Econometric analysis of Internet Banking Development in the New EU Countries

Vilnius University
Faculty of Mathematics and Informatics
Department of Econometric Analysis
Bachelor Thesis
Econometric analysis of Internet Banking Development in the New EU
Member Countries
Ekonometrinė internetinės bankininkystės paplitimo naujosiose ES narėse analizė
Žygimantas Matijošaitis
Vilnius 2016

Vilnius University
Faculty of Mathematics and Informatics
Department of Econometric Analysis
Thesis supervisor lecturer Dr. Dmitrij Celov .............................................................
Reviewer ..................................................................................................................
Thesis defended (Date) ............................................................................................
Thesis evaluation......................................................................................................
Registered number...................................................................................................
Date of submission to the department.....................................................................

1
Abstract
This thesis aims to study the diffusion process of internet banking in the new EU member states
(joined 2004-2007). The countries were selected on the premise that the economic development in
each state should be homogenous. Moreover, for the selected timeframe these countries had well-
balanced datasets, thus, allowing us to make reliable estimates.
Generalised Bass Diffusion model was studied under panel data modelling frameworks. Due to Bass
Diffusion model formulation, differenced data was used and, naturally, first-differences panel data
models were not considered. Furthermore, because of software limitations (specifically, in R and
EViews) Bass Diffusion formula could not be implemented under a Random Effects (RE) settings and
thus RE models were not considered either. Three special cases of Panel data analysis were studied:
Pooled LS, Fixed Individual Effects and Fixed Period Effects. Different cases of the stated models
were estimated: models with Restricted Percentage of Eventual Adopters, restricted rates of
innovators and imitators and eventually – models with all of the Bass coefficients unrestricted.
Fixed Period Effects were rejected in the initial modelling phase – no further analysis and
comparisons were needed. Out of the remaining two, both pooled LS and Fixed Individual Effects
models were selected with restrictions on percentage of eventual adopters (95%). After F and Chi-
squared tests, pooled LS model was determined to be efficient. Later, robustness check further
proved its efficiency. Moreover, we had no grounds to assume autocorrelation in the pooled LS
model. Therefore the estimated pooled LS model was concluded to be the preferable model for the
Internet Banking diffusion analysis in the new EU member countries.
Key words: Panel data analysis, Fixed Effects, Pooled, Bass, Diffusion Models, Electronic Banking,
Adopters, Innovators.

2
Santrauka
Šis baigiamasis darbas yra skirtas tirti internetinės bankininkystės paplitimą naujosiose ES šalyse
(prisijungusiose 2004-2007 metais). Šalys buvo pasirinktos darant prielaidą, jog jų ekonominis
vystymasis turėtų būti gan panašus. Taip pat, šios šalys turėjo gana gerai subalansuotus duomenis,
o tai būtina sąlyga statistiškai patikimai analizei.
Apibendrintas Bass‘o skvarbos modelis buvo naudojama mūsų paneliniuose modeliuose. Dėl Bass‘o
modelio formuluotės, duomenims teko naudoti pirmuosius skirtumus. Todėl natūralu, jog pirmųjų
skirtumų paneliniai modeliai nebuvo analizuojami. Priedo, dėl programinės įrangos ribotumo (R bei
EViews), Bass‘o skvarbos formulė negalėjo būti pritaikyta atsitiktinių poveikių modeliams. Nagrinėti
trys skirtingi panelinių duomenų analizės atvejai: pastovaus laisvojo nario, fiksuotų individualiųjų bei
fiksuotų laiko poveikių modeliai. Be to, skirtingi būdai skaičiuoti modeliams buvo bandomi kiekvienu
atveju. Pirmiausia - fiksuojant galutinį vartotojų procentą visuomenėje. Toliau – fiksuojant
inovatorių bei imituotojų dalis. Galiausiai nagrinėjome skvarbos modelius neapribojant minėtų
koeficientų.
Fiksuotus laiko poveikius teko atmesti jau pradinėje analizės etape ir tolimesnių testų neprireikė.
Tolimesnei analizei abu: pastovaus laisvojo nario bei fiksuotų individualiųjų poveikių modeliai buvo
parinkti su fiksuotu galutiniu vartotojų procentu visuomenėje (95%). Po F bei Chi-kvadrato testų
pastovaus laisvojo nario modelis buvo parinktas kaip efektyvus modelis. Vėliau pastovumo testas
papildomai patvirtino minėto modelio efektyvumą. Be to, mes neturėjome pagrindo įtarti
autokoreliacijos pastovaus laisvojo nario modelyje. Taigi, buvo nustatyta, jog pastovaus laisvojo
nario modelis yra tinkamiausias internetinės bankininkystės skvarbos panelinei analizei naujosios ES
šalyse.
Raktiniai žodžiai: panelinė analizė, fiksuoti efektai, pastovus laisvasis narys, Bass, skvarbos modeliai,
elektroninė bankininkystė, imituotojai, inovatoriai.

3
Table of contents
1. Introduction
2. Dataset
3. Diffusion of Innovation and Bass Diffusion Process
4. Panel Data Models
4.1. Pooled Model
4.1.1. Pooled LS with Restricted Percentage of Eventual Adopters
4.1.2. Pooled LS with restrictions on the rates of innovators and
imitators
4.1.3. Unrestricted Pooled LS
4.2. Fixed Individual Effects
4.2.1. Fixed Individual Effects with Restricted Percentage of
Eventual Adopters
4.2.2. Fixed Individual Effects with restrictions on the rates of
innovators and imitators
4.2.3. Unrestricted Fixed Individual Effects
4.3. Fixed Time-Period Effects
4.3.1. Fixed Time-Period Effects with Restricted Percentage of
Eventual Adopters
4.3.2. Fixed Time-Period Effects with restrictions on the rates of
innovators and imitators
4.3.3. Unrestricted Fixed Time-Period Effects
5. Tests
5.1. Redundant Fixed Effects cross-section test
5.2. Robustness and stability test of the Pooled LS model
5.3. Durbin-Watson test for Autocorrelation
6. Conclusions
7. References
8. Appendix - Dependent and Independent Variables

