ECONOMETRICS: Solutions PS4

SOLUTIONS PS4:
Dummy Variables

Professor: Rodrigo Alegría

1 We have information about the average annual salary (dollars) for teachers in public
secondary schools in 45 states in the USA. Using this information, the following model was
estimated:

𝑦 = 28,694.918 − 2,954.127𝐷 − 3,112.194𝐷 − 2.34𝑥 𝑅 = 0.4977

(3,262.521) (1,862.576) (1,112.873) (0.359)

Such that 𝑥 is expenditures in public secondary schools per pupil (dollars), 𝐷 is a dummy
variable being 1 if the state is a North-eastern or North central state and 𝐷 is a dummy
variable being 1 if the state is a Southern state.

Interpret this estimated regression model and calculate the appropriate tests to validate the
model at 1% significance level.

If there is an annual increment of 1 dollar in the expenditures per pupil, on average,

there will be a decrease by 2.34 dollars in the annual salary for teachers independently
on the location of the public secondary school.

On average, and if expenditures in public secondary schools per pupil are zero
dollars, the annual salary for teachers working in a public secondary school located
in a Western state (reference category) will be 28,694.918 dollars.

On average, and if expenditures in public secondary schools per pupil are zero
dollars, the annual salary for teachers working in a public secondary school located
in a North-eastern or North central state will be 25,740.791 dollars.

On average, and if expenditures in public secondary schools per pupil are zero
dollars, the annual salary for teachers working in a public secondary school located
in a Southern state will be 25,582.724 dollars.

49.77% of the variability in annual salary is jointly explained by the behaviour of

annual expenditures per pupil and location of the secondary school. Then, we can
say that the above regression model has a medium explanatory power.

1
 ECONOMETRICS: Solutions PS4

𝑯𝟎 : 𝜷𝟏 = 𝟎, 𝑯𝒂 : 𝜷𝟏 ≠ 𝟎
𝜷 𝟐.𝟑𝟒
𝒕𝒔 = 𝒔𝒆(𝜷𝟏 in our example: 𝒕𝒔 = = −𝟔. 𝟓𝟏𝟖
𝟏) 𝟎.𝟑𝟓𝟗
𝒕𝒄 = −𝟐. 𝟕𝟎𝟒 two-tailed t-test with 41 degrees of freedom at 1% significance level.

We reject the null hypothesis since – 𝟔. 𝟓𝟏𝟖 < −2. 𝟕𝟎𝟒 at 1% significance level and
therefore, annual expenditures in public secondary schools per pupil is a statistically
significant explanatory variable that helps us to understand the behaviour of annual
salary for teachers in public secondary schools, given the above econometric
specification (partial t-test).

𝑯𝟎 : 𝜸𝟏 = 𝟎, 𝑯𝒂 : 𝜸𝟏 ≠ 𝟎
𝜸 𝟐,𝟗𝟓𝟒.𝟏𝟐𝟕
𝒕𝒔 = 𝒔𝒆(𝜸𝟏 ) in our example: 𝒕𝒔 = 𝟏,𝟖𝟔𝟐.𝟓𝟕𝟔
= −𝟏. 𝟓𝟖𝟔
𝟏
𝒕𝒄 = −𝟐. 𝟕𝟎𝟒 two-tailed t-test with 41 degrees of freedom at 1% s.l.

We fail to reject the null hypothesis since – 𝟏. 𝟓𝟖𝟔 > −2. 𝟕𝟎𝟒 at 1% significance level
and therefore, there is not a significant difference in the average annual salary
between a teacher working in a Western state and a teacher working in a North-
eastern or North central state, independently on the annual expenditures per pupil in
public secondary schools.

𝑯𝟎 : 𝜸𝟐 = 𝟎, 𝑯𝒂 : 𝜸𝟐 ≠ 𝟎
𝜸 𝟑,𝟏𝟏𝟐.𝟏𝟗𝟒
𝒕𝒔 = 𝒔𝒆(𝜸𝟏 ) in our example: 𝒕𝒔 = 𝟏,𝟏𝟏𝟐.𝟖𝟕𝟑
= −𝟐. 𝟕𝟗𝟔
𝟏
𝒕𝒄 = −𝟐. 𝟕𝟎𝟒 two-tailed t-test with 41 degrees of freedom at 1% s.l.

We reject the null hypothesis since – 𝟐. 𝟕𝟗𝟔 < −2. 𝟕𝟎𝟒 at 1% significance level and
therefore, there is a significant difference in the average annual salary between a
teacher working in a Western state and a teacher working in a Southern state,
independently on the annual expenditures per pupil in public secondary schools.

𝑯𝟎 : 𝜷𝟏 = 𝜸𝟏 = 𝜸𝟐 = 𝟎 that is, none of the explanatory variables have predictive

power.

𝑯𝒂 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly explain the behaviour of the dependent variable.

𝑹𝟐 𝒏 𝒌 𝟏 𝟎.𝟒𝟗𝟕𝟕 𝟒𝟓 𝟑 𝟏
𝑭𝒔 = 𝟏 𝑹𝟐
∗ 𝒌
in our example, 𝑭𝒔 = 𝟏 𝟎.𝟒𝟗𝟕𝟕
∗ 𝟑
= 𝟏𝟑. 𝟓𝟒

2
 ECONOMETRICS: Solutions PS4

𝑭𝒄 = 𝟒. 𝟑𝟏 F-test with n=3 and d=41 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟏𝟑. 𝟓𝟒 > 4.31 at a 1% significance level and
therefore, the three explanatory variables, location (two dummies) and annual
expenditures in public secondary schools per pupil, are jointly statistically
significant, helping us to understand the behaviour of annual salary for teachers in
public secondary schools in the USA.

2 Suppose you have survey data on wages, education, professional experience and
gender. Additionally, you have answers to the following question: how many times have you
smoked marihuana in the last month?

a- Write down an equation that allows us to estimate the effect of marihuana

consumption on wages, considering the effect of other factors. The objective is to
be able to make statements of this short: “Increasing the consumption of marihuana
in %, would change on average wages on %”

𝐥𝐨𝐠(𝐰𝐢 ) = 𝛃𝟎 + 𝛃𝟏 𝐥𝐨𝐠(𝐞𝐝𝐮𝐢 ) + 𝛃𝟐 𝐥𝐨𝐠(𝐩𝐞𝐢 ) + 𝛅𝐃𝐢 + 𝛃𝟑 𝐥𝐨𝐠(𝐦𝐢 ) + 𝐮𝐢

Such that:

𝟏 𝐢𝐟 𝐢 𝛜 𝐦𝐚𝐥𝐞
Dummy variable: 𝐃𝐢 (gender)=
𝟎 𝐢𝐟 𝐢 𝛜 𝐟𝐞𝐦𝐚𝐥𝐞 (𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐜𝐞 𝐜𝐚𝐭𝐞𝐠𝐨𝐫𝐲)

b- Specify a model that allows us to test whether the consumption of drugs has different
effects on males´ wages and females´ wages. How would you test for this difference
to be non-existent?

𝐰𝐢 = 𝛃𝟎 + 𝛃𝟏 𝐞𝐝𝐮𝐢 + 𝛃𝟐 𝐩𝐞𝐢 + 𝛅𝐃𝐢 𝐦𝐢 + 𝛃𝟑 𝐦𝐢 + 𝐮𝐢

Such that:

𝟏 𝐢𝐟 𝐢 𝛜 𝐦𝐚𝐥𝐞
Dummy variable: 𝐃𝐢 (gender)=
𝟎 𝐢𝐟 𝐢 𝛜 𝐟𝐞𝐦𝐚𝐥𝐞 (𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐜𝐞 𝐜𝐚𝐭𝐞𝐠𝐨𝐫𝐲)

I would perform an individual partial t-test for the statistical validity of the interaction
term:

𝐇𝟎 : 𝛅 = 𝟎, 𝐇𝐚 : 𝛅 ≠ 𝟎

c- Assume that marihuana consumption is measured by dividing people into 4

categories: no consumer, occasional consumer (1 to 5 times a month), moderate
consumption (6 to 10 times a month) and regular consumer (more than 10 times a

3
 ECONOMETRICS: Solutions PS4

month). Write down a model that allows us to estimate the effects of consuming
marihuana on wages.

