Dynamic panel data methods for cross-section

panels with an application on a winter tourism

demand model
Franz Eigner
UK Panel data econometrics
Prof. Kunst, SS09
June 21st, 2009
Abstract
This paper deals with theoretical aspects of dynamic panel data anal-
ysis and employs them in a simple application. The theoretical part deals
with the importance of dynamic modelling, the bias of the LSDV estima-
tor and then focuses on the description of consistent estimators. These are
the Anderson/Hsia (1981), the Dierence-GMM from Arellano and Bond
(1991), the System-GMM from Blundell and Bond (1998) and the bias
corrected LSDV estimator suggested by Bruno (2005). It follows an ap-
plication which is taken from Eigner, Toeglhofer, Prettenthaler (2009) and
which models winter tourism demand for 185 ski destinations in Austria,
based on the number of overnight stays from 1973-2006. By using income
and relative purchasing power of the tourists together with snow cover-
age as determinants for tourism demand, both economic and climatologic
aspects are combined in a single framework, based on an autoregressive
distributed lag model. The study especially emphasizes the importance
of climatologic variables in explaining winter tourism demand in Austria.
1
Contents
1 Preliminary considerations 3
1.1 Bias of the LSDV estimator . . . . . . . . . . . . . . . . . . 3
1.2 Advantages of dynamic modelling . . . . . . . . . . . . . . . 4
2 Consistent estimators 4
2.1 First Diererence IV (Anderson/Hsiao, 1981) . . . . . . . . 4
2.2 GMM estimators . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Dierence GMM (Arellano and Bond, 1991) . . . . . 5
2.2.2 System-GMM (Blundell/Bond, 1998) . . . . . . . . . 7
2.2.3 Validity of the instruments . . . . . . . . . . . . . . 7
2.3 Bias corrected LSDV estimator . . . . . . . . . . . . . . . . 8
2.3.1 Bruno (2005) . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Summary of the models . . . . . . . . . . . . . . . . . . . . 9
3 Application - Winter tourism demand model 9
3.1 Data description . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Commands in statistical software packages . . . . . . . . . . 10
3.3 Estimation table of the winter tourism demand model for
Austrian ski destinations from 1973 to 2006 . . . . . . . . . 10
3.4 Interpretation of the results . . . . . . . . . . . . . . . . . . 12
3.5 Final considerations concerning the application . . . . . . . 12
2
1 Preliminary considerations
Dynamic modelling means that the static specication of the linear xed eects
model
y
it
= x
it
+
i
+
it
, u
it
=
i
+
it
is enhanced by including autoregressive coecients (lagged dependent vari-
ables), which allow feedback from current or past shocks to current values of
the dependent variable y
it
, The most trivial case is a AR(1) model specication
y
it
= y
i,t1
+x
it
+
i
+
it
, u
it
=
i
+
it
Dynamic modelling can be seen as adequate in the presence of
temporal autocorrelation in the residuals
it
or
high persistency in the dependent variable y
it
Unfortunately, as will be shown later, the inclusion of an autoregressive coef-
cient makes the LSDV estimator inconsistent and biased. One is therefore
interested in nding a consistent estimator for N and T xed (cross-
section panel), When the only aim is to cope with temporal autocorrelation in
the data, Parks method could be assessed. Parks method is also a dynamic
approach, assuming an autoregressive specication of the error term and esti-
mated by a FGLS procedure. Alternatively, one could get rid of autocorrelation
by using a well known procedure like the Prais Winston Transformation. It is
implemented as a feature in the Panel corrected standard errors model (PCSE)
in Stata (StataCorp, 2007).
1.1 Bias of the LSDV estimator
Strictly speaking, the strict exogeneity assumption of the LSDV estimator
E(
i,t
| x
i
,
i
) = 0, t = 1, . . . , T; i = 1, . . . , N
is obviously violated by inclusion of y
i,t1
. Only a weaker form of exogeniety still
holds, that is sequential exogeneity. It means that variables x
it
and y
i,t1
are
predetermined, therefore only uncorrelated with the subsequent (future) error
disturbances. The LSDV model can be estimated by OLS using the within-
transformed variables. For the within-transformed variable holds:
y
i,t1
= y
i,t1