4
1. Introduction
Electronic banking has benefits to both customer’s and bank’s sides. From the viewpoint of
customers, electronic banking is usually cheaper, time-efficient and can be even easier than ordinary
banking [1]. When it comes to banks, the main advantage is lower maintenance costs. Also,
electronic banking is advantageous from the viewpoint of environment – internet banking provides
the possibility to reduce the usage of paper, energy and other resources as most daily operations
can be done online. Besides, internet banking, not only providing a possibility to save money and
environment, helps improve the customer service. [2]
The European Union has experienced a substantial increase of its members in the years 2004 – 2007.
The new countries were as follows:
1. Lithuania (1 May 2004)
2. Latvia (1 May 2004)
3. Estonia (1 May 2004)
4. Hungary (1 May 2004)
5. Cyprus (1 May 2004)
6. Malta (1 May 2004)
7. Poland (1 May 2004)
8. Czech Republic (1 May 2004)
9. Slovakia (1 May 2004)
10. Slovenia (1 May 2004)
11. Bulgaria (1 January 2007)
12. Romania (1 January 2007)
All of the stated members (except for Malta and Cyprus) had been part of the Soviet Union or the
Eastern bloc which had greatly hampered their development. However, after the accession to the
EU, a rapid economic development took place in these countries, which was followed by internet-
based service expansion. A lot of services nowadays can be provided via the internet and online
banking is no exception. During the recent years it has gained a significant amount of popularity in
most of the selected countries. However, it can be easily seen that the numbers greatly vary across
countries. The following figure 1.1 depicts the percentage of individuals, aged from 16 to 74, using
Internet Banking services in 2005 and 2014.

5
Fig 1.1.: Internet banking users in 2005 and 2014
In 2005 we can see an outlier (Estonia) and some other countries that are ahead of their pack –
specifically, Latvia and Malta. Logically, those countries that started from the top, had a higher
chances of staying at the top in 2014. However, some differences are quite miniscule in 2005 (i.e.
Lithuania and Hungary) but the gap widened significantly during the forthcoming 10 years.
Furthermore, some other states managed to overpass the former front-runners (i.e. Slovakia outrun
Slovenia)
Fig. 1.2: Internet Banking Growth

6
From the figure 1.2, we can now see that for most of the time, Internet Banking kept growing;
however, the growth rates greatly vary in the selected states. Moreover, Internet Banking kept
growing quite steadily (even if differently) in most of the sample countries. The online banking
popularity growth seems to have taken an S-curve shape. Except maybe for Bulgaria and Romania,
which have delayed growth of online banking,
It does not necessarily mean that Romania and Bulgaria are bound to lag behind forever. They were
the latest to join the EU among our sample countries. A gradual change in the growth rate of the
two lagging countries would be expected in the future.
Fig. 1.3: Internet Banking Spread in the Sample Countries
Not surprisingly, because of these heterogeneous development rates, the countries reached totally
different rates of development and it is obvious that economic convergence due to the EU
membership, alone, cannot explain these varying rates of development. If anything, the gaps of
Internet banking popularity even widened in most of the cases. Therefore, culture and some
country-specific factors are likely to have had a substantial impact towards these developments.
The objective of this thesis is to investigate what kind of models are the most appropriate for
explaining these developments and what factors play a major role in this technology diffusion
process.

7
2. Dataset
The sample period of the data is 2005 – 2014 at annual frequency. The main reasons for this choice
are the fact that there were a substantial amount of certain explanatory data missing in 2004 and
2015 years and the fact that most countries joined the EU in 2004, two others in 2007, which is also
relatively not far from 2005. This way, the models are more useful for explaining the developments
internet banking of countries for the recent past – after their accession to the EU.
The data sample limitations explain why among the latest EU members Croatia was not included.
There is a substantial amount of gaps in Croatia’s dataset (especially until 2007) making the dataset
unbalanced and diminishing the reliability of models, especially the Pooled least squares (later -
pooled LS). In turn, it would be harder to determine which of the panel data technique is most
suitable for this analysis.
Romania and Bulgaria’s datasets have a few missing observations, which were interpolated to avoid
dealing with unbalanced panel dataset. The variables were IntBank, HousInt, IntD and IntW (see –
Appendix). These variables were interpolated for 2005 year – the only unobserved year in the
selected time period of the dataset. It should be stated that the observations of these variables for
the year 2004 were available and were included in the formula. The selected method for
interpolation was moving averages, specifically:
Xt = (3 ∙ (Xt−1 + Xt+1) + Xt+2)/7
The idea was that in the beginning, development process is usually very fast; however, with time it
slows down. Therefore, this formula seems to satisfactory summarize an augmenting process,
where
∆xt ≥ ∆xt+1
Furthermore, IntBank was extrapolated for Malta for 2004 in order to acquire the first differences
of the IntBank variable. The earliest known value was 2005; therefore for an augmenting time series,
the selected formula for extrapolation into the past was chosen as follows:
Xt = (2 ∙ (Xt+1 + Xt+2) + Xt+3)/6
As we have an augmenting series, the final sum is divided by 6 in order to get a value that is a little
lower than the ones in the succeeding years.
When looking for other possibly useful variables, a few interesting conclusions from one previous
study [3] are worth keeping in mind:
 Educational level raises the coefficient of innovation, whereas ageing of individuals does
the opposite
 Educational level and ageing of individuals with primary education lowers the coefficient of
imitation.
 Coefficient of innovation is diminished by requirement of special financial knowledge