Let´s first define the reference category which is going to be no consumer. Then,
since we have four categories, we need to define three dummy variables as follows:

Dummy variables (categorical explanatory variables):

𝟏 𝐢𝐟 𝐢 𝛜 𝐨𝐜𝐜𝐚𝐬𝐢𝐨𝐧𝐚𝐥
𝐃𝟏 (drugs)=
𝟎 𝐢𝐟 𝐢 𝛜 𝐨𝐭𝐡𝐞𝐫𝐰𝐢𝐬𝐞
𝟏 𝐢𝐟 𝐢 𝛜 𝐦𝐨𝐝𝐞𝐫𝐚𝐭𝐞
𝐃𝟐 (drugs)=
𝟎 𝐢𝐟 𝐢 𝛜 𝐨𝐭𝐡𝐞𝐫𝐰𝐢𝐬𝐞
𝟏 𝐢𝐟 𝐢 𝛜 𝐫𝐞𝐠𝐮𝐥𝐚𝐫
𝐃𝟑 (drugs)=
𝟎 𝐢𝐟 𝐢 𝛜 𝐨𝐭𝐡𝐞𝐫𝐰𝐢𝐬𝐞
Next, write down the econometric specification given the above dummy variables:

𝐰𝐢 = 𝛃𝟎 + 𝛃𝟏 𝐞𝐝𝐮𝐢 + 𝛃𝟐 𝐩𝐞𝐢 + 𝛄𝟏 𝐃𝟏𝐢 + 𝛄𝟐 𝐃𝟐𝐢 + 𝛄𝟑 𝐃𝟑𝐢 + 𝐮𝐢

3 Using the data of eight firms, a regression model was estimated to analyse the
relationship between investment in thousand Euros (𝑦 ) and production growth rate in %
(𝑥 ):

𝑦 = 3.841 − 0.0812𝑥 𝑅 = 0.466 𝑆𝑆𝑅 = 39.21 𝑛=8

(2.12) (0.038)

Additionally, two different regressions are estimated. The first one only takes into account
European firms within the original sample:

𝑦 = −0.372 + 0.108𝑥 𝑅 = 0.976 𝑆𝑆𝑅 = 0.949 𝑛=4

(0.782) (0.012)

And the second one only takes into American firms within the original sample:

𝑦 = 1.259 + 0.171𝑥 𝑅 = 0.933 𝑆𝑆𝑅 = 1.407 𝑛=4

(1.43) (0.032)

4
 ECONOMETRICS: Solutions PS4

Find whether making the distinction between European and American firms helps to
understand better the behaviour of investment and interpret your results.

𝑯𝟎 : Estimating one single regression line is more efficient.

𝑯𝒂 : Estimating two different regression lines is more efficient.

𝑺𝑺𝑹𝑻 − 𝑺𝑺𝑹𝟏 − 𝑺𝑺𝑹𝟐 𝒏 − 𝟐𝒌 − 𝟐 𝟑𝟗. 𝟐𝟏 − 𝟎. 𝟗𝟒𝟗 − 𝟏. 𝟒𝟎𝟕 𝟖 − 𝟐 − 𝟐

𝑭= ∗ = ∗ =
𝑺𝑺𝑹𝟏 + 𝑺𝑺𝑹𝟐 𝒌+𝟏 𝟎. 𝟗𝟒𝟗 + 𝟏. 𝟒𝟎𝟕 𝟐

𝑭 = 𝟑𝟏. 𝟐𝟖

𝑭𝒄 = 𝟏𝟖 F-test with n=2 and d=4 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟑𝟏. 𝟐𝟖 > 18 at a 1% significance level and
therefore, the distinction between American and European firms (structural break)
is statistically significant and therefore, regressing two different regressions is more
efficient in order to understand the behaviour of investment than regressing one
single regression without making the above distinction.

4 We have the following estimated regression model that explains the behaviour of
profits:

𝑝𝑟𝑜𝑓𝚤𝑡 = 215 − 25𝑝𝑐 + 14𝑠𝑒𝑐𝑡𝑜𝑟 − 22ℎ𝑜𝑚𝑒 − 50𝑠𝑜𝑢𝑡ℎ + 45𝑢𝑟𝑏𝑎𝑛

Such that profit is monthly profits in thousand dollars, pc is monthly production costs in
thousand dollars, sector is a sector dummy variable with a value of 1 if the sampled company
belongs to the tertiary sector, home is a nationality dummy variable equals to 1 if the sampled
company is a national company, south is a dummy variable with a value of 1 if the sampled
company is located in the south of the country and urban is a dummy variable with a value
of 1 if the sampled company is located in an urban area.

a- Find the predicted average profit for a foreign manufacturing company that is located
in a rural area at the north of the country independently of pc.

𝒑𝒓𝒐𝒇 𝒕 = 𝟐𝟏𝟓 − 𝟐𝟓(𝟎) + 𝟏𝟒(𝟎) − 𝟐𝟐(𝟎) − 𝟓𝟎(𝟎) + 𝟒𝟓(𝟎)

5
 ECONOMETRICS: Solutions PS4

𝒑𝒓𝒐𝒇 𝒕 = 𝟐𝟏𝟓 = 𝟐𝟏𝟓, 𝟎𝟎𝟎 dollars per month would be the average
predicted profits given the above firm characteristics and the above regression
model

b- Taking two companies of our sample with the same production costs, find the
estimated average difference in their monthly profit if we know that one of them is
a national manufacturing company located in a southern city of the country and the
other one is a foreign services company located in a northern city of the country.

𝒑𝒓𝒐𝒇 𝒕𝟏 = 𝟐𝟏𝟓 − 𝟐𝟓(𝟎) + 𝟏𝟒(𝟎) − 𝟐𝟐(𝟏) − 𝟓𝟎(𝟏) + 𝟒𝟓(𝟏)

𝒑𝒓𝒐𝒇 𝒕𝟏 = 𝟐𝟏𝟓 − 𝟐𝟐 − 𝟓𝟎 + 𝟒𝟓 = 𝟏𝟖𝟖, 𝟎𝟎𝟎 dollars

𝒑𝒓𝒐𝒇 𝒕𝟐 = 𝟐𝟏𝟓 − 𝟐𝟓(𝟎) + 𝟏𝟒(𝟏) − 𝟐𝟐(𝟎) − 𝟓𝟎(𝟎) + 𝟒𝟓(𝟏)

𝒑𝒓𝒐𝒇 𝒕𝟐 = 𝟐𝟏𝟓 + 𝟏𝟒 + 𝟒𝟓 = 𝟐𝟕𝟒, 𝟎𝟎𝟎 dollars

𝑫𝒊𝒇𝒇𝒆𝒓𝒆𝒏𝒄𝒆 = 𝟐𝟕𝟒 − 𝟏𝟖𝟖 = 𝟖𝟔, 𝟎𝟎𝟎 dollars per month would be the
average difference in profits between those two firms given the above
regression model.

5 The Chinese Ministry of Education is performing an analysis about the recurrent

expenditures in secondary schools in the city of Shanghai. Using a sample of 74 secondary
schools, the following estimation results are obtained:

𝑦 = 23,953.3 + 339.0432𝑥 𝑅 = 0.394 𝑆𝑆𝑅 = 8.916 𝑛 = 74

(27,167.96) (49.551)

Where the dependent variable is recurrent expenditures and the explanatory variable is
number of students in each secondary school.

However, it is believed that the type of school affects completely the behaviour of recurrent
expenditures and two different regression models are estimated distinguishing between
regular secondary schools (40 observations) and occupational secondary schools (34
observations) such that:

𝑦 = 47,974.07 + 436.7769𝑥 𝑅 = 0.634 𝑆𝑆𝑅 = 3.4895 𝑛 = 34

(33.879,03) (58,621)

6
 ECONOMETRICS: Solutions PS4

𝑦 = 51,475.25 + 152.2982𝑥 𝑅 = 0.263 𝑆𝑆𝑅 = 1.215 𝑛 = 40

(21,599.14) (41.398)

Is there a significant difference in the behaviour of recurrent expenditures between the two
types of schools? Interpret your result at 1% significance level.

𝑯𝟎 : Estimating one single regression line is more efficient.

𝑯𝒂 : Estimating two different regression lines is more efficient

𝑺𝑺𝑹𝑻 − 𝑺𝑺𝑹𝟏 − 𝑺𝑺𝑹𝟐 𝒏 − 𝟐𝒌 − 𝟐 𝟖. 𝟗𝟏𝟔 − 𝟑. 𝟒𝟖𝟗 − 𝟏. 𝟐𝟏𝟓 𝟕𝟒 − 𝟐 − 𝟐

𝑭= ∗ = ∗
𝑺𝑺𝑹𝟏 + 𝑺𝑺𝑹𝟐 𝒌+𝟏 𝟑. 𝟒𝟖𝟗 + 𝟏. 𝟐𝟏𝟓 𝟐

𝑭 = 𝟑𝟏. 𝟑𝟑

𝑭𝒄 = 𝟒. 𝟗𝟐 F-test with n=2 and d=70 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟑𝟏. 𝟑𝟑 > 4.92 at a 1% significance level and
therefore, the distinction between regular and occupational schools (structural break)
is statistically significant and therefore, regressing two different regressions is more
efficient in order to understand the behaviour of recurrent expenditures than
regressing one single regression without making the above distinction.

