The STATESPACE Procedure

Example 28.1 Series J from Box and Jenkins

This example analyzes the gas furnace data (series J) from Box and Jenkins. (The data are not shown; see Box and Jenkins (1976) for the data.)

First, a model is selected and fit automatically using the following statements.

title1 'Gas Furnace Data';
title2 'Box & Jenkins Series J';
title3 'Automatically Selected Model';

proc statespace data=seriesj cancorr;
   var x y;
run;

The results for the automatically selected model are shown in Output 28.1.1.

Output 28.1.1: Results for Automatically Selected Model

Gas Furnace Data
Box & Jenkins Series J
Automatically Selected Model

The STATESPACE Procedure

Number of Observations 296

Variable Mean Standard
Error
x -0.05683 1.072766
y 53.50912 3.202121

Gas Furnace Data
Box & Jenkins Series J
Automatically Selected Model

The STATESPACE Procedure

Information Criterion for Autoregressive Models
Lag=0 Lag=1 Lag=2 Lag=3 Lag=4 Lag=5 Lag=6 Lag=7 Lag=8 Lag=9 Lag=10
651.3862 -1033.57 -1632.96 -1645.12 -1651.52 -1648.91 -1649.34 -1643.15 -1638.56 -1634.8 -1633.59

Schematic Representation of Correlations
Name/Lag 0 1 2 3 4 5 6 7 8 9 10
x +- +- +- +- +- +- +- +- +- +- +-
y -+ -+ -+ -+ -+ -+ -+ -+ -+ -+ -+
+ is > 2*std error,  - is < -2*std error,  . is between



Output 28.1.2: Results for Automatically Selected Model

Schematic Representation of Partial
Autocorrelations
Name/Lag 1 2 3 4 5 6 7 8 9 10
x +. -. +. .. .. -. .. .. .. ..
y -+ -- -. .+ .. .. .. .. .. .+
+ is > 2*std error,  - is < -2*std error,  . is between

Yule-Walker Estimates for Minimum AIC
  Lag=1 Lag=2 Lag=3 Lag=4
  x y x y x y x y
x 1.925887 -0.00124 -1.20166 0.004224 0.116918 -0.00867 0.104236 0.003268
y 0.050496 1.299793 -0.02046 -0.3277 -0.71182 -0.25701 0.195411 0.133417



Output 28.1.3: Results for Automatically Selected Model

Gas Furnace Data
Box & Jenkins Series J
Automatically Selected Model

The STATESPACE Procedure
Canonical Correlations Analysis

x(T;T) y(T;T) x(T+1;T) Information Criterion Chi-Square DF
1 1 0.804883 292.9228 304.7481 8



Output 28.1.4: Results for Automatically Selected Model

Gas Furnace Data
Box & Jenkins Series J
Automatically Selected Model

The STATESPACE Procedure
Selected Statespace Form and Preliminary Estimates

State Vector
x(T;T) y(T;T) x(T+1;T) y(T+1;T) y(T+2;T)

Estimate of Transition Matrix
0 0 1 0 0
0 0 0 1 0
-0.84718 0.026794 1.711715 -0.05019 0
0 0 0 0 1
-0.19785 0.334274 -0.18174 -1.23557 1.787475

Input Matrix for Innovation
1 0
0 1
1.925887 -0.00124
0.050496 1.299793
0.142421 1.361696



Output 28.1.5: Results for Automatically Selected Model

Variance Matrix for Innovation
0.035274 -0.00734
-0.00734 0.097569



Output 28.1.6: Results for Automatically Selected Model

Gas Furnace Data
Box & Jenkins Series J
Automatically Selected Model

The STATESPACE Procedure
Selected Statespace Form and Fitted Model

State Vector
x(T;T) y(T;T) x(T+1;T) y(T+1;T) y(T+2;T)

Estimate of Transition Matrix
0 0 1 0 0
0 0 0 1 0
-0.86192 0.030609 1.724235 -0.05483 0
0 0 0 0 1
-0.34839 0.292124 -0.09435 -1.09823 1.671418

Input Matrix for Innovation
1 0
0 1
1.92442 -0.00416
0.015621 1.258495
0.08058 1.353204



Output 28.1.7: Results for Automatically Selected Model

Variance Matrix for Innovation
0.035579 -0.00728
-0.00728 0.095577

Parameter Estimates
Parameter Estimate Standard
Error
t Value
F(3,1) -0.86192 0.072961 -11.81
F(3,2) 0.030609 0.026167 1.17
F(3,3) 1.724235 0.061599 27.99
F(3,4) -0.05483 0.030169 -1.82
F(5,1) -0.34839 0.135253 -2.58
F(5,2) 0.292124 0.046299 6.31
F(5,3) -0.09435 0.096527 -0.98
F(5,4) -1.09823 0.109525 -10.03
F(5,5) 1.671418 0.083737 19.96
G(3,1) 1.924420 0.058162 33.09
G(3,2) -0.00416 0.035255 -0.12
G(4,1) 0.015621 0.095771 0.16
G(4,2) 1.258495 0.055742 22.58
G(5,1) 0.080580 0.151622 0.53
G(5,2) 1.353204 0.091388 14.81



The two series are believed to have a transfer function relation with the gas rate (variable X) as the input and the CO$_{2}$ concentration (variable Y) as the output. Since the parameter estimates shown in Output 28.1.1 support this kind of model, the model is reestimated with the feedback parameters restricted to 0. The following statements fit the transfer function (no feedback) model.

title3 'Transfer Function Model';
proc statespace data=seriesj printout=none;
   var x y;
   restrict f(3,2)=0 f(3,4)=0
            g(3,2)=0 g(4,1)=0 g(5,1)=0;
run;

The last two pages of the output are shown in Output 28.1.8.

Output 28.1.8: STATESPACE Output for Transfer Function Model

Gas Furnace Data
Box & Jenkins Series J
Transfer Function Model

The STATESPACE Procedure
Selected Statespace Form and Fitted Model

State Vector
x(T;T) y(T;T) x(T+1;T) y(T+1;T) y(T+2;T)

Estimate of Transition Matrix
0 0 1 0 0
0 0 0 1 0
-0.68882 0 1.598717 0 0
0 0 0 0 1
-0.35944 0.284179 -0.0963 -1.07313 1.650047

Input Matrix for Innovation
1 0
0 1
1.923446 0
0 1.260856
0 1.346332



Output 28.1.9: STATESPACE Output for Transfer Function Model

Variance Matrix for Innovation
0.036995 -0.0072
-0.0072 0.095712

Parameter Estimates
Parameter Estimate Standard
Error
t Value
F(3,1) -0.68882 0.050549 -13.63
F(3,3) 1.598717 0.050924 31.39
F(5,1) -0.35944 0.229044 -1.57
F(5,2) 0.284179 0.096944 2.93
F(5,3) -0.09630 0.140876 -0.68
F(5,4) -1.07313 0.250385 -4.29
F(5,5) 1.650047 0.188533 8.75
G(3,1) 1.923446 0.056328 34.15
G(4,2) 1.260856 0.056464 22.33
G(5,2) 1.346332 0.091086 14.78