Beruflich Dokumente
Kultur Dokumente
Szymon Borak
Rafał Weron
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Motivation 1-2
Motivation
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Motivation 1-3
Agenda
1. Motivation X
5. Conclusions
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-1
Futures and
Forward
Futures,
Markets Financial
Bilateral Forward
and Option
Time
Contracts
Bilateral Markets Day-Ahead Physical delivery
Contracts
Market
Day-Ahead
Intraday
(spot)
Market
Market
Balancing
Balancing
Market
Market
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-2
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-3
Low
demand
curve
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-4
High
demand
curve
Low
demand
curve
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-5
High
demand
curve
Low
demand
curve
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-6
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-7
Futures contracts
Forward contracts
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Nordic power market 2-8
19990104
300
280
260
240
220
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-1
The model
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-2
m
b l is a nonparametric estimator of ml obtained directly from
the data points Xt,j
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-3
Estimation
L
X K
X
Z>
t m(Xt ) = Zt,l al,k ψk (Xt ) = Z>
t Aψ(Xt )
l=0 k=1
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-4
Estimation cont.
X Jt n
T X o2
b > Aψ(X
Yt,j − Z b t,j ) w (Xt,j ) = min
t
b t ,A
Z b
t=1 j=1
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-5
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-6
Identification
If (Z
bt , m)
b are estimates then also
e >Z
(B e −1 m)
bt , B b
define equivalent
estimates.
Here Be is an arbitrary matrix of the
1 0
form Be= , where B is an invertible matrix.
0 B
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
The Dynamic Semiparametric Factor Model 3-7
Identification cont.
The estimates can be shifted by a constants
L
X L
X
m0 (x) + Zt,l ml (x) = m0 (x) + {(Zt,l − cl )ml (x) + cl ml (x)}
l=1 l=1
XL L
X
new
= m0 (x) + cl ml (x) + Zt,l ml (x)
l=1 l=1
L
X
= m0 (x)new + new
Zt,l ml (x).
l=1
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-1
The data
Nord Pool futures and forward prices from the period Jan. 4,
1999 – Dec 30, 2004, i.e. 1487 (business) days
This leaves us with 1487 − 500 − 125 = 862 days for which
the procedure (calibration + forecast) is repeated
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-2
1000 1000
0 0
−1000 −1000
−2000 −2000
−3000 −3000
Jan99 Jan00 Jan01 Jan02 Jan99 Jan00 Jan01 Jan02
DSFM PCA
0.1 0.1
0.05 0.05
0 0
−0.05 −0.05
−0.1 −0.1
100 200 300 400 500 600 700 100 200 300 400 500 600 700
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-3
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-4
In-sample errors
16000
PCA(3)
14000 DSFM(3)
12000
10000
PCA(4)
PCA(5)
8000
PCA(6)
DSFM(4)
6000
DSFM(5)
4000
DSFM(6)
2000
0
0 200 400 600 800 1000
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-5
14000
12000
10000
8000
6000
4000
2000
0
DSFM(3) DSFM(4) DSFM(5) DSFM(6) PCA(3) PCA(4) PCA(5) PCA(6)
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-6
Forecasting setup
For a 500 day window the DSFM and PCA models (with
L = 3, . . . , 6 factors) are calibrated
For l = 2, 3, Z
bt,l generally exhibit a seasonal pattern
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-7
Forecasting models
Trend update (STr1 for DSFM and PTr1 for PCA): the
sinusoidal trend gl is forecasted for l = 2, 3 and added to the
forward price forecast
3
X
YT∗ +h (x) = YT (x) + b l (x) {b
m gl (T + h) − gbl (T )} .
l=2
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-8
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-9
L
X 3
X
YT∗ +h (x) = e∗
Z T +h,l m
b l (x) + m gl (T + h).
b l (x)b
l=0 l=2
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-10
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-11
5 days-ahead forecasts
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-12
25 days-ahead forecasts
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-13
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-14
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Modelling and forecasting 4-15
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Conclusions 5-1
Conclusions
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Conclusions 5-2
Conclusions cont.
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700
Conclusions 5-3
References
19990104
300
280
260
240
200
180
160
140
120
100
80
0 100 200 300 400 500 600 700