8
Therefore, it seems only reasonable to test Education-related variables and money spent on
research and development (usually abbreviated as R&D).
3. Diffusion of Innovation and Bass Diffusion Process
When a new technology is introduced to the market, it takes a certain amount of time for it to gain
in popularity. There are a lot of factors determining how quickly and to what extent this technology
takes off: how well the good is being communicated through different channels, the pertained
usefulness of it, how much time has passed and etc. In the beginning it is very important how many
people are willing to try the new technology. These people are called innovators. They help
popularize the product and then are followed by other customers – adopters. Everett M. Rogers [4]
proposed five stages of technology adoption in a society in this order:
1. Innovators (2.5%),
2. Early adopters (13.5%)
3. Early majority (34%)
4. Late majority (34%)
5. Laggards (16%)
It could be stated that Internet banking in 2005 was just starting to gain popularity in most of the
sample countries and typically would be in the first or the second stage of the diffusion of innovation
process. During the following 10 years we saw the continuing adoption of this technology. From the
figure 1.1 it can be seen that in 2014 the three Baltic States had already managed to reach the fourth
stage of the diffusion of innovations process, whereas two other, most lagging sample countries:
Bulgaria and Romania had just reached the second phase of early adoption. The other states were
in between - the third stage. Thus, we can see that even at cross-country level there are leaders,
core countries and laggards.
Despite the fact that internet banking provides more or less the same benefits for every country, it
is clear that it has taken different rates of development in each of them. Hence, the Generalised
Bass model [5] could be a good starting point for the Internet Banking modelling. When compared
to the original Bass model, the generalized formula incorporates explanatory variables which might
be helpful to explain the diffusion process of Internet banking popularity and account for lagging
and leading cross-country differences. For example, Internet banking may also be greatly influenced
not only by the number of people eager to try new technologies but also by the number of
households actually having access to the Internet. Besides, such data as education or rate of young
people versus total population are other factors worth consideration. Younger people are typically
more eager to try new technologies, while more educated citizens are usually more tech savvy.
These are just a couple of examples how such and similar data might help explain the rates of
adopters and imitators in a society of interest.
The Bass model assumes a linear relationship between the individuals who have already adopted
the new technology and the ones that are yet to do the same. According to the model, the number

9
of new Internet Banking would be greatly affected by the existing rates of innovators and imitators
in any given society. While the former are not influenced by the number of already existing internet
banking users, the latter – imitators - are. Therefore, innovators would play a more important role
in the beginning, foreseeing the new possibilities, when Internet banking had not reached high
numbers of customers and is just starting to gain popularity. They could be companies, government
institutions, bigger clients- the ones that in some ways influence other people and/or have the
biggest advantage of the new technology. Naturally, the percentages of innovators and adopters
vary across countries, thus contributing to different rates of technology adoption.
The original formula of the Generalised Bass model could be rewritten for a percentage change of
internet banking users as follows:
𝑓𝑡 = (𝑝 + 𝑞𝐹𝑡−1)(𝑀 − 𝐹𝑡−1) ∙ 𝐺 𝑋𝑡
, where
𝑓t - Change of customers in a period t
𝑝 – Rate of innovators
𝑞 – Rate of imitators
𝐹𝑡−1 – Cumulative percent of adoptions having occurred before period t
𝑀 – Percent of eventual adopters
𝐺 𝑋𝑡
– Chosen explanatory data function
The number of eventual adopters could make an interesting study itself. In this case M stands for a
percentage of individuals in a society who will eventually adopt internet banking services. It would
be natural to assume the peak of M has not been reached anywhere in the world yet: even in
countries, where online banking is very popular, the number of users is still continuing to grow. Let
us consider Norway for example – the leader of all European countries in the number of Internet
Banking users – in 2014 89% of its population aged 16 to 74 had used online banking within the last
three months. Nevertheless, in 2015 the number of users increased even further to 90%. It poses
an interesting question – what is the highest achievable rate of population using these services?
Noticing the slowdown of new users’ growth in already developed countries, we could presume the
percent of eventual adaptors being around 95%. Or we could try predicting this rate using the Bass
model, which, however, would make the model a little more complicated and estimates unstable.
Two other numbers of interest are p and q. Usually these coefficients are expected to satisfy the
following inequality:
𝑞 > 𝑝 > 0
However, it should be noted that special cases when 𝑞 = 0 or 𝑝 = 0 are also possible. Then we
would study an exponential and logistic cases, correspondingly.

10
4.Panel data models
Four different panel data estimation techniques will be studied: pooled LS, fixed individual effects
(FIE) and fixed time-period effects (FTPE) models.
4.1. Pooled LS
It is a restrictive model which specifies constant coefficients. If the model is correctly specified and
errors are not correlated with regressors, it the model can be consistently estimated using Pooled
LS. However, if fixed or random effects models are appropriate, Pooled LS estimator is inconsistent.
[6]
The model in its error term: 𝑦𝑖𝑡 = 𝛼 + 𝑥𝑖𝑡
′
𝛽 + 𝜀𝑖𝑡 (1)
After the transformation to the Bass model, we have:
𝑦𝑖𝑡 = 𝛼 + (𝑝 + 𝑞 ∑ 𝑦𝑖𝑗
𝑡−1
𝑗=1 )(𝑀 − ∑ 𝑦𝑖𝑗
𝑡−1
𝑗=1 ) ∙ (1 + ∑ 𝜷𝑖
𝑛
𝑖=1 𝑿𝒊𝒕) + 𝜀𝑖𝑡 (2)
The dataset comprised data from 12 countries. All of these countries joined the EU in the time period
of 2004-2007 and the majority of them had undergone the Soviet Union oppression, which might
lead to a viable hypothesis that their online banking development factors were to be similar
additionally keeping in mind the theory of economic convergence and the nearness to the more
advanced Western European countries.
Practical application
4.1.1. Pooled LS with Restricted Percentage of Eventual Adopters
The plm package of R does not allow for restrictions on coefficients; for this reason, R was not
particularly suitable for Bass Diffusion panel data modelling. The problem arises as one cannot put
restrictions on the parameters p, q, M and simply has to presume that the coefficients would be in
accordance with the Generalised Bass diffusion formula. Unfortunately, such a naive assumption
proved itself wrong and thus EViews was chosen for modelling.
First of all, to simplify the model, we equalise M to 95% - a presumed maximum percentage of
Internet Banking users in any of the selected societies. Then the model is selected using Forward
selection process.