1
T1
(y
i2
+. . . +y
iT
) and
it
=
it

1
T1
(
i2
+. . . +
iT
)
One can show that the correlation between y
i,t1
and
it
is negative, due to
cor(y
i,t1
,
1
T1

i,t1
) < 0 and cor(
1
T1
y
it
,
it
) < 0, Therefore the LSDV
estimator is inconsistent and biased in dynamic models (for N and xed
T). The bias of the LSDV estimator was rst estimated by Nickell (1981). His
bias approximation follows:

= plim
N
(
lsdv
) =

2

h(, T)
(1
2
x y1
)
2
y1
3

, where =
x y1
/
2
x
h(, T) =
(T1)T+
T
T(T1)(1)
2
and
x y1
=

x y1
/
x

y1
annot.: variables denoted as x and y are within-transformed
Because h(, T) is always positive (when > 0), the LSDV estimate is down-
ward biased. It can be seen that the bias for the autoregressive coecient is
especially severe, when
1. the autoregressive coecient is high
2. the number of time periods T is low
3. the ratio of
2

/
2
y1
is high
Notice that the bias of the exogenous variables is proportional to the bias of the
autoregressive coecient, where simply indicates the regression coecient of
y
i,t1
on x
it
.
1.2 Advantages of dynamic modelling
However, dynamic modelling includes several advantages. One not only takes
into account (temporal) autocorrelation in the residuals, but one is also able to
reduce the amount of potential spurious regression, which may lead to wrong
inferences and inconsistent estimation in static models. Especially in the context
of tourism demand models, static models may lead to an overestimation of the
eects of the exogenous variables.
1
Furthermore the coecient of the lagged
dependent variable itself may be of interest. In the tourism demand context, it
indicates habit formation due to mouth-to-mouth eects.
2 Consistent estimators
This section deals with consistent estimators for dynamic panel data models.
2.1 First Diererence IV (Anderson/Hsiao, 1981)
A specic case of a GMM estimator is the First-Dierence IV. Individual xed
eects
i
are eliminated by dierencing instead of within-transforming.
y
it
= y
i,t1
+x

it
+
i
+
it
y
it
= y
i,t1
+x

it
+
it
In matrix notation:
F
y
= F
y1
+FX +F
1
Spurious regression was also present in the application described in this paper. Because
when using a static specication of the winter tourism demand model, a negative coecient
for snow cover is obtained, probably resulting from a positive evolution of the overstay nights
together with a negative evolution of the snow cover variable. In dynamic specication, the
more intuitive positive coecient is obtained, meaning that higher snow cover leads to higher
winter tourism demand.
4
where F = I
N

F
T
and F
T
=
_
_
_
_
_
1 1 0 . . . 0
0 1 1 . . . 0
.
.
. 0
.
.
.
.
.
.
0
0 0 0 1 1
_
_
_
_
_
However y
i,t1
is correlated with the error term
i,t1
. Consistent estima-
tion is still possible by using the IV method with y
i,t2
as instrument for y
i,t1
, because
2
E(y
i,t2

it
) = 0
Even when all moments are employed, this means if all valid lagged internal
variables are included as instruments, estimation is inecient, because not all
information, e.g.
it
MA(1), is used. Only for the case of homoskedasticity
of the residuals it can be shown that Andersion/Hsiao (1981) is the most ecient
GMM estimator.
2.2 GMM estimators
2.2.1 Dierence GMM (Arellano and Bond, 1991)
Ecient estimates are obtained using a GMM framework. Following moments
are exploited:
E[y
i,ts

it
] = 0 and E[X
i,ts

it
] = 0 for s 2; t = 3, . . . T
X
i
=
2
6
6
6
4
y
i2
y
i1
x

i3
x

i2
y
i3
y
i2
x

i4
x

i3
.
.
.
.
.
.
y
i,T1
y
i,T2
x

iT
x

i,T1
3
7
7
7
5
Z
i
=
2
6
6
6
6
6
4
[y
i1,
x

i1
, x

i2
] 0 0
0
.
.
.
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 [y
i1
, . . . , y
i,T2
, x

i1
, . . . , x

i,T1
]
3
7
7
7
7
7
5
X = (y
1
, X), Z = (Z

1
, Z

2
, . . . , Z

N
)