6 Male babies tend to weigh more than female babies. If we define a dummy variable
𝑀 = 1 for male babies and 𝑀 = 0 for female babies, the regression that explains baby´s
weigh in grams (𝑌) as a function of the number of cigarettes per day smoked by the mother
(𝑋) and the dummy variable 𝑀 is the following (sample size 𝑛 = 964):

𝑦 = 3,354 + 119𝑀 − 7𝑥 𝑅 = 0.033

(20) (26) (2.1)

Interpret this estimated regression model and calculate the appropriate tests to validate the
model.

Male babies weight, on average, 3,473gr if the mother smokes 0 cigarettes per day
whereas female babies weight, on average, 3,354gr if the mother smokes 0 cigarettes
per day. That is, male babies weight, on average, 119gr more than female babies
independently on the number of cigarettes smoked by the mother.
7
 ECONOMETRICS: Solutions PS4

If the mother smokes 1 additional cigarette per day, on average, the weight of the
baby will decrease by 7gr, independently on the baby´s gender.

3.3% of the variability of baby´s weight is jointly explained by the behaviour of gender
and number of cigarettes smoked per day by the mother. That is, the above model
has a low explanatory power.

𝐇𝟎 : 𝛃𝟏 = 𝟎, 𝐇𝐚 : 𝛃𝟏 ≠ 𝟎
𝜷 𝟕
𝒕𝒔 = 𝒔𝒆(𝜷𝟏 in our example: 𝐭 𝐬 = 𝟐.𝟏 = −𝟑. 𝟑𝟑
𝟏)
𝐭 𝐜 = −𝟐. 𝟓𝟕𝟔 two-tailed t-test with 961 degrees of freedom at 1% s. l.

We reject the null hypothesis since – 𝟑. 𝟑𝟑 < −2. 𝟓𝟕𝟔 at 1% significance level and
therefore, number of cigarettes per day smoked by the mother is a statistically
significant explanatory variable that helps us to understand the behaviour of baby´s
weight, given the above econometric specification (partial t-test).

𝐇𝟎 : 𝛄𝟏 = 𝟎, 𝐇𝐚 : 𝛄𝟏 ≠ 𝟎
𝜸 𝟏𝟏𝟗
𝒕𝒔 = 𝒔𝒆(𝜸𝟏 ) in our example: 𝐭 𝐬 = 𝟐𝟔 = 𝟒. 𝟓𝟕
𝟏
𝐭 𝐜 = 𝟐. 𝟓𝟕𝟔 two-tailed t-test with 961 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟒. 𝟓𝟕 > 2. 𝟓𝟕𝟔 at 1% significance level and
therefore, gender is a statistically significant explanatory variable that helps us to
understand the behaviour of baby´s weight, given the above econometric
specification (partial t-test).

𝐇𝟎 : 𝛃𝟏 = 𝛄𝟏 = 𝟎 that is, none of the explanatory variables have predictive power.

𝐇𝐚 : The 2 explanatory variables introduced in the above regression model are

appropriate variables that jointly help to explain the behaviour of the dependent
variable.

𝐑𝟐 𝐧 𝐤 𝟏 𝟎.𝟎𝟑𝟑 𝟗𝟔𝟒 𝟐 𝟏
𝐅𝐬 = 𝟏 𝐑𝟐
∗ 𝐤
in our example, 𝐅𝐬 = 𝟏 𝟎.𝟎𝟑𝟑
∗ 𝟐
= 𝟏𝟔. 𝟒𝟑

𝐅𝐜 = 𝟒. 𝟔𝟑 F-test with n=2 and d=961 degrees of freedom at 5% significance level.

We reject the null hypothesis since 𝟏𝟔. 𝟒𝟑 > 4.63 at a 1% significance level and
therefore, both explanatory variables, gender and number of cigarettes per day
smoked by the mother, are jointly statistically significant, helping us to understand
the behaviour of baby´s weight.
8
 ECONOMETRICS: Solutions PS4

9 Using the data of the previous exercise, a new regression model is estimated such
that (strategy 1):

𝑦 = 3,418 − 7.2𝑥 𝑅 = 0.012 𝑆𝑆𝑅 = 158.6 𝑛 = 964

(143) (2.1)

Strategy 2 consists on performing two different regressions. The first one only takes into
account babies that are first-born (their mothers have not got previous births):

𝑦 = 3,363 − 4.0𝑥 𝑅 = 0.004 𝑆𝑆𝑅 = 91.2 𝑛 = 584

(18) (2.8)

And the second one only takes into account babies that are not first-born (their mothers
have got previous births):

𝑦 = 3,506 − 12.1𝑥 𝑅 = 0.039 𝑆𝑆𝑅 = 63.5 𝑛 = 380

(23) (3.1)

Find the most appropriate strategy to better understand the behaviour of the dependent
variable (structural break?) and interpret your results.

𝐇𝟎 : Estimating one single regression line is more efficient.

𝐇𝐚 : Estimating two separate regression lines is more efficient

𝐒𝐒𝐑 𝐓 − 𝐒𝐒𝐑 𝟏 − 𝐒𝐒𝐑 𝟐 𝐧 − 𝟐𝐤 − 𝟐 𝟏𝟓𝟖. 𝟔 − 𝟗𝟏. 𝟐 − 𝟔𝟑. 𝟓 𝟗𝟔𝟒 − 𝟐 − 𝟐

𝐅= ∗ = ∗ =
𝐒𝐒𝐑 𝟏 + 𝐒𝐒𝐑 𝟐 𝐤+𝟏 𝟗𝟏. 𝟐 + 𝟔𝟑. 𝟓 𝟐

𝐅 = 𝟏𝟐. 𝟏

𝐅𝐜 = 𝟒. 𝟔𝟑 F-test with n=2 and d=960 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟏𝟐. 𝟏 > 4.63 at a 1% significance level and
therefore, the distinction between first born and non first born babies (structural
break) is statistically significant and therefore, regressing two different regressions is
more efficient in order to understand the behaviour of baby´s weight than regressing
one single regression without making the above distinction.

9
 ECONOMETRICS: Solutions PS4

8 The variable s denotes the time invested in sleeping at night (minutes per week), w is
the time invested in working (minutes per week), e (level of education) and a (age of the
individual) are measured in years and m is a dummy variable with a value of 1 if the individual
is a male. Sample size is 706 individuals.

𝑠 = 3,840.83 + 0.163𝑤 − 11.7𝑒 − 8.7𝑎 − 0.128𝑎 + 87.75𝑚 𝑅 = 0.117

(235.22) (0.018) (5.86) (11.21) (0.134) (29.33)

a- Interpret this estimated regression model.

On average, and independently of the rest of variables, female individuals sleep

3,840.83 minutes per week while male individuals sleep 87.75 minutes per week more
than females.

On average, and independently of gender, if we increase by 1 minute per week the

time invested in working, sleeping time increases by 0,163 minutes per week
maintaining the rest of the model as a constant term.

On average, and independently of gender, if we increase by 1 year the level of

education, sleeping time decreases by 11.7 minutes per week holding the rest of the
model as a constant term.

On average, and independently of gender, if we increase by 1 year the age of the

individual, sleeping time decreases by 8.7 minutes per week at a decreasing rate
holding the rest of the model as a constant term.

The estimated regression line explains 11.7% of the total variability of the time
invested in sleeping at night meaning the model has a low explanatory power.

a- Are the effects of the variable age statistically significant?

𝐇𝟎 : 𝛃𝟑 = 𝟎, 𝐇𝐚 : 𝛃𝟑 ≠ 𝟎
𝜷 𝟖.𝟕
𝒕𝒔 = 𝒔𝒆(𝜷𝟑 ) in our example: 𝐭 𝐬 = 𝟏𝟏.𝟐𝟏 = −𝟎. 𝟕𝟕𝟔
𝟑
𝐭 𝐜 = −𝟐. 𝟓𝟕𝟖 two-tailed t-test with 700 degrees of freedom at 1% s. l.

We fail to reject the null hypothesis since – 𝟎. 𝟕𝟕𝟔 > −2. 𝟓𝟖𝟒 at 1% significance level
and therefore, age is not a statistically significant explanatory variable not helping us
to understand the behaviour of sleeping time at night, given the above econometric
specification (partial t-test).

10
 ECONOMETRICS: Solutions PS4

𝐇𝟎 : 𝛃𝟒 = 𝟎, 𝐇𝐚 : 𝛃𝟒 ≠ 𝟎
𝜷 𝟎.𝟏𝟐𝟖
𝒕𝒔 = 𝒔𝒆(𝜷𝟒 ) in our example: 𝐭 𝐬 = 𝟎.𝟏𝟑𝟒
= −𝟎. 𝟗𝟓𝟓
𝟒
𝐭 𝐜 = −𝟐. 𝟓𝟕𝟖 two-tailed t-test with 700 degrees of freedom at 1% s. l.