11
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽1 ∙ 𝐷𝐼𝑛𝑡𝐷𝑡 + 𝛽2
∙ 𝐷𝑃𝑜𝑝𝑌𝐹𝑡)
Coefficient Estimate Standard Error Pr(>|t|)
𝑝 0.0031 0.0018 0.0727
𝑞 0.0008 0.0002 0.0012
𝛽1 0.3521 0.1354 0.0105
𝛽2 0.9468 0.4555 0.0399
Table 3.1: Coefficients of Pooled LS with Restricted Percentage of Eventual Adopters
Adjusted 𝑅2
0.4222
AIC 4.2594
BIC 4.3539
Table 3.2 - Statistics
It can be seen that the significant explanatory variables were dintd and popYf (description of data
and other variables in the Appendix 1). Both dintd and popYf have positive coefficients which seems
very intuitive. Yet, one surprising fact stands out – while p is around 0.03 and perfectly fits its
theoretical estimates in technology diffusion processes, q value is not only a lot lower than its usual
average value (0.38) but also lower than p. This indicates that there are more innovators than
adopters.
It should be noted that the intercept was removed due to its insignificance (p-value ≈ 0.64).
However, if one expands the (2) formula, he may notice that p in a way may serve as an intercept in
Pooled LS models.
4.1.2. Pooled LS with restrictions on the rates of innovators and imitators
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + (0.03 + (0.38 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)
𝛼 -191.2933 19.8751 0.0000
𝑀 58.5857 1.8497 0.0000
Table 3.3: Coefficients of Pooled LS with restrictions on the rates of innovators and imitators
Adjusted R2
-2592.63
AIC 12.6423
BIC 12.6888
Table 3.4: Statistics
This pooled model was estimated fixing p and q coefficients but releasing M. In previous studies the
average values of p and q were found to be 0.03 and 0.38, respectively [7].
A few problems can be easily noticed which make the model invalid. First of all, the R2
is not only
negative but also its modulus greater abnormally large: > 630. This indicates that the model does

12
not fit the data. Moreover, we may see that the c(2) coefficient is far from any possible real M-value.
For example, in 2014 Internet banking in a lot of countries has already surpassed this value. Last but
not least, Akaike of this Pooled LS model is bigger than that of the last model, which helps us once
again to rule out the idea of fixing the p and q but releasing the M coefficient.
One possible explanation of these problems could be that the actual rate of imitators is far from the
estimated average value of q. Consequently, the first Pooled LS model would preferable as it allow
for estimations of actual p and q. Moreover, with rejection of this model, there is no need for
further estimation of a Pooled LS with restricted p, q and M values, as it would be simply a subset
of this rejected idea of restricting p and q values.
4.1.3. Unrestricted Pooled LS
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽1 ∙ 𝐷𝐼𝑛𝑡𝐷𝑡 + 𝛽2
𝑝 -0.0795 0.0250 0.0019
𝑞 0.0009 0.0003 0.0055
𝑀 -3.0943 2.5722 0.2315
𝛽1 0.3579 0.1397 0.0117
𝛽2 0.8844 0.4677 0.0612
Table 3.5: Coefficients of Unrestricted Pooled LS
Adjusted R2
0.419846
AIC 4.2717
BIC 4.3897
The unrestricted Pooled LS model lets us estimate q, p and M without relying on guesses. Ideally,
this would be a perfect approach when dealing with uncertainties. However, when estimating the
model without restrictions, again, a few serious problems arise.
First of all, the M coefficient becomes insignificant (p − value ≈ 0.23). Moreover, the estimated
percentage of eventual adopters is -3.094 which is incorrect for obvious reasons.
Secondly, while the rate of imitators marginally increases when compared to the Pooled LS with
Restricted Percentage of Eventual Adopters (0.000905 and 0.000779, respectively), the rate of
innovators becomes negative.
Finally, when comparing for the Akaike criterions, we can safely reject the Unrestricted Pooled LS
model.
The intercept was again removed due to insignificance (p-value ≈ 0.95)

13
Conclusions
Among the Pooled LS models, the first one - Pooled LS with Restricted Percentage of Eventual
Adopters is preferable
4.2. Fixed Individual Effects
Even if one ignores the flaws of the economic convergence theory, it is statistically evident that the
sample countries were developing at contrasting paces, had different standards of living, different
rates of GDP growth, levels of tertiary education were also greatly varying and so on. Therefore
there might exist some behavioural or/and country specific factors in each country leading to
different rates of Internet banking popularity augmentation.
The original FIE model in its error term looks as follows:
yit = αi + xit
′
β + εit (3), where:
xit
′
β = xit,1β1+. . . +xit,k βk and xit,j indicates variable j at time t for i − th individual
αi - unobserved individual effect
Again, we transform the (3) equation to a Bass Diffusion model form:
yit = αi + (p + q ∑ yij
t−1
j=1 )(M − ∑ yij
t−1
j=1 ) ∙ (1 + ∑ 𝛃i
n
i=1 𝐗 𝐢𝐭) + εit (4)
It should be stated that αi in the FIE model can be correlated with xit
′
. Nevertheless, the assumption of
strict exogeneity ( E(εit|xi, αi) = 0, t = 1,2, … T ) has to be maintained.
The assumptions and further details about FIE can be found in [8].
4.2.1. Fixed Individual Effects model with Restricted Percentage of Eventual
Adopters
It can be seen that the only significant regressor was dintd. As expected, dintd had a positive
coefficient which means that the increase in daily internet users was positively correlated with the
use of online banking.
Similarly to the Pooled LS model, p outweighs q, with the estimated rates of innovators and
imitators being 0.081366 and 0.001970, respectively.