= (,

)
The idea of the GMM framework is the following: L instruments imply a set of
L moments, i.a. g
i
(

) = y
i,ts

it
, where exogeneity holds when E(g
i
()) = 0.
Each of the L moment equations corresponds to a sample moment g(

) =
1
n

n
i=1
g
i
(

). Estimator is typically obtained by solving g(

) = 0. But here
this is not possible due to model overidentication. More equations are present
than unknown variables, or equivalently speaking, more excluded instruments
than endogenous variables. As a consequence one is not able to solve the equa-
tion exactly. Instead, the criterion J
N
is minimized.
J
N
(

) = g(

W
N
g(

)
or in our notation:
J
N
=
_
1
N
N

i=1

i
Z
i
_

W
N
_
1
N
N

i=1
Z

i
_
2
valid under the assumption, that errors do not depend on time
5
where

W
N
is an estimated weighting matrix. The weighting matrix is the weak
point of the Anderson/Hsiao estimator, which uses the identity matrix as weight-
ing matrix and therefore assumes conditional homoskedasticity. Such a matrix
however overweights moments with higher variances and moments which are
correlated with each other. Optimal weighting and therefore ecient estima-
tion is assessed by using the inverse of the moment covariance matrix:
W
N
= V ar(Z

)
1
= (Z

Z)
1
This procedure is similar to the GLS procedure, but where GLS minimizes the
weighted sum of the second moments of the residuals, GMM minimizes the
weighted sum of the covariance structure of the moments. Unless is known,
ecient GMM is not feasible. A two-step procedure is therefore assessed. First
replace with some simple G (here: assuming
it
i.i.d., equivalent to conditional
homoskedasticity)

W
1N
=
_
N

i=1
Z

i
G
T
Z
i
_
1
= (Z

GZ)
1
where G =
`
I
N
G

and G
T
= F
T
F

T
=
2
6
6
6
6
6
4
2 1 0 0
1 2
.
.
.
0
0
.
.
.
.
.
.
1
0 0 1 2
3
7
7
7
7
7
5
which delivers (consistent) rst-step estimates. Its residuals
1i
are used for
the two-step estimation of

W.

W =
_
N

i=1
Z

i

1i

1i
Z
i
_
1
Ecient estimates for the Di-GMM are then obtained with:

EGMM
=
_
X

Z

WZ

X
_
1
X

Z

WZ

y
One can show that under homoskedasticity one-step estimates are asymptoti-
cally equivalent to two-step estimates. However when there is a high persistency
of y
it
, instruments are weak. (Blundell and Bond, 1998, Kitazawa, 2001). Be-
cause the dierences of a time series which is persistent (near a unit root) are
near to innovations and therefore dicult to instrument. This is known as the
weak instrument problem.
Briey I also discuss the estimation of the variance matrix.
one-step procedure
replacing with a sandwich-type proxy

1
delivers consistent and robust vari-
ances.
d
V ar[

1
] =

Z

W
1
Z

1
X

Z

W
1
Z

1
Z

W
1
Z

Z

W
1
Z

1
two-step procedure
6
using the optimal weighting matrix W = (Z

Z)
1
, above formula reduces to
d
V ar[

2
] =

Z(Z

1
Z)
1
Z

1
however:

V ar[

2
] can be heavily downward biased. For many years econome-
tricians therefore printed out the t-statistics of the one-step procedure, when
estimating two-step GMMs. However nowadays, windmeyers correction (2005)
is mostly used for two-step t-statistics which have proven to perform very well
in Monte Carlo simulations.
2.2.2 System-GMM (Blundell/Bond, 1998)
It was invented to tackle the weak instrument problem. It builds up a system of
two equations, the level equation and the dierence equation. Endogenous
variables in the level equation are instrumented by lagged dierences. Summa-
rizing, where Arellano-Bond instruments dierences [. . . ] with levels, Blundell-
Bond instruments levels with dierences. [...] For random walklike variables,
past changes may indeed be more predictive of current levels than past levels
are of current changes (Roodman, 2006). Following additional moments are
explored:
E[y
i,t1
(
i
+
it
)] = 0 and E[X
i,t1
(
i
+
it
)] = 0, for t = 3, . . . , T
In matrix notation:
X
i
=
_