We fail to reject the null hypothesis since – 𝟎. 𝟗𝟓𝟓 > −2. 𝟓𝟖𝟒 at 1% significance level
and therefore, the decreasing returns in the effect of age is not a statistically
significant explanatory variable not helping us to understand the behaviour of
sleeping time at night, given the above econometric specification (partial t-test).

b- Is there evidence to say that males sleep more than females?

𝐇𝟎 : 𝛃𝟓 = 𝟎, 𝐇𝐚 : 𝛃𝟓 ≠ 𝟎
𝜷 𝟖𝟕.𝟕𝟓
𝒕𝒔 = 𝒔𝒆(𝜷𝟓 ) in our example: 𝐭 𝐬 = 𝟐𝟗.𝟑𝟑 = 𝟐. 𝟗𝟗𝟏
𝟓
𝐭 𝐜 = 𝟐. 𝟓𝟕𝟖 two-tailed t-test with 700 degrees of freedom at 1% s. l.

We reject the null hypothesis since 𝟐. 𝟗𝟗𝟏 > 2. 𝟓𝟖𝟒 at 1% significance level and
therefore, there is enough evidence to say that males tend to sleep more than females
independently of other factors (partial t-test).

c- Is the above model globally significant?

𝐇𝟎 : 𝛃𝟏 = 𝛃𝟐 = 𝛃𝟑 = 𝛃𝟒 = 𝛃𝟓 = 𝟎that is, none of the explanatory variables have

predictive power.

𝐇𝐚 : The 5 explanatory variables introduced in the above regression model are

appropriate variables that jointly help to explain the behaviour of the dependent
variable.

𝐑𝟐 𝐧 𝐤 𝟏 𝟎.𝟏𝟏𝟕 𝟕𝟎𝟔 𝟓 𝟏
𝐅𝐬 = 𝟏 𝐑𝟐
∗ 𝐤
in our example, 𝐅𝐬 = 𝟏 𝟎𝟏𝟏𝟕
∗ 𝟓
= 𝟏𝟖. 𝟒𝟖

𝐅𝐜 = 𝟑. 𝟎𝟒 F-test with n=5 and d=700 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟏𝟖. 𝟒𝟖 > 3.04 at a 1% significance level and
therefore, all the explanatory variablesare jointly statistically significant, helping us
to understand the behaviour of sleeping time at night.

11
 ECONOMETRICS: Solutions PS4

9 We have the following estimated regression model that explains the behavior of
salaries:

log (𝑤𝑎𝑔𝑒 ) = 300 + 0.05𝑒𝑑𝑢 + 0.07𝑒𝑥𝑝 + 0.14𝑚𝑎𝑙𝑒 − 0.12𝑏𝑙𝑎𝑐𝑘

− 0.002𝑠𝑜𝑢𝑡ℎ + 0.06𝑢𝑟𝑏𝑎𝑛 + 0.08𝑚𝑎𝑟𝑟𝑖𝑒𝑑

Such that wage is the weekly salary in dollars, edu is years of education, exp is years of
professional experience, male is a gender dummy variable with a value of 1 if the sampled
individual is a male, black is a race dummy variable with a value of 1 if the sampled individual
is black, south is a dummy variable with a value of 1 if the sampled individual lives in the
south of the country, urban is a dummy variable with a value of 1 if the sampled individual
lives in an urban area and married is a dummy variable with a value of 1 if the sampled
individual is a married individual.

a- Which would be the predicted average salary difference for a black single female that
lives in a urban area at the north of the country respect the reference category and
independently of edu and exp?

𝐥𝐨𝐠(𝐰𝐚𝐠𝐞 ) = 𝟎. 𝟏𝟒(𝟎) − 𝟎. 𝟏𝟐(𝟏) + 𝟏𝟒(𝟎) − 𝟎. 𝟎𝟎𝟐(𝟎) + 𝟎. 𝟎𝟔(𝟏)

+ 𝟎. 𝟎𝟖(𝟎)

∆𝐰𝐚𝐠𝐞 =6% per week less would be the average predicted weekly salary
difference of the above individual respect to the reference category (all the
dummies being zero) given the above individual characteristics and the above
regression model.

b- Taking two males from our sample with the same years of education and the same
years of professional experience, which would be the estimated average difference in
their weekly salary if we know that one of them is black, single and lives in a southern
city of the country and the other one is white, married and lives in a northern city of
the country?

Following the same reasoning as before, the first individual will have a weekly
salary that is, on average, 20.2% lower than the second individual.

12
 ECONOMETRICS: Solutions PS4

10 Using the data of the mid-term exam results, the Econometrics teacher estimates the
following regression model (strategy 1):

𝑒𝑥𝑎𝑚 = −1.338 + 0.648𝑝𝑎𝑟𝑡 + 0.284𝑝𝑠 𝑅 = 0.433 𝑆𝑆𝑅 = 465.1 𝑛 = 171

Strategy 2 consists on performing two different regressions. The first one only takes into
account students that are foreign individuals:

𝑒𝑥𝑎𝑚 = −0.485 + 0.521𝑝𝑎𝑟𝑡 + 0.0.326𝑝𝑠 𝑅 = 0.432 𝑆𝑆𝑅 = 175.4 𝑛 = 73

And the second one only takes into account Spanish students:

𝑒𝑥𝑎𝑚 = −1.961 + 0.794𝑝𝑎𝑟𝑡 + 0.198𝑝𝑠 𝑅 = 0.442 𝑆𝑆𝑅 = 266.1 𝑛 = 98

a- Find the most appropriate strategy to better understand the behavior of the mid-
term exam gradesand interpret your results.

𝐇𝟎 : Estimating one single regression line is more efficient.

𝐇𝐚 : Estimating two separate regression lines is more efficient

𝐒𝐒𝐑 𝐓 − 𝐒𝐒𝐑 𝟏 − 𝐒𝐒𝐑 𝟐 𝐧 − 𝟐𝐤 − 𝟐 𝟒𝟔𝟓. 𝟏 − 𝟏𝟕𝟓. 𝟒 − 𝟐𝟔𝟔. 𝟏 𝟏𝟕𝟏 − 𝟒 − 𝟐

𝐅= ∗ = ∗
𝐒𝐒𝐑 𝟏 + 𝐒𝐒𝐑 𝟐 𝐤+𝟏 𝟏𝟕𝟓. 𝟒 + 𝟐𝟔𝟔. 𝟏 𝟑
= 𝟐. 𝟗𝟏𝟓

𝐅𝐜 = 𝟑. 𝟗𝟏 F-test with n=3 and d=165 degrees of freedom at 1% significance level.

We fail to reject the null hypothesis since 𝟐. 𝟗𝟏 < 3.91 at a 1% significance level and
therefore, the distinction between Spanish and foreign students (structural break
given by nationality) is statistically insignificant and therefore, regressing two
different regressions is less efficient in order to understand the behaviour of grades
than regressing one single regression without making the above distinction.

b- Specify a model in which you could test directly if there is a difference in the
performance of the mid-term exam depending on whether the student is Spanish or
a foreign student and independently of other factors.

𝒆𝒙𝒂𝒎𝒊 = 𝜷𝟎 + 𝜷𝟏 𝒑𝒂𝒓𝒕𝒊 + 𝜷𝟐 𝒑𝒔𝒊 + 𝜸𝑫𝒊 + 𝒖𝒊

𝟏𝐢𝐟𝐢 𝛜 𝐒𝐩𝐚𝐧𝐢𝐬𝐡
Dummy variable: 𝐃𝐢 (nationality)=
𝟎𝐢𝐟𝐢 𝛜 𝐟𝐨𝐫𝐞𝐢𝐠𝐧 (𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐜𝐞𝐜𝐚𝐭𝐞𝐠𝐨𝐫𝐲)

13
 ECONOMETRICS: Solutions PS4

11 We have the following estimated regression model:

log (𝑤 ) = 1.6 − 0.32𝑓𝑒𝑚 + 0.16log (𝑠𝑖𝑧𝑒 ) + 0.05𝑒𝑑𝑢 𝑅 = 0.31 𝑆𝑆𝑅 = 359

(0.02) (0.02) (0.02) (0.002)

Such that 𝑤 measures salaries in thousand dollars for each of our 2,000 sampled individuals,
𝑒𝑑𝑢 measures education in years, 𝑓𝑒𝑚 is a gender dummy variable with a value of 1 if the
individual i is a female, and size is a variable measuring the number of workers working in
the company.

a- Interpret the above estimation results.

No matter the gender and the level of education, 1% increase in the size of the
company produces, on average, a 0.16% increase in salaries. Additionally, no matter
the gender and independently of the company size, an additional year of education
produces, on average, salaries to increase by 5%. Finally, and independently of
education and company size, a female salary 32% lower than a male salary on
average.

b- Are gender and company size categorical factors statistically significant?