14
One could note that the adjusted R2
is higher in the the Fixed Individual Effects model, whereas,
AIC and BIC criterion values are lower in the Pooled LS model.
DIntBankt = α + αi + (p + (q ∙ LagIntBank 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽 ∙ 𝐷𝐼𝑛𝑡𝐷𝑡)
𝛼 -7.2851 3.5247 0.0413
𝑝 0.0814 0.0361 0.0264
𝑞 0.0020 0.0008 0.0134
𝛽 0.0472 0.0232 0. 0450
Table 3.7: Coefficients of fixed individual effects model with restricted percentage of eventual
adopters
Adjusted 𝑅2
0.4401
AIC 4.3135
BIC 4.6676
Date Effect
Bulgaria -1.7366
Czech Republic 0.5577
Estonia 2.8153
Cyprus -1.0119
Latvia 1.3855
Lithuania 1.3585
Hungary -0.6404
Malta -0.1820
Poland -0.2444
Romania -1.4144
Slovenia -0.9188
Slovakia -0.2695
Table 3.9: Individual Fixed Effects
From the 3.9 table we can see that the value of individual effects for a country is positively correlated
with its level of attained diffusion. Front-runners have higher values and vice versa. It is possible
that there exist unobserved factors which affect the rate of advancement of online banking in our
sample countries. Fixed Effects help to account for these factors.
4.2.2. Fixed Individual Effects with restrictions on the rates of innovators and
imitators
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼𝑖 + (0.03 + (0.38 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)

15
𝛼̅ -174.6002 20.9015 0.0000
𝑀 56.6192 2.2514 0.0000
Table 3.10: Coefficients of Fixed Individual Effects with restrictions on the rates of innovators and
imitators
Adjusted 𝑅2
-1250.25
AIC 11.9989
BIC 12.3008
Date Effect
Bulgaria 126.2121
Czech Republic -42.4761
Estonia 236.8397
Cyprus -30.1439
Latvia -20.4276
Lithuania -58.6136
Hungary -37.8218
Malta -81.3121
Poland -55.3329
Romania 129.0612
Slovenia -91.8014
Slovakia -74.1836
The idea of restricting the rates of innovators and imitators while releasing the number of eventual
adopters has to be rejected once more. The 𝑅2
(both the regular and adjusted) value remain
negative, fluctuating around -2000, which means that such model cannot be estimated, moreover,
after noticing the high values of Akaike and BIC, we render this model invalid.
4.2.3. Unrestricted Fixed Individual Effects
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼𝑖 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽 ∙ D𝐷𝑒𝑛𝑠)
𝛼 2.4158 1.6872 0.1552
𝑝 -0.1091 0.0537 0.0445
𝑞 0.0027 0.0009 0.0031
𝑀 -1.1100 15.7600 0.9440
𝛽 0.1307 0.1761 0.4597
Table 3.13: Unrestricted Fixed Individual Effects

16
Adjusted 𝑅2
0.2737
AIC 4.5687
BIC 4.9404
Date Effect
Bulgaria -2.3501
Czech Republic 0.8047
Estonia 4.4563
Cyprus -1.5329
Latvia 2.1149
Lithuania 1.7648
Hungary -0.5452
Malta -0.6019
Poland -0.4692
Romania -2.3037
Slovenia -1.1941
Slovakia -0.1436
Due to collinearity problems, which arise because of a complicated formula, the stated model was
the only estimative model. However, once again, we declare that the unrestricted model (this time
FIE) is useless because of illogical estimated coefficients and insignificant 𝑀’s p-value, thus, further
analysis is not required.
Conclusions
Of all Fixed Individual Effects models, only the Fixed Individual Effects model with restricted
percentage of eventual adopters was appropriate. It was selected for further analysis
4.3. Fixed Time-Period Effects
As we already established, it is possible to control for individual effects. However, it is also likely
that, because of unobserved causes, countries in a given year 𝑡 ∈ 𝑇 perform better than in another
year 𝑔 ∈ 𝑇, 𝑔 ≠ 𝑡. Therefore, with time-period effects may also be useful in explaining the online
banking developments.
While the assumptions of Fixed time-period effect (FTE) model remain the same, it is denoted
slightly differently from the FIE model. The error term looks as the following:

17
𝑦𝑖𝑡 = 𝛼 𝑡 + 𝑥𝑖𝑡
′
𝛽 + 𝜀𝑖𝑡 (5), where:
𝑥𝑖𝑡
′
𝛽 = 𝑥𝑖𝑡,1 𝛽1+. . . +𝑥𝑖𝑡,𝑘 𝛽 𝑘 and 𝑥𝑖𝑡,𝑗 indicates variable 𝑗 at time 𝑡 for 𝑖 − 𝑡ℎ individual
𝛼 𝑡 - unobserved time-period effect
Transformation of the (5) equation to a Bass Diffusion model form gives us:
𝑦𝑖𝑡 = 𝛼 𝑡 + (𝑝 + 𝑞 ∑ 𝑦𝑖𝑗
𝑡−1
𝑗=1 )(𝑀 − ∑ 𝑦𝑖𝑗
𝑡−1
𝑗=1 ) ∙ (1 + ∑ 𝜷𝑖
𝑛
𝑖=1 𝑿𝒊𝒕) + 𝜀𝑖𝑡 (6)
4.3.1. Fixed Time-Period Effects with Restricted Percentage of Eventual Adopters
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼 𝑡 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽 ∙ 𝐷𝐷𝑒𝑛𝑠𝑡)
𝛼 -1.9332 2.6126 0.4609
𝑝 0.0236 0.0259 0.3643
𝑞 0.0023 0.0006 0.0003
Table 3.16: Coefficients of Fixed Time-Period Effects with Restricted Percentage of Eventual
Adopters
Adjusted 𝑅2
0.2146
AIC 4.6180
BIC 4.8968
Date Effect
2005 0.7059
2006 0.7403
2007 1.4834
2008 0.6545
2009 0.5500
2010 0.6673
2011 -0.4595
2012 -2.3186
2013 -0.9220
2014 -1.1013
Table 3.18: Period Fixed Effects
We can see that the Fixed Time-Period Effect is insignificant (p-value ≈ 0.46), thus searching for
possible regressors is meaningless and we reject this model.