_
y
i2
y
i1
x

i3
x

i2
y
i3
y
i2
x

i4
x

i3
.
.
.
.
.
.
y
i,T1
y
i,T2
x

iT
x

i,T1
y
i2
x

i2
.
.
.
.
.
.
y
i,T1
x

iT
_

_
Z
i
=
_
Z
D
i
0
0 Z
L
i
_
Z
L
i
=
_

_
[y
i2
, x

i2
, x

i3
] 0 0
0
.
.
.
0
.
.
.
.
.
.
.
.
.
0
0 0 [y
i2
, . . . , y
i,T2
, x

i,2
, . . . , x

i,T
]
_

_
2.2.3 Validity of the instruments
Performance depends strongly on the validity of the instruments. A valid in-
strumental variable z requires:
1. E[ | z] = 0 (exogeneity)
2. cov(z, x) = 0 (relevance)
These assumptions can be tested, using procedures like the overidentifying re-
strictions test which is also used in the context of IV estimation.
Overidentifying restrictions test (Sargan/Hansen Test)
The assumption of exogeneity of all instruments is tested using the Hansen-
Statistic J(

EGMM
)
2
LK
or the Sargan-Statistic, which, as opposed
7
to Hansen, assumes conditional homoskedasticity. The main problem with
the Hansen-statistic is that it is seriously weakened by the use of a large
amount of instruments. By denition, if the number of instruments is
above the number of (cross-section) observations, it will deliver a p-value
of one, which is meaningless. The Sargan-statistic is known as relatively
robust to the use of a large number of instruments, but may be inconsistent
under heteroskedasticity or at least, tests based on Sargan-statistic have
little power. One could also test subset of instruments, which is especially
helpful in analyzing the new instruments employed for the System-GMM.
This can be done by the Dierence-in-Sargan statistic: DS = S
u
S
r

2
Arellano and Bond - Autocorrelation test
It examines the dierenced residuals to nd autocorrelation of the resid-
uals in levels, which would be an indicator of weak, not exogenous instru-
ments.
It is dicult to detect how many instruments should be used in GMM esti-
mation. Typically one employs more instruments to increase eciency of the
estimation. However it was shown that more instruments also increase the -
nite sample bias (Bun/Kiviet, 2002). Thus it exists a trade-o between small
sample bias and eciency. It is suggested by Hahn that a restricted subset of
instruments should be used, using a number of instruments at least below the
number of observations.
2.3 Bias corrected LSDV estimator
Consistent estimation is obtained by additive bias correction. One could es-
timate LSDV coecients and correct these by subtracting their (estimated)
biases. The nowadays relatively large amount of bias corrected LSDV estima-
tors may be distinguished according to their use of external information. Bias
corrected LSDV estimators, which
use a preliminary consistent estimator
Kiviet (1995)
Hansen (2001)
Hahn and Kuersteiner (2002) - not for short T
Bruno (2005) - for unbalanced panels and short T
do not need a preliminary consistent estimator
Bun/Carree (2005)
2.3.1 Bruno (2005)
This paper will explain briey the procedure suggested by Bruno (2005). Bias
approximations emerge with an increasing level of accuracy:
B
1
= c
1
(

T
1
), B
2
= B
1
+c
2
(N
1

T
1
) and B
3
= B
2
+c
3
(N
1

T
2
)
where c
1
, c
2
and c
3
depend i.a. on
2

and . Because bias approximations

are not yet feasible.
2

and have to obtained from a consistent estimator.

