�ݺ�ߣ

Industrial Economics EC4417
1
Department of Economics
EC 4417 - Industrial Economics Project
Autumn Semester 2014/15
Lecturer: Bernadette Andreosso-O’Callaghan
Teaching Assistant: OlubunmiIpinnaiye
Project Title:
Relationship between Market Share, R&D expenditure and Advertising
Group Members: Names ID
Lonan O Cearbhaill 11135069
Gearoid Dowling 10080414
Brian Mullins 10116052
(To be completed by T.A.)
Group Number:_______________

2
Contents
1. Introduction:........................................................................................................................ 3
2. Data: ....................................................................................................................................3
3. Methodology:....................................................................................................................... 4
4. Results:................................................................................................................................ 4
4.1 Report on the coefficients. ............................................................................................ 4
4.2 Measures of Fit............................................................................................................. 5
4.3 Violations of the assumptions of the Basic Regression Model:........................................6
5. Interpretation of Results and Conclusion:.............................................................................. 7
6. References:.............................................................................. Error! Bookmark not defined.
7. Appendix:.................................................................................................................... 10. /10

3
1. Introduction:
The technology industry in Ireland is becoming increasingly important to contributing to
growth in the Irish Economy. Med-tech and Information/Communications technology have
been identified as key industries that are driving the Irish economic recovery (Ahearne
2014).
Forty per cent of Ireland’s GDP (€72bn) comes from its technology sector, which employs
over 105,000 people. Since 2010, more than 18,000 jobs have been announced in the
sector, making Ireland the go-to place to locate international tech headquarters. Ireland is
one of the main locations worldwide where a career in IT can be advanced. Having said that
our indigenous IT firms only account for €2bn of that €72bn but we make up 28.6% of the
employment with 30,000 of the IT jobs (Silicon republic 2013).
Key stats:
 9/10 global ICT companies maintain a presence in Ireland.
 The top 5 software companies have a significant presence in Ireland.
 The total number of ICT enterprises in Ireland is approximately 5,400; 233 of which
are foreign owned.
(Enterprise Europe Network 2012).
The aim of this project is to construct a statistical model to assess the relationship between
the dependent variable (Market share) and the two independent variables (advertising and
R&D) within the technology industry. We will examine, through our statistical estimates,
what companies and to a broader extent the government could learn from this report about
how these variables interact in the technology sector and the implications it has on their
decisions.
Using the Ordinary Least Squares method, we will assess the relationship between the
variables on the basis of the below model:
Mkt Share= 𝑏1 + 𝑏2𝑅&𝐷 + 𝑏3𝑎𝑑𝑣
This will provide an estimate of the relationship between the variables which we expect to
be positive. The reason for this is the presumption that an increase in R&D and/or
advertising expenditure would also increase the % of Mkt Share a company holds in an
industry. Volker Grossman carried out an empirical driven model to study the implications of
R&D expenditure and advertising expenditure finding there to be a positive relationship
between those two variables and the size of firms and we will use these findings as the basis
for our models expected outcome (Grossman 2008).
2. Data:
We are using data from the FAME databases to guide our model on the technology
industry. We have chosen to use the % of market share held within each company when
taking the total number of observations (30) as the entire industry. We have calculated
this through taking a company’s turnover/industry total turnover. This may be a
limitation of the data as it is a small sample size within the entire industry but we believe