18
4.3.2. Fixed Time-Period Effects with restrictions on the rates of innovators and
imitators
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼 𝑡 + (0.03 + (0.38 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)
𝛼 -207.0243 22.2651 0.0000
𝑀 60.4389 2.18179 0.0000
Table 3.19: Coefficients of Fixed Time-Period Effects with restrictions on the rates of innovators
and imitators
Adjusted 𝑅2
-2666.85
AIC 12.74116
BIC 12.99668
Date Effect
2005 67.4281
2006 46.6295
2007 19.5190
2008 -12.6655
2009 -25.3048
2010 -27.2868
2011 -26.4477
2012 -18.6216
2013 -22.9811
2014 -0.26917
The model is again non-estimative (Negative 𝑅2
).
4.3.3. Unrestricted Fixed Time-Period Effects
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼 𝑡 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (𝑀 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙(1 + 𝛽 ∙ 𝐷𝐼𝑛𝑡𝐷)
𝛼 -0.2628 1.6024 0.8700
𝑝 0.0030 0.0092 0.7496
𝑞 0.0010 0.0006 0.1039
𝑀 89.604 10.587 0.0000
𝛽 0.2952 0.2899 0.3110
Table 3.22: Coefficients of Unrestricted Fixed Time-Period Effects

19
Adjusted 𝑅2
0.4310
AIC 4.3223
BIC 4.6528
Date Effect
2005 0.5534
2006 0.6477
2007 1.0362
2008 0.5125
2009 -0.4322
2010 0.6539
2011 0.1125
2012 -1.3093
2013 -0.8778
2014 -0.7507
The Fixed Time-Period Effects are insignificant (p-value ≈ 0.87). Besides, multicollinearity problems
do not let us remove the regressor.
Conclusions
Neither of the Fixed Time-Period Effects models was satisfactory; therefore, now we will only
concentrate on the Pooled LS and Fixed Individual Effects models.
5. Tests
The next step in the study will be testing which model is the most appropriate for internet banking
analysis in the chosen countries. Now we have two models:
a) Pooled LS with Restricted Percentage of Eventual Adopters:
b) Fixed Individual Effects model with Restricted Percentage of Eventual Adopters:
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + 𝛼𝑖 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽
∙ 𝐷𝐼𝑛𝑡𝐷𝑡)

20
5.1. Redundant Fixed Effects cross-section test
Cross-section Chi-squared and F tests
Due to EViews limitations, we cannot directly test the chosen form of our Pooled LS model against
the selected Fixed Effects model. However, we can test for redundant fixed effects in the latter
model. Firstly, we take away the 𝐷𝑃𝑜𝑝𝑌𝐹𝑡 regressor and thus reduce our LS model to such form:
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 0 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽 ∙ 𝐷𝐼𝑛𝑡𝐷𝑡)
Now the pooled LS estimates look like this:
𝑝 0.0034 0.0017 0.0423
𝑞 0.0005 0.0002 0.0023
𝛽 0.4875 0.2058 0.0195
Table 5.1: Chosen pooled LS
Adjusted 𝑅2
0.4115
AIC 4.2694
BIC 4.3403
The Coefficients in the simplified model remain significant. Even though the adjusted 𝑅2
decreases
and AIC and BIC criterions decrease, now we can compare the adjusted LS and Fixed Effects modes.
If our tests show that fixed effects are redundant, this will serve as a good indicator that the initial
Pooled LS should be preferred.
The fixed individual effects model with Restricted Percentage of Eventual Adopters is tested against
its Pooled LS alternative:
𝐷𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡 = 𝛼 + (𝑝 + (𝑞 ∙ 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡)) ∙ (95 − 𝐿𝑎𝑔𝐼𝑛𝑡𝐵𝑎𝑛𝑘 𝑡) ∙ (1 + 𝛽 ∙ 𝐷𝐼𝑛𝑡𝐷𝑡)
Chi-squared statistics Probability
18.4254 0.0722
Table 5.3: Chi-squared test
The p-value of the Chi-squared test is above the 0.05 significance level indicating that the fixed
effects are likely to be redundant but the result does not yet seem to be decisive.
Now let us test the same hypothesis with the F-test
F-statistic Probability
1.5816 0.1153
Table 5.4: F-test

21
The p-value of the F-test once again tells us that the fixed individual effects are rather likely to be
insignificant. Furthermore, the p-value this time is higher than the more forgiving level of 0.1,
reassuring us again that the Pooled LS is preferable.
Now we can finally come back to the initial pooled LS model with Restricted Percentage of Eventual
Adopters. This model has an additional significant regressor - a change in ratio of younger
population against the entire society. Furthermore, it has a higher value of adjusted 𝑅2
and lower
values of AIC and BIC than the simplified LS version without this independent variable. As a result,
we opt for the initial Pooled LS model:
5.2. Robustness and stability test of the Pooled LS model
One additional way of deciding on the selected Pooled LS model could be drawing different
subsamples out of our dataset and then re-estimating our model. In this case we create 12 different
LS models with the same specifications as our main Pooled LS model. However, each of them
includes data only from 11 of the 12 sample countries. Following are our new estimated coefficients.
Coefficient
Exlcuded
country
Original
model
Without
Bulgaria
Without
Cyprus
Without
Czech
Republic
Without
Estonia
Without
Hungary
Without
Latvia
𝑝 0.0032 0.0043 0.0034 0.0022 0.0028 0.0031 0.0028
𝑞 0.0008 0.0007 0.0009 0.0009 0.0007 0.0007 0.0007
𝛽1 0.3521 0.3621 0.3104 0.3236 0.3989 0.3816 0.3863
𝛽2 0.9468 0.8016 1.0981 1.4781 1.0213 0.9675 0.9456
Coefficient model Without
Lithuania
Without
Malta
Without
Poland
Without
Romania
Without
Slovakia
Without
Slovenia
𝑝 0.0029 0.0034 0.0028 0.0038 0.0035 0.0035
𝑞 0.0007 0.0008 0.0008 0.0007 0.0009 0.0008
𝛽1 0.3998 0.3280 0.3730 0.3527 0.2929 0.3163
𝛽2 0.8691 0.6997 0.8671 0.8764 0.9380 0.7924
Table 5.5: Adjusted Sample Models
When deciding between the Pooled LS and FIE models, one may notice that 𝑝 coefficient may serve
as an intercept in the Pooled model. Therefore, its variance is most significant as other coefficients
would be far less affected by selection of the estimator. The original estimated value is 0.003168