(Anderson/Hsiao, Arellano/Bond, Blundell/Bond)
8
LSDV C
i
= LSDV

B
i
, i = 1, 2 and 3
2.4 Summary of the models
Model Transformation Regressors Consistency
LSDV/FE Within y
i,t1
, x
it
no
Bias corrected LSDV Within y
i,t1
, x
it
yes
First-dierence IV y
i,t1
, x
it
yes
First-dierence GMM y
i,t1
, x
it
yes
System-GMM y
i,t1
, x
it
, y
i,t1
, x
it
yes
Table 1: Summary of the models
There is a vast literature dealing with the performance of these estimators under
a variety of conditions. According to Monte Carlo Simulations, one can state
that GMMs are more adequate for large N dimension and are more accurate in
case of heteroskedasticity, whereas the bias corrected LSDV performs in general
better for small data sets.
3 Application - Winter tourism demand model
Theoretical model descriptions are applied in this chapter on a winter tourism
demand model for 185 Austrian ski destinations. The study is taken from Eigner
et al. (2009).
3.1 Data description
The analyzed panel data set contains the number of overnight stays of tourists
from the 15 most important countries (including Austria) in 185
3
dierent ski-
ing destinations in Austria in the winter season for the time period from 1973
to 2006. Data can be described as typical cross-section panel data, with a mod-
erately large number of cross-section units (N=185), each observed for a small
number of time periods (T=34).
Nights number of overnight stays in winter season
Snow snow cover
GDP income variable
Beds infrastructure variable
PP relative purchasing power
Table 2: Description of the variables
The number of overnight stays measures winter tourism demand, which is
assumed to depend on relative purchasing power (PP)
4
and income (GDP)
5
of
3
Austrian ski resort database. JOANNEUM Research (2008)
4
OECD (2008)
5
OECD (2008)
9
the tourists, on infrastructure, measured by the number of beds (BEDS)
6
and
on the climate variable snow (SNOW)
7
. To account for unobserved time eects,
time dummies are included in the demand function.
3.2 Commands in statistical software packages
In STATA, GMMs can be estimated by the ocal command xtabond (Arel-
lano/Bond) and xtdpdsys (Blundell/Bond) command, where the later is only
available in STATA 10. However more testing procedures are available by using
xtabond2 (Roodman, 2006). The bias corrected LSDV estimator is available
under the command xtlsdvc (Bruno, 2005). The cross section autocorrelation
test by Pesaran (2004) can be assessed using xtcsd.
Using R (2008), a specic package named plm (Yves/Giovanni, 2008) is avail-
able, containing the function pgmm for GMMs. Currently, this function fails
when dealing with large instrument matrices. Furthermore, it is not able to
exactly reproduce the estimation results of the STATA functions.
Current versions of other statistical software packages like SAS, Eviews and
LIMDEP are also able to calculate GMMs for dynamic panel models. Un-
fortunately, for all these programs (together with R and STATA) holds, that
bias corrected LSDV estimators are currently not present as ocial commands
in statistical software packages. Luckily, Bruno (2005) not only brought his
considerations into a paper but also implemented his estimator as an inocal
command for STATA.
3.3 Estimation table of the winter tourism demand model
for Austrian ski destinations from 1973 to 2006
--------------------------------------------------------------------------------------------------------------------
pool fe fe_tw fe_tw_bc diffgmm2 sysgmm2 sysgmm_v sysgmm_g
--------------------------------------------------------------------------------------------------------------------
L.NIGHTS 0.716*** 0.609*** 0.596*** 0.637*** 0.475*** 0.634*** 0.600*** 0.632***
(11.97) (10.54) (10.62) (67.01) (6.35) (11.63) (4.64) (11.78)
L2.NIGHTS 0.215*** 0.174*** 0.187*** 0.161*** 0.166*** 0.163*** 0.268*** 0.179***
(3.92) (3.82) (4.36) (16.72) (5.44) (3.93) (3.69) (3.78)
SNOW / 100 0.067*** 0.076*** 0.070*** 0.071*** 0.071*** 0.097*** 0.153*** 0.095***
(5.70) (6.49) (4.04) (4.44) (3.65) (4.41) (2.70) (5.35)
log(BEDS) 0.086*** 0.113*** 0.132*** 0.119*** 0.202*** 0.202*** 0.134 0.222***
(7.40) (5.53) (5.80) (10.34) (4.41) (6.19) (1.38) (7.87)
log(GDP) 0.039 0.013 0.407*** 0.436*** 0.977 0.700 0.361 0.663***
(0.73) (1.42) (3.29) (5.85) (1.45) (1.49) (0.97) (2.72)
log(PP) -0.041*** -0.035*** -0.030** -0.028** -0.024 -0.010 -0.056 0.010
(-3.35) (-2.76) (-2.31) (-2.36) (-0.38) (-0.16) (-1.17) (0.30)
--------------------------------------------------------------------------------------------------------------------
R2_within 0.776 0.785
corr(x_i,mu_i) 0.954 0.924
sigma_u 0.193 0.174
sigma_e 0.147 0.145
rho 0.632 0.592
Pesaran AR 74.3 2.8
Pesaran p_value 0.000 0.005
t-statistics Robust Robust Robust Corrected Corrected Corrected Corrected
F 15133.0 1129.5 355.7 102.7 37285.5 274410.0 211497.0
diff AR(2) 0.621 0.901 0.345 0.926
Sargan test 0.000 0.000 0.000 0.000
Hansen test 1.000 1.000 0.202 1.000
Diff_Sarg IV 1.000 0.752 1.000
Diff_Sarg GMM 1.000 0.510 1.000
No. of instruments 562 594 147 893
No. of groups 185 185 185 185 185 185 185
No. of observations 5920 5920 5920 5920 5735 5920 5920 5920
--------------------------------------------------------------------------------------------------------------------
* p<0.1, ** p<0.05, *** p<0.01