We need to perform two individual t-tests:

𝐇𝟎 : 𝛃𝟏 = 𝟎, 𝐇𝐚 : 𝛃𝟏 ≠ 𝟎
𝜷 𝟎.𝟑𝟐
𝒕𝒔 = 𝒔𝒆(𝜷𝟏 in our example: 𝐭 𝐬 = = −𝟏𝟔
𝟏) 𝟎.𝟎𝟐
𝐭 𝐜 = −𝟐. 𝟓𝟕𝟔 two-tailed t-test with 1,996 degrees of freedom at 1% s. l.

We reject the null hypothesis since – 𝟏𝟔 < −2. 𝟓𝟕𝟔 at 1% significance level and
therefore, gender is statistically significant explanatory variable that helps us to
understand the behaviour of salaries, given the above econometric specification
(partial t-test).

𝐇𝟎 : 𝛃𝟐 = 𝟎, 𝐇𝐚 : 𝛃𝟐 ≠ 𝟎
𝜷 𝟎.𝟏𝟔
𝒕𝒔 = 𝒔𝒆(𝜷𝟐 in our example: 𝐭 𝐬 = 𝟎.𝟎𝟐 = 𝟖
𝟐)
𝐭 𝐜 = 𝟐. 𝟓𝟕𝟔 two-tailed t-test with 1,996 degrees of freedom at 1% s. l.

We reject the null hypothesis since 𝟖 > 2. 𝟓𝟕𝟔 at 1% significance level and therefore,
company size is a statistically significant explanatory variable that helps us to
understand the behaviour of salaries, given the above econometric specification
(partial t-test).

14
 ECONOMETRICS: Solutions PS4

We estimate an alternative model changing the previous specification such that:

log (𝑤 ) = 1.6 − 0.26𝑓𝑒𝑚 + 0.18log (𝑠𝑖𝑧𝑒 ) + 0.05𝑒𝑑𝑢 − 0.16𝑓𝑒𝑚 ∗ log (𝑠𝑖𝑧𝑒 )

𝑅 = 0.32 𝑆𝑆𝑅 = 341

c- Do small companies discriminate against women more or less than larger firms? Is
the discrimination statistically significant?

According to the coefficient associated to the interaction effect, small companies, on

average, discriminate less than big companies.

To see whether it is a significant effect we need to perform a t-test for the interaction
effect. However, we do not know the standard error associated to this coefficient. An
alternative test that can be used is the F-test for the addition of new regressors such
that:

𝑯𝟎 : 𝜷𝟒 = 𝟎 that is, this new explanatory variable has not predictive power.

𝑯𝒂 : The 1 explanatory variable introduced in the above regression model is an

appropriate variable that explains the behaviour of the dependent variable.

𝑺𝑺𝑹𝒌 − 𝑺𝑺𝑹𝒎 𝒏 − 𝒎 − 𝟏 𝟑𝟓𝟗 − 𝟑𝟒𝟏 𝟐𝟎𝟎𝟎 − 𝟒 − 𝟏

𝑭𝒔 = ∗ = ∗ = 𝟏𝟎𝟓. 𝟑𝟎𝟕
𝑺𝑺𝑹𝒎 𝒎−𝒌 𝟑𝟒𝟏 𝟒−𝟑

𝑭𝒄 = 𝟔. 𝟑𝟔 F-test with n=1 and d=1,995 degrees of freedom at 1%

We reject the null hypothesis since 𝟏𝟎𝟓. 𝟑𝟎𝟕 > 𝟔. 𝟔𝟑 at a 1% significance level and
therefore, the new explanatory variable (interaction effect) introduced in the second
model is improving the prediction power of our model or help us to understand better
the behaviour of our dependent variable. As a consequence, our preferred
specification will be the second model.

Therefore, the interaction effect can be considered as a significant effect.

12 The following model is regressed using data in quarterly form from 1990 to 2005 (64
observations) for Malaysian stock prices against output knowing that there was an economic
crisis in 1997.

𝑌 = α +𝛽𝑋 + 𝑈

The first regression using all the data produced a RSS of 0.56. Then, two regressions were
run. The first one on a subsample of the data from 1990-1997, giving a RSS of 0.23. The

15
 ECONOMETRICS: Solutions PS4

second one was on the simple from 1998 to 2005, producing a RSS of 0.17. Test whether the
crisis in 1997 produced a significant shock in the behaviour of Malaysian stock prices.

𝑯𝟎 : Estimating one single regression line is more efficient.

𝑯𝒂 : Estimating two separate regression lines is more efficient.

𝟎. 𝟓𝟔 − 𝟎. 𝟐𝟑 − 𝟎. 𝟏𝟕 𝟔𝟒 − 𝟐 ∗ 𝟏 − 𝟐
𝑭= ∗ = 𝟏𝟐
𝟎. 𝟐𝟑 + 𝟎. 𝟏𝟕 𝟏+𝟏
𝒅𝒇𝒏 = 𝟐 (𝐧𝐮𝐦𝐞𝐫𝐚𝐭𝐨𝐫) 𝐚𝐧𝐝 𝒅𝒇𝒅 = 𝟔𝟎 (𝐝𝐞𝐧𝐨𝐦𝐢𝐧𝐚𝐭𝐨𝐫)

𝑭𝒄 = 𝑭𝟐, 𝟔𝟎 (𝟏 − 𝟎, 𝟎𝟓) = 𝟑. 𝟏𝟓

Since 𝟏𝟐 > 𝟑. 𝟏𝟓, I reject the null hypothesis at 5% significance level.

We conclude that there is a significant structural break in this model and therefore
we need to split the data into two sub-samples in order to overcome the break.
Alternatively, regressing two different models is more efficient in order to understand
the behaviour of Malaysian stock prices during the above time period of time, and
taking into account the Asian financial crisis that took placed in 1997.

13 We have the following estimated regression model:

𝑦 = 186.4 + 2.33𝑥 − 126𝐷 − 1.29𝐷 𝑥 𝑅 = 0.5055 𝑛 = 34

(45.67) (0.86) (37.01) (1.02)

Such that 𝑦 measures annual expenditure on beer in dollars for each of our 34 sampled
individuals, 𝑥 measures individual annual income in thousand dollars and 𝐷 is a dummy
variable with a value of 1 if the individual i is a female and 0 if the individual is a male.

a- What will be the difference in consumption between a male and a female with the
same annual income?

On average, and independently on annual income, females annually consume 126

dollars less in beer than males.

16
 ECONOMETRICS: Solutions PS4

b- Test at 1% level the following: there are no differences in beer consumption across
gender.

𝐇𝟎 : 𝛄 = 𝟎, 𝐇𝐚 : 𝛄 ≠ 𝟎
𝜸 𝟏𝟐𝟔
𝒕𝒔 = 𝒔𝒆(𝜸) in our example: 𝐭 𝐬 = 𝟑𝟕.𝟎𝟏 = −𝟑. 𝟒
𝐭 𝐜 = −𝟐. 𝟕𝟓 two-tailed t-test with 30 degrees of freedom at 1% significance level.

We reject the null hypothesis since – 𝟑. 𝟒 < −2.75 at 1% significance level and
therefore, there is a significant additive difference in the average beer consumption
between females and males, independently on the annual income. On average, males
consume significantly more beer than females.

c- Test at 5% level the following: there are no differences in the marginal propensity to
consume beer respect to income across gender.

𝐇𝟎 : 𝛅 = 𝟎, 𝐇𝐚 : 𝛅 ≠ 𝟎
𝜹 𝟏.𝟐𝟗
𝒕𝒔 = 𝒔𝒆(𝜹) in our example: 𝐭 𝐬 = 𝟏.𝟎𝟐
= −𝟏. 𝟐𝟔
𝐭 𝐜 = −𝟐. 𝟎𝟒𝟐 two-tailed t-test with 30 degrees of freedom at 5% significance level.

We fail to reject the null hypothesis since – 𝟏. 𝟐𝟔 > −2.042 at 1% significance level
and therefore, there is not a significant gender difference in the effect of annual
income on beer consumption between females and males. On average, the effect of
annual income on beer consumption is significantly the same for males and females.

14 We have a housing price model with the following variables: price (house prices), sqrft
(house size), bdrms (number of bedrooms) and colonial (dummy variable equal to one if the
house is of the colonial style. The estimation results are the following (sample size is 88
houses):

log (𝑝𝑟𝚤𝑐𝑒 ) = 5.56 + 0.707 log(𝑠𝑞𝑟𝑓𝑡 ) + 0.027𝑏𝑑𝑟𝑚𝑠 + 0.054𝑐𝑜𝑙𝑜𝑛𝑖𝑎𝑙

(0.65) (0.093) (0.029) (0.045)

a- Interpret this estimated regression model.