22
and while it is quite different from the model with Bulgaria and Czech Republic excluded (0.004318
and 0.002235, respectively), it does not vary very much in general. Differences in 𝑝 and 𝑞 could be
explained by the varying number of innovators and adopters across countries. For example a higher
than the average rate of innovators would indicate that the rate of innovators is lower in the
excluded state and vice versa. Interestingly, Lithuania has a higher rate of both innovators and
adopters than the average of the remaining countries.
The only coefficient that varies quite noticeably is 𝛽2. It looks that in Czech Republic the change of
ratio of Younger people to the remaining population impacts the developments of online banking a
lot less than in Malta for example. Other models’ 𝛽2 , while still varying, are more in line with the
original estimated value. However, the variance of q, β1 and β2 are not very important when
deciding between the Pooled LS and FIE models as they do not serve as intercepts and thus are little
affected by that choice.
Given the estimated values, it could be said that the Pooled LS model is satisfactory. The rate of
innovators is generally quite stable. Besides, even when it does vary, the estimated values are very
low and the derived intercept: p ∙ 95 would not affect our predictions very significantly. We have
no grounds to oppose the results of the F and Chi-squared tests – that Pooled LS gives us efficient
estimates.
5.3. Durbin-Watson test for Autocorrelation
The statistics of Durbin-Watson are as follows:
Model Durbin-Watson statistic
Pooled LS with Restricted Percentage of Eventual
Adopters
2.5449
Fixed Individual Effects model with Restricted
Percentage of Eventual Adopters
2.7684
Table 5.6 Durbin-Watson Statistics
While the Durbin-Watson statistics of the FIE model are out of range (1.4 – 2.6), we do not have
enough evidence to suspect serial correlation in the Pooled LS model.
6. Conclusions
The goal of the study was to find the best model to explain online banking developments in the new
EU member states. 12 countries were selected on the basis of their accession date to the European
Union. That could have been one of the reasons to suspect that a pooled estimation method would
prove satisfactory. On the other hand, as it quite often occurs in panel data analysis, fixed effects –

23
let it be time or individual – are more preferable to the pooled models. This way they allow for
country or time specific intercepts.
However, in this analysis we find that the Pooled LS model gives us efficient estimates. It is
interesting to note that the estimated rate of innovators was found to be very close to the average
value of 0.03, while the estimated rate of adopters was significantly far from the average number
found in literature (0.38). Nevertheless, this does not contradict the Generalised Bass Model theory
in any way.
F and Chi-squared tests were not the only ones to show that the Pooled LS model is preferable.
Robustness check - excluding a country then re-estimating the model for the remaining 11 states -
showed us that the variability in p (rate of innovators, partially serving as an intercept in the Pooled
LS model as the usual intercept in the model was insignificant) is rather low and the value is quite
close to zero. Therefore it was another confirmation for the Pooled LS model choice.
Efficiency of the Pooled LS model was likely achieved due to differenced regressors’ data. Despite
similar (or in most of the times – exactly the same) accession dates to the EU, the member states
had very different numbers on their datasets. Differencing the data helped us concentrate only on
the recent changes and not the aggregated levels of development.
Different independent variables were considered (see Appendix) but only two of them were
significant in our Pooled LS model. Not surprisingly, Internet banking and Internet access were
closely related and thus a change of frequency of Internet access daily was included in the model as
a significant regressor. Interestingly, the change of ratio of population from 15 to 64 years against
the entire population was not only significant but also had a higher weight than DIntD in explaining
the developments of online banking in the countries. Its coefficient was almost three times bigger
than that of the previous one (0.9468 and 0.3521, respectively). This tells us that the structure of a
society is crucially important when it comes to internet banking popularity. Younger people tend to
be more tech savvy, more of them enter the workforce and thus need internet banking services.
Therefore, banks should focus on the structure of the society when making decisions about
investment in their online banking platforms.
It is worth noting that in this analysis our time series was not very long (10 observations). Longer
time series data might have given us different estimates. However, a lot of the sample countries
were missing observations prior to 2005 or 2004 and the aim of the thesis was to study the recent
developments in the specified field.
Also, there were limitations in the model implementation. The plm package in R would not allow
restricting the parameters, therefore R gave us coefficients which were not in line with the Bass
diffusion theory. EViews, on the other hand, does not have heteroscedasticity tests for panel data
models, moreover, there was only Durbin-Watson statistics for autocorrelation detection. If such
limitations were to be overcome, further and more in depth analysis would be possible.