Table 3: Estimation table of the winter tourism demand model for all estimators
6
Statistik Austria (2008)
7
ZAMG (2009)
10
Equation Type DIFF_GMM SYS_GMM SYS_GMM_valid SYS_GMM_gdp
First
difference
equation
IV

Diff.
(SNOW
log_PP
log_GDP
log_BEDS)
time_dummies
Diff.
(SNOW log_PP
log_GDP
log_BEDS)
time_dummies
- Diff.
(log_BEDS SNOW
log_PP)
time dummies
GMM Lag(2-.).
log_NIGHTS
Lag(2-.).
log_NIGHTS
Lag(4-7).
log_NIGHTS
Lag(2-.).
log_NIGHTS
log_GDP
IV - log_GDP
SNOW log_PP
log_GDP SNOW
log_PP
Level equation
GMM - Diff.Lag.
log_NIGHTS
Diff.(Lag(3).
log_NIGHTS)
Diff.Lag.(log_NIG
HTS log_GDP)

Table 4: Instrument list for Table 3
The OLS coecient of the lagged dependent variable is expected to suer from
an upward bias due to its ignorance of individual specic eects (Hsiao 1986),
whereas the within estimator of the xed eects model is expected to be down-
ward biased (Nickell 1981). [...] According to Blundell and Bond (1995), a
plausible parameter estimate should therefore lie between the within and the
OLS estimate. (Eigner et al., 2009). This initial plausability assumption is
fullled in our estimations. Controlling for time eects in the xed eects model
results in a large increase in the coecient for GDP, making it theoretically more
reasonable, while leaving all other coecients practically unchanged.
The bias corrected estimator produces only minor changes to the two-way xed
eects model, which seems to suer therefore only from minor biases. The dif-
ference GMM suers from a substantial downward bias of the sum of the lagged
dependent variables, which is consistent with the nite sample bias found in
Blundell and Bond (1998) in case of near persistent series, which could be even
greater than the one for the within estimator, especially in the case of weak in-
struments. Indeed this seems to be the case here. The Sargan test indicates for
both models (Di-GMM and System-GMM) that instruments are weak, even
though this test statistic is quite restrictive and has to be interpreted with care
on account of its missing robustness to heteroskedasticity.
The problem of a large amount of instrumental variables, which are weaken-
ing the Hansen-Test and could lead to nite sample bias, is tackled in the
SYS_GMM_v model. This model aims at fullling the overidentifying restric-
tions test, needing less instrumental variables by leaving out the time dummies
as standard instrumentals and using only lag 4 to 7 of the dependent variable
as GMM-instrumentals. Estimations seem to t well of what one is expecting.
However, model specication for SYS_GMM_v is quite arbitrary and the co-
ecients for BEDS and GDP are not signicant, which is unreliable and hints
at specication problems.
Theoretical considerations led to the construction of SYS_GMM_g. Until now
GDP has been dealt as strictly exogenous in our tourism demand model, which
does not have to be plausible. As a higher GDP is expected to increase the
number of overnight stays, increasing tourism earnings lead to an increase of the
11
Austrian GDP, which is relatively strongly weighted (9%) in the GDP variable.
Although GMM estimators are in general assumed to be relatively robust to
endogeneity problems, one could treat GDP not as standard but as GMM style
instrumental variable. Assessing the SYS_GMM_v estimator with all available
moments leads to estimation results which are similar to the bias-corrected two-
way xed eects model, with reliable and signicant coecients for SNOW and
GDP. The main weakness of this model is the use of a vast number of instrumen-
tal variables, which may enhance nite sample bias. It therefore breaks the rule
that the number of groups should be larger than the number of instrumental
variables. As a consequence, due to the weakening of the Hansen Test no clear
statements can be given for the validity of the instruments. One has to accept