Independently of the number of bedrooms and colonial, increasing by 1% the size of

the house will produce, on average, 0.707% increment in house prices. Holding the
size as a constant and independently of colonial, if we increase by 1 the number of
bedrooms, on average, house prices will increase by 2.7%. Finally and no matter the

17
 ECONOMETRICS: Solutions PS4

size and the number of bedrooms, a colonial house will be, on average, 5.5% more
expensive than a non-colonial house.

b- Is the effect of the variable bdrms statistically significant?

𝐇𝟎 : 𝛃𝟏 = 𝟎, 𝐇𝐚 : 𝛃𝟏 ≠ 𝟎
𝜷 𝟎.𝟎𝟐𝟕
𝒕𝒔 = 𝒔𝒆(𝜷𝟏 in our example: 𝐭 𝐬 = 𝟎.𝟎𝟐𝟗 = 𝟎. 𝟗𝟑𝟏
𝟏)
𝐭 𝐜 = 𝟐. 𝟔𝟑𝟗 two-tailed t-test with 84 degrees of freedom at 1% significance level.

We fail to reject the null hypothesis since 𝟐. 𝟔𝟑𝟗 > 0.931 at 1% significance level and
therefore, number of bedrooms is a statistically insignificant explanatory variable.
This might be due to the inclusion of a similar size effect with the size variable (first
independent factor).

c- Is there a significant evidence to say that colonial houses are more expensive than
the rest of the houses independently of the rest of the factors?

𝐇𝟎 : 𝛄𝟏 = 𝟎, 𝐇𝐚 : 𝛄𝟏 ≠ 𝟎
𝜸 𝟎.𝟎𝟓𝟒
𝒕𝒔 = 𝒔𝒆(𝜸) in our example: 𝐭 𝐬 = 𝟎.𝟎𝟒𝟓 = 𝟏. 𝟐
𝐭 𝐜 = 𝟐. 𝟔𝟑𝟗 two-tailed t-test with 84 degrees of freedom at 1% significance level.

We fail to reject the null hypothesis since 𝟐. 𝟔𝟔 > 1.2 at 1% significance level and
therefore, there is not a significant difference in house prices between a colonial and
a non-colonial house.

d- Is the above model globally significant knowing that 𝑅 = 0.649?

𝐇𝟎 : 𝛃𝟏 = 𝛃𝟐 = 𝛄𝟏 = 𝟎 that is, none of the explanatory variables have predictive

power.

𝐇𝐚 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly help to explain the behaviour of the dependent
variable.

𝐑𝟐 𝐧 𝐤 𝟏 𝟎.𝟔𝟒𝟗 𝟖𝟖 𝟑 𝟏
𝐅𝐬 = 𝟏 𝐑𝟐
∗ 𝐤
in our example, 𝐅𝐬 = 𝟏 𝟎.𝟔𝟒𝟗
∗ 𝟑
= 𝟓𝟏. 𝟕𝟕

𝐅𝐜 = 𝟒. 𝟎𝟒 F-test with n=3 and d=84 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟓𝟏. 𝟕𝟕 > 4.04 at a 1% significance level and
therefore, the three explanatory variables, size, number of bedrooms and colonial, are
jointly statistically significant, helping us to understand the behaviour of housing
prices.

18
 ECONOMETRICS: Solutions PS4

15 The following stock price model was regressed using monthly data from 1980m1 to
1989m12:

𝑠 = 𝛼 + 𝛽𝑦 + 𝑢

It is believed there is a structural break at 1987m11, following a stock market crash. The
regression using all the data produced a SSR of 0.97. Then two further regressions were run
from 1980m1 to 1987m11, which produced a SSR of 0.58 and another regression from
1987m12 to 1989m12 produced a SSR of 0.32.

a- Do you think the stock market crash at 1987m11 was statistically significant?

𝐇𝟎 : Estimating one single regression line is more efficient.

𝐇𝐚 : Estimating two separate regression lines is more efficient

𝐒𝐒𝐑 𝐓 − 𝐒𝐒𝐑 𝟏 − 𝐒𝐒𝐑 𝟐 𝐧 − 𝟐𝐤 − 𝟐 𝟎. 𝟗𝟕 − 𝟎. 𝟓𝟖 − 𝟎. 𝟑𝟐 𝟏𝟐𝟎 − 𝟐 − 𝟐

𝐅= ∗ = ∗ =
𝐒𝐒𝐑 𝟏 + 𝐒𝐒𝐑 𝟐 𝐤+𝟏 𝟎. 𝟓𝟖 + 𝟎. 𝟑𝟐 𝟐

𝐅 = 𝟒. 𝟓𝟏

𝐅𝐜 = 𝟒. 𝟖𝟐 F-test with n=2 and d=116 degrees of freedom at 1% significance level.

We fail to reject the null hypothesis since 𝟒. 𝟖𝟐 > 4.04 at a 1% significance level and
therefore, the stock market crash was not statistically significant and the parameters
are stable. Therefore, one regression model, taking into account all the observation
is the most efficient solution.

b- Why are structural breaks a problem for financial econometrics? Give examples of
some recent structural breaks.

They might be a problem because they are exogenous shocks in the model that may
produce the model to become unstable affecting the structure of the model.
Examples: any economic crisis you may think about.

19
 ECONOMETRICS: Solutions PS4

16 We have the following estimated regression model:

𝑦 = 100 − 0.8𝑥 − 0.3𝐷 𝑥 𝑅 = 0.47 𝑛 = 55

(27) (0.05) (0.01)

Such that 𝑦 measures profits in thousand dollars for each of our 55 sampled companies, 𝑥
measures production costs in thousand dollars and 𝐷 is a dummy variable with a value of 1
if the company i is a manufacturing firm and 0 if the company is a services firm.

a- Interpret the estimated regression model.

Average profits when production costs are zero will be 100,000 dollars. In addition,
for a services company, an increment of 1,000 dollars in production
costs will produce, on average, a decrease of 800 dollars in profits whereas if
the company is a manufacturing firm, 1,000 dollars increment in production
costs will produce, on average, a decrease of 1,100 dollars in profits.

b- Is 𝐷 𝑥 a significant explanatory variable? Why? Explain your answer.

𝐇𝟎 : 𝛅 = 𝟎, 𝐇𝐚 : 𝛅 ≠ 𝟎
𝜹 𝟎.𝟑
𝒕𝒔 = 𝒔𝒆(𝜹) in our example: 𝐭 𝐬 = 𝟎.𝟎𝟏
= −𝟑𝟎
𝐭 𝐜 = −𝟐. 𝟔𝟕𝟖 two-tailed t-test with 52 degrees of freedom at 1% significance level.

We reject the null hypothesis since – 𝟐. 𝟔𝟕𝟖 > −30 at 1% significance level and
therefore, therefore, the interaction effect is a statistically significant explanatory
variable that helps us to understand the behaviour of profits, given the above
econometric specification (partial t-test).That is, there is a significant difference in
the effect of production costs on profits depending on the sectoral type of the
company.

c- Find the predicted profits for a manufacturing company with 55,000 dollars as
production costs.

𝒚 = 𝟏𝟎𝟎, 𝟎𝟎𝟎 − 𝟎. 𝟖(𝟓𝟓, 𝟎𝟎𝟎) − 𝟎. 𝟑 ∗ (𝟏) ∗ (𝟓𝟓, 𝟎𝟎𝟎) = 𝟑𝟗, 𝟓𝟎𝟎 𝑫𝒐𝒍𝒍𝒂𝒓𝒔

20
 ECONOMETRICS: Solutions PS4

17 Let´s consider the following regression model using a sample of annual data from
1970 until 2001 (both included) for the Castilla-León economy:

𝑦 = 134.6 + 10.89𝑥 + 21.6𝐷 + 3.91𝐷 𝑥 𝑅 = 0.921

(56.2) (2.06) (3.99) (0.91)

Such that 𝑦 are annual regional exports, 𝑥 is the annual exchange rate (pts/$) and 𝐷 is a
dummy variable equals to 1 if 𝑡 ≤ 1985 and equals to 0 if 𝑡 > 1985 (Spain being a
European Union member).

a- Interpret the above regression model.

On average and assuming that the annual exchange rate is zero, regional exports are
134.6 units if 𝒕 > 1985. Similarly, if 𝒕 ≤ 𝟏𝟗𝟖𝟓, regional exports are, on average, 156.2
units assuming that the annual exchange rate is zero.

For 𝒕 ≤ 𝟏𝟗𝟖𝟓, if the exchange rate increases by 1 unit, on average, regional exports
will increase by 14.8 units whereas for 𝒕 > 1985, the same unit change in the
exchange rate produces, on average, an increment in regional exports of 10.89 units.

92.1% of the variability in regional exports is explained through the above

econometric specification. That is, we can say the regression model has a high
explanatory power.

b- Test the validity of the additive and multiplicative dummy effects in the above model
at 1% significance level.