24
Išvados
Šis baigiamojo darbo tikslas buvo ištirti ir paaiškinti internetinės bankininkystės plitimą naujosiose
ES šalyse. 12 Šalių buvo parinktos pagal jų įstojimo į ES datas. Panaši įstojimo data galėjo tapti viena
priežasčių įtarti, kad pastovaus laisvojo nario panelinis modelis bus efektyvus. Visgi, verta paminėti,
jog fiksuoti efektai – individualieji ar laiko - gan dažnai būna efektyvesni už fiksuotojo laisvo nario
modelius. Jie leidžia įvertinti skirtingus laisvuosius narius kiekvienai šaliai ar laikotarpiui.
Atlikdami šią analizę išsiaiškinome, jog fiksuotojo laisvo nario modelis mums duoda efektyvius
įverčius. Taip pat, reikia paminėti, jog įvertinta inovatorių dalis visuomenėje buvo labai arti vidutinės
reikšmės randamos literatūroje (0.03), tačiau įvertinta imituotojų dalis gerokai nutolo nuo tipinės
reikšmės randamos literatūroje (0.38). Visgi, nutolusios nuo įvertintų vidurkių reikšmės jokiu būdu
nepaneigia Bass‘o modelio tinkamumo.
F ir Chi-squared testai nebuvo vieninteliai parodantys, jog fiksuotojo laisvo nario modelis yra
tinkamiausias. pastovumo testas – sukūrimas 12 papildomų pastovaus laisvojo nario modelių su ta
pačia formule bet viena iš imties šalių praleista – parodė, jog p reikšmė (iš dalies tarnaujanti kaip
laisvasis narys minėtame modelyje, nes tradicinis laisvąjam nariui priskirtas koeficientas buvo
nereikšmingas) kinta gana nežymiai ir yra gana netoli nulio. Tai buvo dar viena priežastis įsitikinti
fiksuotojo laisvojo nario modelio pasirinkimu.
Fiksuotojo laisvojo nario modelio efektyvumas buvo galimai pasiektas dėl pirmųjų skirtumų
duomenims naudojimo. Nepaisant panašių (ar daugeliu atvejų – vienodų) įstojimo į ES datų, šalys
narės turėjo labai skirtingus duomenis. Jų išsivystymo lygiai gerokai varijavo visose naudotose srityse.
Diferencijavimas leido koncentruotis tik ties pastaraisiais pokyčiais, o ne į bendrus išsivystymo lygius.
Skirtingi regresoriai buvo svarstomi. (žr. Appendix), visgi tik du iš jų pasirodė reikšmingi mūsų
fiksuotojo laisvojo nario modelyje. Internetinės bankininkystės populiarumo pokyčiai buvo susiję su
kasdienio interneto naudojimo šalyse pokyčiais. Įdomu pastebėti tai, jog šalies gyventojų amžiaus
struktūros pokyčiai modelyje buvo ne tik reikšmingi, bet dar ir svaresni už DIntD aiškinant
internetinės bankininkystės vartotojų pokyčiaus kitimą. Jų koeficientas buvo beveik triskart didesnis
už kasdienį interneto naudojimą (0.9468 and 0.3521, atitinkamai). Tai mums leidžia tarti, kad darant
sprendimus apie investicijas į internetinės bankininkystės plėtrą, bankams būtina atsižvelgti į šalies
gyventojų amžių. Jaunesni žmonės yra labiau imlūs naujovėms, jų daugiau įsilieja į darbo jėgos gretas,
todėl jiems minėtos paslaugos yra galimai svarbesnės.
Verta pastebėti, jog šioje analizėje laiko eilutė nebuvo labai ilga (10 stebinių). Naudojant ilgesnes
laiko eilutės galimai būtų buvę tikslesni bei kitokie rezultatai. Visgi nemažai šalių neturėjo ankstesnių
nei 2004 ar 2005 stebėjimų; taip pat, šio darbo tikslas buvo analizuoti pastarųjų internetinės
bankininkystės plėtros pokyčių priežastis
Verta paminėti ir programinės įrangos trūkumus darant šią analizę. Plm paketas programoje R
neleido fiksuoti reikiamų parametrų, todėl buvo gauti koeficientai neatitinkantys Bass‘o
technologijų skvarbos teorijos. EViews programoje paneliniams duomenims trūko testų tikslesnei

25
analizei. Trūko heteroskedastiškumo testo, taip pat autokoreliacijai nustatyti teko tenkintis tik
Durbin-Watson statistika. Šiuos trūkumus pašalinus būtų galima padaryti tikslesnę analizę.
7. References
[1] Dalia Kriksciuniene, Ferenc Kiss, Electronic banking in Lithuania and Hungary - a co-operative
research, 2006
[2] Mei Xue, Lorin M. Hitt, Pei-yu Chen, Determinants and Outcomes of Internet Banking Adoption,
2011
[3] Javier Alonso Alfonso Arellano, Banco Bilbao Vizcaya Argentaria, Heterogeneity and diffusion in
the digital economy: Spain’s case, Working paper 2015
[4] Everett M. Rogers, Diffusion of Innovations 1962,
[5] Frank M. Bass, Trichy V. Krishnan and Dipak C. Jain, Why the bass model fits without decision
variables 1994,
[6] Microeconometrics Methods and Applications, A. Colin Cameron, Pravin K. Trivedi, 2005
[7] F. Sultan, J.U. Farley and D.R.Lehmann, A meta-analysis of applications of diffusion models
Journal of marketing research, 1990
[8] Econometric Analysis of Cross Section and Panel Data, Jeffrey Wooldridge, 2002

26
8.Appendix
Dependent and Independent Variables
Here are the variables used in the study. Note that not all of them might have been mentioned in
the work. However, all of them were studied and tested in different models. The Letter D in front of
a variable means that the first differences were taken of original variable. For example DIntBank
and IntBank would differ only in a way that the latter variable was not differenced.
Following is the description of the dataset. It should be noted that IntBank is the variable of interest
– the dependent variable. The sample period is 2005 – 2014. All of the data is taken from Eurostat,
except of UrbPop, which was taken from United Nations (Department of Economic and Social Affairs,
Population Division (2014). World Urbanization Prospects: The 2014 Revision, custom data acquired
via website). Furthermore, the descriptions of data are also taken from Eurostat and United Nations
but the variable names were created for convenient usage.
Variable
DIntBank
Description
Change in percentage of individuals using the internet for internet banking - % of
individuals aged 16 to 74
Short Description: Within the last 3 months before the survey. Internet banking
includes electronic transactions with a bank for payment etc. or for looking up
account information.
DHousIntTot Change in percentage of households with Internet access
DUrbPop Change in annual percentage of population at mid-year residing in urban areas
DEduc1 Change in less than primary, primary and lower secondary education (levels 0-2)
from 18 to 64 year
DEduc2 Change in upper secondary and post-secondary non-tertiary education (levels 3
and 4) from 18 to 64 years
DEduc3 Change in tertiary education (levels 5-8) from 18 to 64 years
DGDP Change in gross domestic product at market prices
DPopY Change in population From 15 to 64 years
DPopYf Change in ratio of PopY to Pop
PopY population From 15 to 64 years
PopO Change in population 65 years or over

27
DPopOf Change in ratio of PopO to Pop
DDens Change in population density by NUTS 3 region [demo_r_d3dens]
DRnD Change in total intramural R&D expenditure (GERD) by sectors of performance
euro per inhabitant
DPop Change in population on 1 January by age and sex [demo_pjan]
DIntD Change in frequency of Internet access daily; percentage of individuals
DIntW Change in frequency of Internet access: once a week (including every day)
percentage of individuals
Table 7.1 Dependent and Independent Variables

Econometric analysis of Internet Banking Development in the New EU Countries

Recommended

Recommended

More Related Content

Similar to Econometric analysis of Internet Banking Development in the New EU Countries

Similar to Econometric analysis of Internet Banking Development in the New EU Countries (20)

Econometric analysis of Internet Banking Development in the New EU Countries