that the overidentifying-restrictions test of Sargan is highly rejected, indicating
weak instruments, which is though weakened by heteroskedasticity. However,
because the value of the lagged dependent variable is only slightly above the
one of the bias corrected FE model, nite sample bias seems to be small and
therefore the gain in eciency is preferred here.
3.4 Interpretation of the results
According to estimations of the last model SYS_GMM_g, short-term income
elasticity amounts to 0.55, which would indicate the presence of inelastic demand
to Austrian skiing destinations. However, in the long term elasticity increases to
3.01 and thus is strongly elastic as it is common for luxury goods. Estimations
of the bias corrected FE model reveal a short-term income elasticity of 0.44
and a long-term income elasticity of 2.18. Snow estimation results deliver that
additional 10 days with a snow height of more than 1 cm lead to an increase of
overnight stays to an extent of about 0.7 to 1 percent. At an aggregate level
for the winter season, 79-82 percent of total overnight stays in Austria can be
attributed to habit persistence and/or word-of-mouth eects, indicating a very
good reputation of Austrian skiing destinations.
3.5 Final considerations concerning the application
The biases in the estimates seem to follow the theory. However cross-section
dependency in the data may still worse the estimations. In detail, cross-section
dependence may lead to potential loss in eciency (Phillips and Sul, 2003) and
may even lead to biased and inconsistent estimates. Therefore the regression
model should be extended to account for cross-section dependence, this means
for spatial patterns in the data. This can be done in spatial interaction models
with spatial weighting matrices.
12
References
.
Blundell, Richard & Bond, Stephen (2000): GMM Estimation with persis-
tent panel data: an application to production functions. Econometric Reviews,
Taylor and Francis Journals, vol. 19(3), 321-340.
Bond, Stephen (2002): Dynamic panel data models: a guide to microdata
methods and practice. CeMMAP working papers CWP09/02, Centre for
Microdata Methods and Practice, Institute for Fiscal Studies.
Bruno, G.S.F. (2005): Estimation and inference in dynamic unbalanced
panel data models with a small number of individuals. CESPRI WP n.165 ,
University Bocconi-CESPRI, Milan.
Croissant, Yves & Millo, Giovanni (2008): Panel Data Econometrics
in R: The plm Package. Journal of Statistical Software 27(2). URL:
http://www.jstatsoft.org/v27/i02/.
Eigner, F., Toeglhofer, C., & Prettenthaler, F. (2009). Tourism demand in
Austrian ski destinations. A dynamic panel data approach (in preparation).
Kitazawa, Y. (2001): Exponential regression of dynamic panel data models.
Economics Letters 73, 7-13.
Kiviet, J. F. (1995): On Bias, Inconsistency and Eciency of Various
Estimators in Dynamic Panel Data Models. Journal of Econometrics 68, 5378.
Nickell, Stephen J. (1981): Biases in Dynamic Models with Fixed Eects.
Econometrica, Econometric Society, vol. 49(6), 1417-26, November.
Pesaran, M. H. (2004): General diagnostic tests for cross section depen-
dence in panels. Cambridge Working Papers in Economics, 0435, University of
Cambridge.
R Development Core Team (2008): R: A language and environment for
statistical computing. R Foundation for Statistical Computing,Vienna. URL:
http://www.R-project.org
Roodman, D. (2006): How to Do xtabond2: An Introduction to "Dier-
ence" and "System" GMM in Stata. Working Paper 103. Center for Global
Development, Washington.
StataCorp. (2007): Stata Statistical Software: Release 10. College Station,
TX: StataCorp LP.
Windmeijer, F. (2005): A Finite Sample Correction for the Variance of Lin-
ear Ecient TwoSteps GMM Estimators. Journal of Econometrics, 126, 2551.
13