NTNU Norwegian University of Science and Technology Faculty of Information Technology and Electrical Engineering Department of Computer Science
Master ’s thesis
Olav Kaada
Continuous lossless compression of streams of high-frequency multivariate financial market data
Master’s thesis in Computer Science Supervisor: Svein Erik Bratsberg June 2020
Olav Kaada
Continuous lossless compression of streams of high-frequency
multivariate financial market data
Master’s thesis in Computer Science Supervisor: Svein Erik Bratsberg June 2020
Norwegian University of Science and Technology
Faculty of Information Technology and Electrical Engineering Department of Computer Science
64
◦
D D = 1
D >1
N D
0 200 400 600 800 1000 1200 1400
N = 273793
N D > 0
Ar,c c r Ar,1
r D Ar,2, ..., Ar,D+1
A1,1, ..., A1,D+1, ..., AN,1, ..., AN,D+1
A1,1, ..., AN,1, ..., A1,D+1, ..., AN,D+1
δi =xi−xi−1
[x1,δ2 =x2−x1, ...,δN =xN −xN−1]
xi =x1+
!i
j=2
δj
δi2 =δi −δi−1 = (xi−xi−1)−(xi−1−xi−2)
[δ1,δ22 =δ2−δ1, ...,δN2 =δN −δN−1]
xi =xref +iδ1+
!i
j=2
(i−j+ 1)δj2
xi =axi−1+bxi−2+...+"
a b "
[0−15]
log2(16) = 4
32 log2(32) = 5
n >0
1 +log2(E) +"log2(n)#=b, E
b → E
log2(E) =N
64
−1
−1
t−1 t0
t1, ..., tn
96
−28 28 −216 216 −232 232 −264 264 0.0
0.2 0.4 0.6 0.8 1.0
N = 1633158
64
−22 22 −24 24 −28 28 −216 216 0.0
0.2 0.4 0.6 0.8 1.0
N = 3266316
−21 21 −22 22 −24 24 −28 28 0.0
0.1 0.2 0.3 0.4
N = 3266316
T ime|Bid P rice|Bid Size|Of f er P rice|Of f er Size,
log2(64) = 6
log2(32) = 5
N N
N + 1
2, ..., N N
1.1M
103 1
2 3
103
103
103
103
64/2.7≈24
1/4 1/2 2.7
1 + 6 = 7
103 1
2 3
103
103
103
103
64/3.0 ≈ 21
1.5
103 2
4 6
103
103
103
103
32/6.6≈5
103 5
10
103
103
103
103
32/11.5≈3
103 2
4 6
103
103
103
103
32/5.0≈6
1.7
5 4
103 5
10
103
103
103
103
32/4.2 ≈ 8
103 1
2 3 4 5
103
103
103
103
192/4.1≈47
103 1
2 3 4 5
103
103
103
103
192/4.4≈ 44
85×103
103 3.0
3.5 4.0 4.5 5.0
103
103
103
103
25×103 100×103
25×103 100×103
4.4
Nblocks >1
103 3.0
3.5 4.0 4.5 5.0
103
103
103
103
D
x1 =x0+δ1, x0
x1
x2 =x1 + (δ1+δ22).
x3 =x2+ (δ1+δ22+δ32).
xi δ1+"i
j=2δj2 xi−1
xi = (x0+δ1) + (δ1+δ22) + (δ1+δ22+δ23) +...+ (δ1+δ22+δ32+...+δ2i),
xi =x0+iδ1+ (i−1)δ22+ (i−2)δ23+...+δi2.
xi =x0+iδ1+
!i
j=2
(i−j+ 1)δj2.
NTNU Norwegian University of Science and Technology Faculty of Information Technology and Electrical Engineering Department of Computer Science
Master ’s thesis
Olav Kaada
Continuous lossless compression of streams of high-frequency multivariate financial market data
Master’s thesis in Computer Science Supervisor: Svein Erik Bratsberg June 2020