4
on the basis of comparison between the variables’ relationship that it will give us
readable results that would be a good estimate of the industry as a whole.
Administration costs will be used to form the advertising variables as this information
will contain the total spend and as Research and Development costs are specified we
can use them directly in the model.
It is also important to note that information regarding R&D expenditure is not readily
available for Irish companies within the FAME database. We will therefore use UK
companies in our model to form an understanding of the relationships between the
variables in our model and explain what we can conclude from this with reference to the
importance of the Technology industry within Ireland.
3. Methodology:
OLS Method
We will examine the relationships between the variables of our model by using
regression analysis. For a population regression, we would assume the equation of a
multi-variable regression to be as follows:
𝑌 = 𝐵1+ 𝐵2𝑥 + 𝐵3𝑥 + 𝑒
Of course the purpose of statistical analysis is to provide us with an estimate of the
overall population as it is nearly impossible to accurately assess the population mean
of the dependent variable with respect to available information. Instead, we use
sample data where we say that 𝑏1 and 𝑏2 will be estimates of 𝐵1 and 𝐵2(the
intercept and slope).
Ordinary Least Squares (OLS) is the best linear estimate as it minimizes the squared
error giving us a model that best fits the data we will use. To hold true, the OLS
involves the following assumptions:
 The regression should show the following relationship between the
dependent variable (Y) for each independent variable (X): 𝑌 = 𝐵1 + 𝐵2𝑋 + 𝑒
 The average value of the random error is: 𝐸( 𝑒) = 0
 That there is constant variance of the random error (Homoskedasticity):
𝑣𝑎𝑟( 𝑒) = 𝜎2
= 𝑣𝑎𝑟(𝑦)
 The covariance between any pair of errors is 0: 𝐶𝑜𝑣( 𝑒𝑖, 𝑒𝑗) = 0
4. Results:
4.1 Report on the coefficients.
As we recall, we are testing our model using a sample regression for the population based
on the below equation:
Mkt Share= 𝑏1 + 𝑏2𝑅&𝐷 + 𝑏3𝑎𝑑𝑣

5
We need to determine the coefficients for the parameters 𝑏1 𝑎𝑛𝑑 𝑏2 in order to determine
the slope of the line, which will tell us about the relationship between market share and the
explanatory variables r&d and advertising. After running the regression (appendix 1), we can
see that the coefficients are 𝑏2𝑟&𝑑 = .0306753and 𝑏3𝑎𝑑𝑣 = .0039812. The important
thing to note here is the sign of the coefficients; they are both positive which suggests a
positive relationship between dependent and explanatory variables meaning an increase in
r&d and advertising expenditure is expected to lead to an increase in market share.
When we look at the individual magnitude of the independent variables we can see in real
terms how a unitincrease (taken as a %) affects the dependent variable. We can explain this
more clearly that if R&D expenditure increases by 20% (20 units), then we can work out the
increase in market share as 20 ∗ 0.0306753 = .613506% increase in market share. Equally,
we can apply the same formula to measure a 20% increase in advertising expenditure to
increase market share as 20 ∗ .0039812 = .079624%. We can deduce from these results
that it takes quite a large increase in R&D and Advertising expenditure to make a significant
increase in market share.
The next step is to analyse the significance of each of the independent variables using
hypothesis testing. We need to set up two hypotheses tests as follows:
𝐻0: 𝐵2 = 0, 𝐻1: 𝐵2 ≠ 0
𝐻0: 𝐵3 = 0, 𝐻1: 𝐵3 ≠ 0
We know from economic theory that we need to use a T-test in order to accept/reject H0
(the null hypothesis). We calculated the degrees of freedom as df=N-K-1, or 30-1-1=28.
Using a .05 significance level in a two-tailed test we found that 𝑇𝑐𝑟𝑖𝑡 = 2.048. From our
regression, we can see that for our R&D variable that 𝑇𝑠𝑡𝑎𝑡 = 8.43.
Our other variable, Advertising, shows a value of𝑇𝑠𝑡𝑎𝑡 = 1.81. This means we can reject the
null hypothesis H0: B2=0 as 8.43>2.048.
We cannot, however, reject the null hypothesis H0: B3=0 using a 95% confidence level given
that 1.81<2.048.
Further confirmation of this can be seen through our p-value, the exact level of significance.
For our R&D variable, the p-value=0.00 which proves 0.05>0.00 meaning that there is a
positive relationship between R&D expenditure and Market Share. The Advertising
expenditure variable gives a p-value=0.082, which is 0.05<0.082 meaning it does not fall in
the reject region and the null hypothesis is true at a 95% confidence level. In summary, we
conclude that R&D variable is significant while our Advertising variable is not statistically
significant.
4.2 Measures of fit:
Fit:
R² is known as the coefficient of determination and is the common estimate used to test the
goodness of fit of our statistical model. The premise is that the closer to R²=1 the model is

