ECON 482 / WH Hong Answer Key

Answer Key: Problem Set 2

1. Consider the standard simple regression model y = β 0 + β1 x + u under the Gauss-

Markov Assumptions SLR. 1 through SLR. 5. The usual OLS estimators βˆ0 and βˆ1

are unbiased for their respective population parameters. Let β1 be the estimator of β1

obtained by assuming the intercept is zero.

n

∑x y i i
i. Verify that β1 = i =1
n
.
∑x i =1
2
i

∑ u = ∑ ( y − β x )
n n
β1 is obtained by minimizing the loss function,
2
2
[Hint: i i 1 i ]
i =1 i =1

(Ans)

( ) ( )
n n
Q β1 = min ∑ u 2 = ∑ yi − β1 xi
2

{β1} i =1 i i =1

( )
∂Q β1
( )
n n n

∂β
= −2∑ yi − β1 xi xi = 0
i =1
=> ∑ y x − β ∑ x
i =1
i i 1
i =1
2
i =0
1

∑x y i i
Thus, β1 = i =1
n

∑x
i =1
2
i

ii. Find β1 in terms of the x , β 0 and β1 . Verify that β1 is unbiased for β1 when

the population intercept ( β 0 ) is zero. Are there other cases where β1 is unbiased?
(Ans)
Plugging in yi = β0 + β1xi + ui gives

⎛ n ⎞ ⎛ n 2⎞
β1 = ⎜ ∑ xi ( β 0 + β1 xi + ui ) ⎟ / ⎜ ∑ xi ⎟ .

⎝ i =1 ⎠ ⎝ i =1 ⎠

After standard algebra, the numerator can be written as

n n n
β 0 ∑ xi +β1 ∑ x 2 + ∑ xi ui .
i
i =1 i =1 i =1

1
ECON 482 / WH Hong Answer Key

Putting this over the denominator shows we can write β1 as

⎛ n ⎞ ⎛ n 2⎞ ⎛ n ⎞ ⎛ n 2⎞
β
1 = β0 ⎜ ∑ xi ⎟ / ⎜ ∑ xi ⎟ + β1 + ⎜ ∑ xi ui ⎟ / ⎜ ∑ xi ⎟ .
⎝ i =1 ⎠ ⎝ i =1 ⎠ ⎝ i =1 ⎠ ⎝ i =1 ⎠

Conditional on the xi, we have

⎛ n ⎞ ⎛ n ⎞
E( β1 ) = β0 ⎜ ∑ xi ⎟ / ⎜ ∑ xi2 ⎟ + β1
⎝ i =1 ⎠ ⎝ i =1 ⎠
because E(ui) = 0 for all i. Therefore, the bias in β1 is given by the first term in

this equation. This bias is obviously zero when β0= 0. It is also zero when
n

∑ x = 0, which is the same as

i =1
i x = 0. In the latter case, regression through the

origin is identical to regression with an intercept.

iii. Find the variance of β1 . [Hint: The variance does not depend on β 0 ]
(Ans)
From the last expression for β1 in part ii, we have that, conditional on the xi,
−2 −2
⎛ n ⎞ ⎛ n ⎞ ⎛ n ⎞ ⎛ n ⎞
Var( β1 ) = ⎜ ∑ xi2 ⎟ Var ⎜ ∑ xi ui ⎟ = ⎜ ∑ xi2 ⎟ ⎜ ∑ xi2 Var(ui ) ⎟
⎝ i =1 ⎠ ⎝ i =1 ⎠ ⎝ i =1 ⎠ ⎝ i =1 ⎠
−2
⎛ n ⎞ ⎛ n
⎞ ⎛ n ⎞
= ⎜ ∑ xi2 ⎟ ⎜ σ 2 ∑ xi2 ⎟ = σ 2 / ⎜ ∑ xi2 ⎟ .
⎝ i =1 ⎠ ⎝ i =1 ⎠ ⎝ i =1 ⎠

iv. ( )
Show that var β1 ≤ var βˆ1 . ( )
n n

∑ xi2 ≥ ∑ ( xi − x ) , with strict inequality unless

2
[Hint: For any sample of data
i =1 i =1

x =0]
(Ans)
⎛ n ⎞
We know that Var( β̂1 ) = σ 2 / ⎜ ∑ ( xi − x ) 2 ⎟ .
⎝ i =1 ⎠
n n
From the hint, ∑ xi2 ≥
i =1
∑ (x − x )
i =1
i
2
, and so Var( β1 ) ≤ Var( β̂1 ). A more direct way to
n n n
see this is to write ∑ ( xi − x ) = 2
∑ x − n( x ) , which is less than
2
i
2
∑x 2
i unless
i =1 i =1 i =1

x = 0.

2
ECON 482 / WH Hong Answer Key

v. Comment on the tradeoff between bias and variance when choosing between β1 and

β̂1 .
(Ans)
For a given sample size, the bias in β1 increases as x increases (holding the sum

of the xi2 fixed). But as x increases, the variance of β̂1 increases relative to

Var( β1 ). The bias in β1 is also small when β 0 is small. Therefore, whether we

prefer β1 or β̂1 on a mean squared error basis depends on the sizes of β 0 , x ,
n
and n (in addition to the size of ∑x
i =1
2
i ).

