Palgrave Handbook of Econometrics: Applied Econometrics

54 Methodology of Empirical Econometric Modeling

1940 1960 1980 2000

–0.10

–0.05

0.00

0.05

0.10 Δ^ef

Δef

1940 1960 1980 2000

–1

0

1

residuals forecast errors

1990 1995 2000

–0.025

0.000

0.025

~Δe

f

Δef

–3 –2 –1 0 1 2 3

0.2

0.4

0.6

residuals

N(0,1)

Figure 1.11 New model on revised data fitted and actual values, residuals and forecasts
foref,t

1.8 Automatic modeling of a VAR 4 ( 25 )

“The Eighth Square at last!” she cried as she bounded across...“Oh, how

glad I am to get here! And what is this on my head?” she exclaimed...It

was a golden crown. (Quote from Alice in Lewis Carroll, 1899)

To illustrate that automatic modeling is not restricted to single equations (see,
e.g., Krolzig, 2003), we now model the four variables in section 1.4.1.1, namely
industrial output per capita,yc,t, numbers of bankruptcies,bt, and patents,pt,
and real equity prices (deflated by a cost of living index),et, using a VAR with 25
lags, augmented by impulse saturation over the common sampleT=1757–1989 at
α=0.0025 (so on average about one variable will be retained by chance as there are
337 candidates in the initial general model). The marginal criticalt-ratio is about
3.1, and only about 3 regressors (other than impulses) were near or below that in the
four finally-selected models. Most diagnostic tests were insignificant in those final
models (but not computable at the start). The entire exercise took under two hours,
including this write-up: technical progress in undertaking empirical econometrics
is huge, as such an analysis would have been simply impossible (conceptually and
practically) when I first started empirical modeling in 1967.

yc,t=−0.128

(0.031)

yc,t− 1 + 0.143

(0.037)

yc,t− 5 − 0.152

(0.037)

yc,t− 19 + 0.135

(0.034)

yc,t− 23

+ 0.016

(0.004)

pt− 0.016

(0.005)

bt− 3 + 0.081

(0.013)

et− 0.084

(0.013)

et− 2