6
than the better it is as an estimateto explain the relationship between the variables or how
well the independent variables predict the explanatory variable. The purpose is for the
model to explain the variation in the dependent variable, in this case Market share. From
our results, we can see that R²=.9096 (appendix 1) which is quite high and suggests the
model is a good fit.
Significance of the Model:
We have used t-tests to test the significance of each individual independent variable and its
relationship with the dependent variable. As this is a multi-variable regression, however, the
f-test will provide us with a better overall assessment of models significance. The test can
also be seen as a test of the significance of R² and the value of that estimator.
The F-test is another form of hypothesis testing which we define as below:
𝐻0: 𝐵1 = 0, 𝐵2 = 0 …
𝐻1: 𝐵1 𝑎𝑛𝑑𝑜𝑟 𝐵2 ≠ 0
The formula to find the F-stat is as follows:
𝐹 = (𝑅2
/𝐾)/[
1− 𝑅2
𝑛 − 𝑘 − 1
]
The calculation shows Fstat=135.83628316 (appendix 2) and we now compare this to the
Fcrit value taken from the table of statistical distributions. Our degrees of freedom will read
as (2,27) and we will again take a 5% critical value. This gives an Fcrit=3.35, and we can see
that 135.83628>3.35 and therefore we can reject the null hypothesis and state that there is
a statistical significance in our models estimates.
Specification error test (Ramsey RESET test)
We have conducted Ramsey’s RESET test to check if we have omitted any significant
variables. To do this we created the new variable,yhatsq. We must present the null and
alternative hypothesis for the value of yhatsq once more:
𝐻0: 𝐵𝑦ℎ𝑎𝑡𝑠𝑞 = 0
𝐻1: 𝐵𝑦ℎ𝑎𝑡𝑠𝑞 ≠ 0
If we reject the null hypothesis then we will conclude that we have omitted significant
variables. Choosing a 5% significance level, we compare the p-value for yhatsq as per the
RESET test (appendix 3) and find that .521>.05 and therefore accept the null and conclude
that we have not omitted significant variables from our model. We have included the
‘ovtest’ (appendix 4) we ran in Stata in our appendix which is consistent with these findings.
4.3 Violations of the assumptions of the Basic Regression Model
Heteroskedasticity:
The OLS model contains one assumption which assumes that the all variances of the random
error will remain constant, our in other words homoscedastic variances. A violation of this is
known as heteroskedasticity meaning we have differing variances in our data and that OLS
will no longer by the best linear estimator for our error.

7
We can test for the variance using graphs to detect if there is a pattern in the change in
variance. On our scatter plot (appendix 5) you can see that the value on the y axis quite high
at beginning but curves towards the x-axis at the end. This seems to indicate the presence of
heteroskedasticity, but we need to conduct a formal test in order to insure it is present in
the model.
The Breusch-Pagan test is the method we used in order to confirm our model was
heteroskedastic. Once again we present the null and alternative hypotheses:
𝐻0: 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝑠 ( 𝐻𝑜𝑚𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦)
𝐻1: 𝐷𝑖𝑓𝑓𝑒𝑟𝑖𝑛𝑔 𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝑠 (𝐻𝑒𝑡𝑒𝑟𝑜𝑠𝑘𝑒𝑑𝑎𝑠𝑡𝑖𝑐𝑖𝑡𝑦)
If the p-value of our chi square (prob>chi2) is smaller than the chosen level of significance
(.05) then we would reject the null and conclude that there is heteroskedasticity in the
model. The results of the Breusch-Pagan test (appendix 6) show p-value=.7812, .7812>.05
and therefore we cannot reject the Null and are satisfied that the assumption of constant
variances hold and that we do not need to correct for heteroskedasticity as it is not present
in our model.
Multicollinearity:
Multicollinearity is a feature of the sample and not the population and is therefore
inevitably present in our model, thus the purpose of testing for it is to measure its degree.
High multicollinearity can lead to large variances and covariances which in turn decreases
the accuracy our estimation. It would mean 𝑅2
could be quite high despite one or more t
ratios or the coefficients being statistically insignificant which would question if this
measure of goodness of fit were actually accurate.
One measure of detection for multicolinearity is if the R² is high but the individual t-tests of
the variables suggest that none or very few are significant which would suggest it could be
there to a large degree. In our model, R²=.9096 which is quite high but the advertising
variable was statistically insignificant when we conducted the t-test.
Formal testing of multicollinearity was required so we conducted the test for the Variance
Inflation Factor (VIF) to measure the extent to which it is present in our model (appendix 7).
VIF is measured as 1/Tolerance and we know that a VIF of over 10 would be a concern and
merit further investigation to see if we could identify why it was so high in our model. The
results show that for both the R&D and Advertising variable that VIF=2.75 which we deem
to be low and thus are satisfied that we can conclude our investigation of mulitcollinearity.
5. Interpretation of Results and Conclusion:
Our findings have shown that our model has yielded a good estimation of the
relationship between the variables. The results indicate that there is a positive
relationship between the dependent and independent variables and thus an increase in
R&D expenditure and advertising will yield a greater % of market share for a company in
the technology industry. We are happy that the model is a good fit, it is overall
statistically significant and that we have not omitted any important variables. Our results