2.
i. Let βˆ0 and β̂1 be the intercept and slope from the regression of yi on xi ., using

n observations. let c1 and c2 , with c2 ≠ 0 , be constants. Let β0 and β1 be the

c
intercept and slope from the regression of c1 yi on c2 xi . Show that β1 = 1 βˆ1 and
c2

β0 = c1βˆ0 .
(Ans)
Note that c1 y = c1 y (the sample average of c1 yi is c1 times the sample average of
yi) and c2 x = c2 x . When we regress c1yi on c2xi (including an intercept) we obtain
the slope:
n n

∑ (c x − c x)(c y − c y ) ∑ c c ( x − x )( y − y )
2 i 2 1 i 1 1 2 i i
β1 = i =1
n
= i =1
n

∑ (c x − c x )
i =1
2 i 2
2
∑ c (x − x )
i =1
2
2 i
2

c ∑
( xi − x )( yi − y )
c1 ˆ
= 1 ⋅ i =1 n
= β1.
∑ (x − x )
c2 2 c2
i
i =1
we obtain the intercept as
β0 = (c1 y )– β1 (c2 x )= (c1 y )– [(c1/c2) β̂1 ](c2 x )= c1( y – β̂1 x )= c1 β̂ 0
because the intercept from regressing yi on xi is ( y – β̂1 x ).

3
ECON 482 / WH Hong Answer Key

ii. Now, let β0 and β1 be from the regression of ( c1 + yi ) on ( c2 + xi ) (with no

restriction on c1 or c2 ). Show that β1 = βˆ1 and β0 = βˆ0 + c1 − c2 βˆ1 .

(Ans)
We use the same approach from part (i) along with the fact that (c1 + y ) = c1+ y

and (c2 + x) = c2+ x . Therefore, (c1 + yi ) − (c1 + y) = (c1 + yi)– (c1+ y )= yi–

y and (c2+ xi)– (c2 + x) = xi– x . So c1 and c2 entirely drop out of the slope

formula for the regression of (c1+ yi) on (c2+ xi), and β1 = β̂1 . The intercept is

β0 = (c1 + y ) – β1 (c2 + x) = (c1 + y ) – β̂1 (c2 + x ) = ( y − βˆ1 x ) + c1 – c2 β̂1 =

β̂ 0 + c1 – c2 β̂1 , which is what we wanted to show.

iii. Now, let βˆ0 and βˆ1 be the OLS estimates from the regression log ( yi ) on xi ,

where we must assume yi > 0 for all i . For c1 > 0 , let β0 and β1 be the

intercept and slope from the regression of log ( c1 yi ) on xi . Show that β1 = βˆ1 and

β0 = log ( c1 ) + βˆ0 .

(Ans)
We can simply apply part (ii) because log(c1 yi ) = log(c1 ) + log( yi ) . In other words,
replace c1 with log(c1), yi with log(yi), and set c2 = 0.

iv. Now, assuming that xi > 0 for all i , let β0 and β1 be the intercept and slope

from the regression of yi on log ( c2 xi ) . How do β0 and β1 compare with the

intercept and slope from the regression of yi on log ( xi ) ?

(Ans)
Again, we can apply part (ii) with c1 = 0 and replacing c2 with log(c2) and xi with
log(xi). If βˆ0 and βˆ1 are the original intercept and slope, then β1 = βˆ1 and

β = βˆ0 − log ( c2 ) βˆ1 .

4
ECON 482 / WH Hong Answer Key

3. Let βˆ0 and βˆ1 be the OLS intercept and slope estimators, respectively, and u be the
sample average of the errors (not the residuals!)
n
di
i. Show that βˆ1 can be written as βˆ1 = β1 + ∑ wi ui where wi = and
i =1 SSTx

di = xi − x .
(Ans)
n

∑( x − x )u
i i
Note that βˆ1 = β1 + i =1
n
. Then, now, define that wi = di / SSTx .
∑( x − x )
2
i
i =1

n
ii. Using part i, along with ∑w
i =1
i = 0 , to show that βˆ1 and u are uncorrelated.

[Hint: You are being asked to show that E ⎡ βˆ1 − β1 u ⎤ = 0 .]

⎣ ⎦ ( )
(Ans)
Because Cov( βˆ1 , u ) = E[( βˆ1 − β1 )u ] , we show that the latter is zero. But, from

part (i), E[( βˆ1 − β1 )u ] =E ⎡⎢

⎣ (∑ n
i =1 )
wi ui u ⎤⎥ = ∑ i =1 wi E(ui u ).
⎦
n
Because the ui are

pairwise uncorrelated (they are independent), E(ui u ) = E(ui2 / n) = σ 2 / n (because

∑ wi E(ui u ) = ∑ i =1 wi (σ 2 / n) = (σ 2 / n)∑ i =1 wi = 0.
n n n
E(ui uh ) = 0, i ≠ h ). Therefore, i =1

iii. Show that βˆ0 can be witten as βˆ0 = β 0 + u − βˆ1 − β1 x . ( )

(Ans)
The formula for the OLS intercept is βˆ0 = y − βˆ x and, plugging in y = β 0 + β1 x + u

gives βˆ0 = ( β 0 + β1 x + u ) − βˆ1 x = β 0 + u − ( βˆ1 − β1 ) x .

