0
0.5
1
1.5
12ϕ
*12ˆlimϕp
21ϕ
m=3 m=3
from a feedback system when , m=3 and
0
2
1
=
=
d
d
*
12
ˆ
lim
j
p
0
22
11
=
=
j
j
21
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-20
-10
0
10
20
)
ˆ
(
*
12
j
t
12
j
12
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-1.5
-1
-0.5
0
0.5
1
1.5
*
12
ˆ
lim
j
p
21
j
-
Case 3: Monte Carlo Simulation*
12ˆ( )t φ from a feedback system when 02211 == ϕϕ , m=12 and 60 and 021 == dd
-1-0.5
00.5
1
-1-0.5
0
0.51
-5
0
5
12ϕ 21ϕ
)ˆ( *12ϕt
-1-0.5
00.5
1
-1-0.5
0
0.51
-4
-2
0
2
4
12ϕ 21ϕ
)ˆ( *12ϕt
m=12 m=60
from a feedback system when , m=12 and 60 and
0
2
1
=
=
d
d
*
12
ˆ
()
t
f
0
22
11
=
=
j
j
12
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-5
0
5
)
ˆ
(
*
12
j
t
21
j
12
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-4
-2
0
2
4
)
ˆ
(
*
12
j
t
21
j
-
Case 3: Monte Carlo Simulationˆ( )t c from a feedback system when 02211 == ϕϕ , m=12 and 60 and 1 2 0d d= =
m=12 m=60
-1-0.5
00.5
1
-1-0.5
0
0.51
-30
-20
-10
0
10
20
30
12ϕ 21ϕ
)ˆ(ct
-1-0.5
00.5
1
-1-0.5
0
0.51
-15
-10
-5
0
5
10
15
21ϕ 12ϕ
)ˆ(ct
Contemporaneous Regression
1 2t tw cw v= +
from a feedback system when , m=12 and 60 and
12
0
dd
==
ˆ
()
tc
0
22
11
=
=
j
j
12
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-30
-20
-10
0
10
20
30
)
ˆ
(
c
t
21
j
21
j
-1
-0.5
0
0.5
1
-1
-0.5
0
0.5
1
-15
-10
-5
0
5
10
15
)
ˆ
(
c
t
12
j
Contemporaneous Regression
12
tt
wcwv
=+
-
Applications
Bi-directional Granger causality has been highlighted in studies using high frequency data (Frijns et al, 2015; Bollen, O’Neill and Whaley, 2016).
Data sourced from Jan 2010 to Dec 2014 from Thompson Reuters SIRCA portal and Bloomberg:-US equities: S&P 500 index (SPX). -Equity index futures: E-Mini futures index (SC1/ES1). -“Investor fear gauge”: CBOE Volatility Index (VIX).-Futures on VIX: S&P VIX futures short term index
(SPVXSTR/VST)
-
Application 1: SPX vs VIX I(0)/I(0)
1-minute 5-minutes 10-minutes
Both SPX VIX None Both SPX VIX None Both SPX VIX None
15 S
ec Both 63 38 20 3 42 33 16 33 26 12 4 82
SPX 81 590 14 96 40 388 8 345 22 107 0 652 VIX 11 12 59 15 5 5 34 53 0 1 9 87 None 7 19 21 200 2 11 10 224 0 3 4 240
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
+
=
−
−
t
t
t
t
t
t
ee
VIXSPX
VIXSPX
2
1
1
1
2221
1211
ϕϕϕϕ
1-minute
5-minutes
10-minutes
Both
SPX
VIX
None
Both
SPX
VIX
None
Both
SPX
VIX
None
15 Sec
Both
63
38
20
3
42
33
16
33
26
12
4
82
SPX
81
590
14
96
40
388
8
345
22
107
0
652
VIX
11
12
59
15
5
5
34
53
0
1
9
87
None
7
19
21
200
2
11
10
224
0
3
4
240
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
-
Application 2:VIX vs SPVXSTR I(0)/I(1)
1-minute 5-minutes 10-minutes
Both VIX VST None Both VIX VST None Both VIX VST None
15 S
ec Both 566 118 322 23 32 176 335 486 14 226 143 646
VIX 2 57 1 2 0 49 0 13 0 35 0 27 VST 9 21 7 2 4 23 4 8 1 17 2 19 None 1 3 2 16 0 2 0 20 0 1 0 21
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
+
∆
=
∆ −
−
t
t
t
t
t
t
ee
VSTVIX
VSTVIX
2
1
1
1
2221
1211
ϕϕϕϕ
1-minute
5-minutes
10-minutes
Both
VIX
VST
None
Both
VIX
VST
None
Both
VIX
VST
None
15 Sec
Both
566
118
322
23
32
176
335
486
14
226
143
646
VIX
2
57
1
2
0
49
0
13
0
35
0
27
VST
9
21
7
2
4
23
4
8
1
17
2
19
None
1
3
2
16
0
2
0
20
0
1
0
21
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
-
Application 3: ES1 vs SPVXSTR I(1)/I(1)
5-minutes 10-minutes
Both ES1 VST NONE Both ES1 VST NONE
1-m
inut
e Both 219 383 237 147 63 176 198 549 ES1 8 54 6 26 5 21 7 61 VST 7 6 19 14 1 2 11 32 NONE 1 0 5 20 0 0 1 25
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
+
∆∆
=
∆∆
−
−
t
t
t
t
t
t
ee
VSTES
VSTES
2
1
1
1
2221
1211 11ϕϕϕϕ
5-minutes
10-minutes
Both
ES1
VST
NONE
Both
ES1
VST
NONE
1-minute
Both
219
383
237
147
63
176
198
549
ES1
8
54
6
26
5
21
7
61
VST
7
6
19
14
1
2
11
32
NONE
1
0
5
20
0
0
1
25
Note: Rejection frequencies of Granger non-causality at the 5% level of significance
-
Conclusion If there does not exist Granger causality between the
basic series then the Granger causality between the systematically sampled series is also absent.
Systematic sampling induces spurious bi-directional Granger causality among the variables if the uni-directional causality runs from a non-stationary series to either a stationary or a non-stationary series.
As m increases VAR(1) becomes VAR(0). All causal inferences concentrate on contemporaneous
relationship among the variables due to systematic sampling of integrated process. However, interestingly, the spurious contemporaneous relationships do not disappear even if the sampling interval is larger.
-
Thank you!
��2017 Australian Conference of Economists, Sydney, 19-21 July 2017��Does Systematic Sampling Preserve Granger Causality with an Application to High Frequency Financial Data?MotivationMotivationSlide Number 4Systematic SamplingSystematic SamplingRelationship between disaggregated and Systematic Sampled SeriesRelationship between cross-covariances of disaggregated and Systematic Sampled SeriesSystematic Sampling and Granger CausalitySystematic Sampling and Granger CausalitySystematic Sampling and Granger CausalityCase 1: No Granger causality between the variables in the disaggregated form�Case 2: Causality between the disaggregated series is one-sidedCase 3: Causality between the disaggregated series is bi-directionalCase 3: Monte Carlo Simulation�Case 3: Monte Carlo Simulation�Case 3: Monte Carlo Simulation�ApplicationsApplication 1: SPX vs VIX I(0)/I(0)Application 2:VIX vs SPVXSTR I(0)/I(1) Application 3: ES1 vs SPVXSTR I(1)/I(1) ConclusionSlide Number 23