𝑯𝟎 : 𝜸 = 𝟎, 𝑯𝒂 : 𝜸 ≠ 𝟎
𝜸 𝟐𝟏.𝟔
𝒕𝒔 = 𝒔𝒆(𝜸) in our example: 𝒕𝒔 = 𝟑.𝟗𝟗 = 𝟓. 𝟒𝟏
𝒕𝒄 = 𝟐, 𝟕𝟔𝟑 two-tailed t-test with 28 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟓. 𝟒𝟏 > 2.763 at 1% significance level and
therefore, there is a significant additive difference in the average regional exports
before and after Spain become a member of the EU, independently on the exchange
rate. On average, regional exports were significantly greater before 1985 than after
1985.

𝑯𝟎 : 𝜹 = 𝟎, 𝑯𝒂 : 𝜹 ≠ 𝟎
𝜹 𝟑.𝟗𝟏
𝒕𝒔 = 𝒔𝒆(𝜹) in our example: 𝒕𝒔 = 𝟎.𝟗𝟏 = 𝟒. 𝟐𝟗

21
 ECONOMETRICS: Solutions PS4

𝒕𝒄 = 𝟐. 𝟕𝟔𝟑 two-tailed t-test with 28 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟒. 𝟐𝟗 > 2.763 at 1% significance level and
therefore, there is a significant multiplicative difference in the effect of exchange rate
on regional exports before and after Spain become a member of the EU. On average,
the effect of the exchange rate on regional exports was significantly greater before
1985 than after 1985.

c- Test the overall fit of the model at 5% significance level.

𝑯𝟎 : 𝜷𝟏 = 𝜸 = 𝜹 = 𝟎 that is, none of the explanatory variables have predictive power.

𝑯𝒂 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly explain the behaviour of the dependent variable.

𝑹𝟐 𝒏 𝒌 𝟏 𝟎.𝟗𝟐𝟏 𝟑𝟐 𝟑 𝟏
𝑭𝒔 = 𝟏 𝑹𝟐
∗ 𝒌
in our example, 𝑭𝒔 = 𝟏 𝟎.𝟗𝟐𝟏
∗ 𝟑
= 𝟏𝟎𝟖. 𝟖

𝑭𝒄 = 𝟐. 𝟗𝟓 F-test with n=3 and d=28 degrees of freedom at 1% significance level.

We reject the null hypothesis since 𝟏𝟎𝟖. 𝟖 > 2.95 at a 1% significance level and
therefore, the three explanatory variables, exchange rate, the additive dummy
variable and the interaction dummy variables, are jointly statistically significant,
helping us to understand the behaviour of regional exports of Castile-Leon economy
during the time period 1970-2001.

18 The following wage equations have been estimated using data on workers from
Vietnam:

log(𝑠𝑎𝑙𝑎𝑟𝑦) = 1.25 + 0.15𝑔𝑒𝑛𝑑𝑒𝑟 + 0.02𝑒𝑥𝑝

(0.35) (0.03) (0.004)

log(𝑠𝑎𝑙𝑎𝑟𝑦) = 1.55 + 0.10𝑔𝑒𝑛𝑑𝑒𝑟 + 0.015𝑒𝑥𝑝 − 0.0005𝑔𝑒𝑛𝑑𝑒𝑟 ∗ 𝑒𝑥𝑝

(0.48) (0.05) (0.005) (0.002)

Where salary is measured in US dollars and gender is a dummy variable taking the value of 1
if the worker is a male and 0 if the worker is a female, exp measures the years of work
experience.

22
 ECONOMETRICS: Solutions PS4

a- Why the coefficients associated to gender and experience are lower in the second
than in the first model?
b- What is the estimated average difference between a man´s salary with 5 years work
experience and that of a woman´s with 10 years work experience according to the
first model?
c- What is the estimated average difference between a man´s salary with 5 years work
experience and that of a woman´s with 10 years work experience according to the
second model?
d- Test that the salary difference between men and women does not depend on
experience.

19 To see whether people living in urban areas spend more on fish than people living
in rural areas, we get the following estimation results:

OLS Estimation results

Dependent: log(expenditure in fish)
Explanatory OLS t-statistic Degrees of Significance t-critical
Variable Coefficient Freedom level
Intercept 6.375 36 0.01 -
log(income) 1.313 5.328 36 0.01 2.719
gender -0.055 -1.378 36 0.01 -2.719
urban 0.143 10.311 36 0.01 2.719

Sample size (n) 40 F-statistic Degrees of Significance F-critical

Freedom level
R-squared 0.750 36 3(n),36(d) 0.01 4.40

Where the dependent variable is expenditure in fish (with log), income is disposable income
(with log), gender is a gender dummy with 1 if male and 0 if female and urban is another
dummy which takes the value 1 if person lives in an urban area. Please, answer the
following three questions:

a- Interpret the above estimations results (only the value of the OLS coefficient for each
of the explanatory variables).

No matter the gender and whether the individual is living in an urban or rural area,
if you increase income by 1%, on average, expenditures in fish will increase by 1.313%.
This seems to be a realistic result as fish tend to be an expensive good within food
goods.

23
 ECONOMETRICS: Solutions PS4

Even if income is zero and no matter whether you live in an urban or rural area,
males tend to consume, on average, a 5% less fish than females. I couldn´t say
whether this result is realistic or not (…).

Independently of the value of income and gender, it seems individuals living in

urban areas consume, on average, a 14.3% more in fish than in rural areas. One
possible explanation of this effect could be higher transportation costs of fish
towards rural areas than towards urban areas.

b- Is the variable gender individually significant to explain the behavior of fish

expenditures (at 1% significance level)? Explain.

This is about performing an individual two tail t-test:

𝑯𝟎 : 𝜷𝟐 = 𝟎, 𝑯𝒂 : 𝜷𝟐 ≠ 𝟎

𝜷
𝒕𝒔 = 𝒔𝒆(𝜷𝟐 in our example: 𝒕𝒔 = −𝟏. 𝟑𝟕𝟖
𝟐)

𝒕𝒄 = −𝟐. 𝟕𝟏𝟗 two-tailed t-test with 36 degrees of freedom at 1%

We fail to reject the null hypothesis since −𝟏. 𝟑𝟕𝟖 ≤ −𝟐. 𝟕𝟏𝟗 at 1% significance level
and therefore, gender is individually insignificant independent factor. It is not
helping you to improve your understanding about fish expenditures. This is
consistent with no real explanation about why you should expect females consuming
more fish than males.

c- Is the model globally significant at 1% significance level? Explain.

𝑯𝟎 : 𝜷𝟏 = 𝜷𝟐 = 𝜷𝟑 = 𝟎 that is, none of the explanatory variables have predictive

power.

𝑯𝒂 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly explain the behaviour of the dependent variable.

𝑹𝟐 𝒏 𝒌 𝟏
𝑭𝒔 = 𝟏 𝑹𝟐
∗ 𝒌
in our example, 𝑭𝒔 = 𝟑𝟔

𝑭𝒄 = 𝟒. 𝟒𝟎 F-test with n=3 and d=36 degrees of freedom at 1%

We reject the null hypothesis since 𝟑𝟔 > 𝟒. 𝟒𝟎 at a 1% significance level and

therefore, the three explanatory variables are jointly statistically significant, helping
us to understand the behavior of fish expenditures. That is, this specification is a
significant regression model.

24
 ECONOMETRICS: Solutions PS4

20 A group of researchers in the field of environmental economics has conducted a

survey to investigate patterns of apples’ consumption, both regular and Eco labeled. The
sample contains the responses of 660 individuals. The following information is available:

Variable Description
regq Quantity demanded regular apples, lbs
ecoq Quantity demanded Eco labeled apples, lbs
regp Price of regular apples, pounds
ecop Price of Eco labeled apples, pounds
educ Years of schooling
age Age in years
hhsize Household size
faminc Family income, thousands
male =1 if the individual is a male

Three different models have been estimated using ecoq as dependent variable. The results
are presented next. (Standard errors in parenthesis)

Model 1 Model 2 Model 3

const 1.965 2.007 1.137
(0.380) (0.387) (0.911)
ecop -2.926 -2.962 -2.889
(0.588) (0.592) (0.596)
regp 3.029 3.063 3.034
(0.711) (0.714) (0.716)
male -0.126 -0.101
(0.221) (0.227)
educ 0.034
(0.045)
age 0.0008
(0.007)
faminc 0.002
(0.003)
hhsize 0.057
(0.069)
N 660 660 660
2
R 0.036 0.036 0.040
RSS 4051.05 4049.05 4033.81
Note: Set 𝛼 equal to 5% when needed.