σ 2 σ (x)
2

( )
2
iv. Use part ii and iii to show that var βˆ0 = + .
n SSTx
(Ans)
Because βˆ1 and u are uncorrelated,

5
ECON 482 / WH Hong Answer Key

Var( βˆ0 ) = Var(u ) + Var( βˆ1 ) x 2 = σ 2 / n + (σ 2 / SSTx ) x 2 = σ 2 / n + σ 2 x 2 / SSTx ,

which is what we wanted to show.

n
σ 2 n −1 ∑ xi2
v. Do the algebra to simplify the expression in part iv to var βˆ0 = ( ) n
i =1
.
∑( x − x )
2
i
i =1

n
SSTx
= n −1 ∑ ( xi − x ) ]
2
[Hint:
n i =1

(Ans)
Using the hint and substitution gives Var( βˆ0 ) = σ 2 [( SSTx / n ) + x 2 ] / SSTx

( ) (
= σ 2 ⎡⎢ n −1 ∑ i =1 xi2 − x 2 + x 2 ⎤⎥ / SSTx = σ 2 n −1 ∑ i =1 xi2 / SSTx .
⎣
n

⎦
n
)
4. Consider the following problem: running a regression and only estimating an intercept
n
Given a sample { yi : i = 1,.., n.} , let β0 be the solution to min ∑ ( yi − b0 ) . Show
2
i.
{b0 } i =1

that β0 = y , that is, the sample average minimizes the sum of squared residuals.
[Hint: You may use one-variable calculus or you can show the result directly by
adding and subtracting y inside the squared residual and then doing a little algebra.]
(Ans)
I will show the result without using calculus. Let be the sample average of the
and write
n n

∑ ( yi − b0 )2 = ∑ [( yi − y ) + ( y − b0 )]2
i =1 i =1
n n n
= ∑ ( yi − y ) 2 + 2∑ ( yi − y )( y − b0 ) + ∑ ( y − b0 ) 2
i =1 i =1 i =1
n n
= ∑ ( yi − y ) 2 + 2( y − b0 )∑ ( yi − y ) + n( y − b0 ) 2
i =1 i =1
n
= ∑ ( yi − y ) 2 + n( y − b0 ) 2
i =1

n
where we use the fact (see Appendix A) that ∑( y − y) = 0
i =1
i always. The first term

6
ECON 482 / WH Hong Answer Key

does not depend on b0 and the second term, n( y − b0 ) 2 , which is nonnegative, is

clearly minimized when b0 = y .

ii. Define residuals u = yi − y . Argue that these residuals always sum to zero.
(Ans)
n n
If we define ui = yi − y then ∑ ui = ∑ ( yi − y ) and we already used the fact that this
i =1 i =1

sum is zero in the proof in part (i).

Computer Exercises
5. The data set in CEOSAL2.dta contains information on chief executive officers for U.S.
corporations. The variable salary is annual compensation, in thousands of dollars, and
ceoten is prior number of years as company CEO.
i. Find the average salary and the average tenure in the sample.
(Ans)
Average salary is about 865.864, which means $865,864 because salary is in
thousands of dollars. Average ceoten is about 7.95.

ii. How many CEOs are in their first year as CEO (that is, ceoten = 0 )? What is the
longest tenure as a CEO?
(Ans)
There are five CEOs with ceoten = 0. The longest tenure is 37 years.

iii. Estimate the simple regression model

log ( salary ) = β 0 + β1ceoten + u ,

And report your results in the usual form. What is the (approximate) predicted
percentage increase in salary given one more year as a CEO?
(Ans)
The estimated equation is
n
log( salary) = 6.51 + .0097 ceoten

n = 177, R2 = .013.

7
ECON 482 / WH Hong Answer Key

We obtain the approximate percentage change in salary given Δceoten = 1 by

multiplying the coefficient on ceoten by 100, 100(.0097) = .97%. Therefore, one
more year as CEO is predicted to increase salary by almost 1%.

6. For the population of firms in the chemical industry, let rd denote annual expenditures
on research and development, and let sales denote annual sales (both are in millions of
dollars.)
i. Write down a model (not an estimated equation) that implies a constant elasticity
between rd and sales . Which parameter is the elasticity?
(Ans)
The constant elasticity model is a log-log model:
log(rd) = β 0 + β1 log(sales) + u,

where β1 is the elasticity of rd with respect to sales.

ii. Now, estimate the model using the data in RDCHEM.dta. Write out the estimated
equation in the usual form. What is the estimated elasticity of rd with respect to
sales ? Explain in words what this elasticity means.
(Ans)
The estimated equation is
n
log( rd ) = –4.105 + 1.076 log(sales)

n = 32, R2 = .910.

The estimated elasticity of rd with respect to sales is 1.076, which is just above one.
A one percent increase in sales is estimated to increase rd by about 1.08%.