8
did show that advertising was not statistically significant and therefore would question
its impact on % of Market share gained through a unit increase in this variable. We are
also satisfied that there has been no great violation of the assumptions made in the OLS
model and that it remains the appropriate and best linear estimate of our analysis
between the variables relationship. Further study in this area is required, however, as
our report is constrained by our data coming from UK companies and that information
on R&D expenditure in particular is difficult to find for Irish companies. It is also
important to note that the figures have been derived from the last available reports of
each individual company which vary and are not necessarily up to date.
The results have been in accordance with a study carried out to investigate the effect of
Executives’ Compensation on R&D expenditure, Advertising and Stock market return in
which it was found that “advertising and R&D spending is positively associated with
stock market return” (Currim et al. 2012). The model has demonstrated that firms in the
technology sector can benefit from greater investment in R&D and advertising. As was
stated at the outset, the technology sector is becoming an increasingly important area
for expanding growth in the Irish economy. The government could be informed by this
to increase incentives in the form of tax allowances, grants etc. to promote greater
investment in R&D in particular within these companies to help increase their
competiveness and performance and promote further expansion in this industry.

9
6. References
Ahearne,A.(2014) ‘The economicrecovery is most visible in the labourmarket’ IrishTimes[online],
Aug9, available:http://www.irishtimes.com/business/economy/the-economic-recovery-is-most-
visible-in-the-labour-market-1.1891630?page=2 [accessed25 Nov2014]
Currim,I. S.,Lim,J. and JoungKim,W. (2012) ‘You Get What You PayFor: The Effectof Top
Executives’CompensationonAdvertisingandR&DSpendingDecisionsandStockMarketReturn’,
Journalof Marketing [online],76(5),33-48, available:EBSCOBusinessSource Complete [accessed03
Dec 2014]
Enterprise Europe Network.(2012) Information,Communicationand Technology Sectorin Ireland
[online],available:http://www.een-
ireland.ie/eei/assets/documents/uploaded/general/ICT%20Fact%20sheet.pdf [accessed13 Nov
2014]
Grossmann,V.(2008) ‘Advertising,In-House R&D,andGrowth’, Oxford EconomicPapers [online],
60(1), 168-191, available:EconlitwithFullText[accessed26Nov2014]
SiliconRepublic(2013), Homegrown Irish tech sector thinksdifferently to overcometalent
bottleneck[online],available: http://www.siliconrepublic.com/careers/item/35189-homegrown-
irish-tech-sector[accessed13Nov2014]

10
7. Appendix:
Appendix 1: FirstRegression
Appendix2: F-Test
𝐹𝑠𝑡𝑎𝑡 = (𝑅2
/𝐾)/[
1 − 𝑅2
𝑛 − 𝑘 − 1
]
𝐹𝑠𝑡𝑎𝑡 = (
.9096
2
)/[
1 − .9096
30 − 2 − 1
]
𝐹𝑠𝑡𝑎𝑡 = .4548/(
.0904
37
)
𝐹𝑠𝑡𝑎𝑡 = 135.836283186
Appendix3: Ramsey RESET Test

11
Appendix4: Ovtest

12
Appendix5: Scatterplot (HeteroskedasticityCheck)
Appendix6: Breusch-Pagan Test (Hettest)

13
Appendix7: VIF test

�ݺ�ߣ

EC4417 Econometrics Project

Recommended

More Related Content

What's hot (11)

Viewers also liked (8)

Similar to EC4417 Econometrics Project (20)

EC4417 Econometrics Project