25
 ECONOMETRICS: Solutions PS4

a- Interpret the coefficients on the price variables from Model 1 and comment on their
signs and magnitudes. Are regular apples and eco-labeled apples substitute goods?
b- Report the individual t-tests from Model 1. At the individual level, are the price
variables statistically significant?
c- Is there a gender difference in the quantity demand for eco-labeled apples? If so, is
the difference statistically significant? Justify your answer.
d- Compare the goodness of fit between Model 1 and Model 2.
e- Explain with your own words how would you extend Model 3 to allow a different
effect of education on apples’ consumption by gender.
f- Model 3 adds the variables faminc, hhsize, educ and age to the regression from part
(b). Test whether these four variables are jointly significant.

21 Gathering data for Michigan manufacturing firms in 2010, we obtain the following
estimation results using a log transformation:

OLS Estimation results

Dependent: log(training per employee)
Explanatory OLS t-statistic Degrees of Significance t-critical
Variable Coefficient Freedom level
Intercept 46.67 101 0.01 -
log(sales) 0.987 7.559 101 0.01 2.626
log(employees) -0.555 -10.378 101 0.01 -2.626
grant 0.125 3.111 101 0.01 2.626

Sample size (n) 105 F-statistic Degrees of Significance F-critical

Freedom level
R-squared 0.237 10.456 3(n),101(d) 0.01 3.98

Such that, the dependent variable is hours of training per employee, the variable sales
represents annual sales, employees is the number of employees and grant variable is a dummy
equals to one if the firm received a job training grant for 2010 and zero otherwise. Please,
answer the following three questions:

a- Interpret the above estimations results (only the value of the OLS coefficient for each
of the explanatory variables).

26
 ECONOMETRICS: Solutions PS4

Maintaining employees as a constant and no matter whether the firm received a

grant, increasing by 1% sales, training per employee will increase, on average,
0.987%. This is realistic since you can think that more sales means more profits and
therefore more opportunities to spend on training.

No matter whether the firm received a grant and holding sales as a constant, if you
increase employees by 1% percent, on average, training per employee will decrease
by 0.555%. This is realistic as you can expect increasing the size of the company
will produce less opportunities to be trained.

Even if we hold sales and employees as constants, a company receiving a grant will,
on average, dedicate more time to training than a company not receiving a grant by
around a 12.5% more. Again, this result is realistic because a grant is similar to a
subsidy.

b- Is the variable grant individually significant to explain the dependent variable (at 1%
significance level)? Explain.

This is about performing an individual two tail t-test:

𝑯𝟎 : 𝜷𝟑 = 𝟎, 𝑯𝒂 : 𝜷𝟑 ≠ 𝟎

𝜷
𝒕𝒔 = 𝒔𝒆(𝜷𝟑 in our example: 𝒕𝒔 = 𝟑. 𝟏𝟏𝟏
𝟑)

𝒕𝒄 = 𝟐. 𝟔𝟐𝟔 two-tailed t-test with 101 degrees of freedom at 1%

We reject the null hypothesis since 𝟑. 𝟏𝟏𝟏 ≥ 𝟐. 𝟔𝟐𝟔 at 1% significance level and
therefore, grant is individually significant independent factor. It is helping you to
improve your understanding about training per employee. Receiving a grant will
subsidy the expenditures for training and therefore being cheaper for the company
to incur on training.

c- Is the model globally significant at 1% significance level? Explain.

𝑯𝟎 : 𝜷𝟏 = 𝜷𝟐 = 𝜷𝟑 = 𝟎 that is, none of the explanatory variables have predictive

power.

𝑯𝒂 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly explain the behaviour of the dependent variable.

𝑹𝟐 𝒏 𝒌 𝟏
𝑭𝒔 = 𝟏 𝑹𝟐
∗ 𝒌
in our example, 𝑭𝒔 = 𝟏𝟎. 𝟒𝟓𝟔

27
 ECONOMETRICS: Solutions PS4

𝑭𝒄 = 𝟑. 𝟗𝟖 F-test with n=3 and d=101 degrees of freedom at 1%

We reject the null hypothesis since 𝟏𝟎. 𝟒𝟓𝟔 > 𝟑. 𝟗𝟖 at a 1% significance level and
therefore, the three explanatory variables are jointly statistically significant, helping
us to understand the behaviour of training. That is, this specification is a significant
regression model that you can use in order to learn something about the behavior of
training.

22 A group of researchers has conducted a survey that contains information on

smoking behavior and other variables for a random sample of 807 single adults from the
United States. The following information is available:

Variable Description
cigs Average number of cigarettes smoked per day
cigprice State cigarette price, cents per pack
educ Years of schooling
age Age in years
income Annual income, dollars
white =1 if the individual is white
restaurn =1 if state restaurant smoking restrictions

Three different models have been estimated using cigs as dependent variable. The results are
presented next. (Notes: Standard errors in parenthesis; l_ stands for natural logarithm)

Model 1 Model 2 Model 3

const -0.983 -3.765 -2.011
(8.685) (8.898) (8.964)
cigprice -0.048 -0.012 -0.005
(0.102) (0.103) (0.103)
l_income 1.429 1.433 1.891
(0.6793) (0.679) (0.713)
white 0.0009 -0.036
(1.483) (1.479)
restaurn -2.961 -2.949
(1.136) (1.134)
educ -0.377
28
 ECONOMETRICS: Solutions PS4

(0.168)
age -0.045
(0.029)
N 807 807 807
R2 0.005 0.013 0.021
RSS 151052.1 149770.7 148568.0
Note: Set 𝛼 equal to 5% when needed.

a- Interpret the coefficients on the variables from Model 1 and comment on their signs
and magnitudes. Is the income effect statistically significant?
b- Interpret the coefficient on the variable restaurn.
c- Is there a race difference in the quantity demanded for cigarettes? If so, is the
difference statistically significant? Justify your answer.
d- Explain with your own words how would you extend Model 3 to allow a different
effect of education on smoking habits by race.
e- Model 3 adds the variables age and educ to the regression from part (b). Test
whether these two variables are jointly significant.

29
 ECONOMETRICS: Solutions PS4

a- No matter the value of the two dummy variables, increasing by 1% inflation

rate would decrease, on average, economic growth by 0.23%.

This is about performing an overall F-test:

𝑯𝟎 : 𝜷𝟏 = 𝜷𝟐 = 𝜷𝟑 = 𝟎 that is, none of the explanatory variables have predictive

power.

𝑯𝒂 : The 3 explanatory variables introduced in the above regression model are

appropriate variables that jointly explain the behaviour of the dependent variable.

30
 ECONOMETRICS: Solutions PS4

𝑹𝟐 𝒏 𝒌 𝟏
𝑭𝒔 = 𝟏 𝑹𝟐
∗ 𝒌
in our example, 𝑭𝒔 = 𝟒. 𝟖𝟒
𝑭𝒄 = 𝟒. 𝟖𝟐 F-test with n=3 and d=96 degrees of freedom at 1%

We reject the null hypothesis since 𝟒. 𝟖𝟒 > 𝟒. 𝟖𝟐 at a 1% significance level and

therefore, the three explanatory variables are jointly statistically significant, helping
us to understand the behaviour of economic growth in a country. That is, this
specification is a significant regression model that you can use in order to learn
something about country level economic growth.

b- If the country is not landlocked and has suffered a civil war, its economic
growth will be, on average, 1.35% lower than a country being not landlocked
and not suffering a civil war in the past ten years, holding inflation rate as a
constant. It is a realistic effect as you should expect suffering a war may
damage the economy of a country

This is about performing an individual two tail t-test:

𝑯𝟎 : 𝜷𝟐 = 𝟎, 𝑯𝒂 : 𝜷𝟐 ≠ 𝟎

𝜷 𝟐
𝒕𝒔 = 𝒔𝒆(𝟐) in our example: 𝒕𝒔 = −𝟔. 𝟒𝟐

𝒕𝒄 = −𝟐. 𝟔𝟐𝟔 two-tailed t-test with 96 degrees of freedom at 1%

We reject the null hypothesis since −𝟐. 𝟔𝟐𝟔 ≥ −𝟔. 𝟒𝟐 at 1% significance level and
therefore, therefore, civil war is an individually significant independent factor. It is
helping you to improve your understanding about economic growth. It can be argued
that this is a realistic effect as one should expect a civil war affecting in a negative
and significant way the economic growth of an economy.

c- A different strategy could be to perform a Chow test about whether a civil war
is a significant structural break changing the entire structure of the model.
That is, we would run three different models with the same specification: (1)
taking into account all the countries no matter whether they have suffered a
civil war, (2) a model taking into account only those countries suffering a civil
war and (3) a model accounting only for those countries not suffering a civil
war in the past ten years. Once the above models are estimated, we could
perform the Chow test and see whether a civil war has a significant effect on
economic growth such that the null hypothesis would be one single regression
model is more efficient (structural break insignificant) and the alternative, two
regression models more efficient (structural break significant).
31