Financial Econometrics
Exam 2. June 9, 2023
[See the Questions sheet for modified words]
Name______________________________ Student ID________________________
1. You can bring one A4 formula sheet.
2. You can bring a calculator.
3. Please submit both your questions and answer sheets.
Chapter 8 (25%)
1. (10%) In a regression model, if heteroskedastic problem exists in the data. We introduce (i) the
ordinary least squares estimates, (ii) the White variance estimates, (iii) the genearlized least
squares (GLS) estimates feasible, and (iv) the feasible generalized least squares (FGLS)
estimates.
(a) (5%) Which estimates prodice the same parameter estimates as the OLS?
(b) (5%) Which estimates are most likely to produce the least stadand errrors?
Ans: (a) (ii) Write variance estimates
(b) (iv) FGLS
2. (5%)Recall the Lagrange multiplier test or Breusch-Pagan test for heteroskedasticity. Consider the general
multiple regression model:
(1) yi = β1 + β2 xi2 + ⋯ + βK xiK + ei .
A general form for the variance function is
(2) var (yi | zi ) = σi2 = E(ei2 | zi ) = exp(α1 + α2 zi2 + ⋯ + αS ZiS ),
Expresss H0: homoskedasticity vs H1: heteroskedasticity using model parameters in (1) and (2)
Ans: H0 : α2 = α3 = ⋯ = αS = 0 vs H1 : not all αi in H0 are all zeros.
3. (10%) Suppose a simple regression model yi = β0 + β1 xi + ei where the errors are statistically
independent from one another (E(ei | xi )) and that heteroskedasticity is present in the form ,
2 2
var (ei | xi ) = σ h(xi ) = σ xi, xi > 0. Derive the transformed model into one with homoscedastic errors.
yi 1 x e
= β0 + β1 i + i
xi xi xi xi xi
Ans: 1. Divide original model by :
2. Define the transformed variables and transformed errors:
y 1 x e
yi* = i ; xi*0 = ; xi*1 = i ; ei* = i
xi xi xi xi
* * *
3. New model is yi = β 0 xi 0 + β1 xi1 + ei
⎛ e ⎞ 1
var (ei* | xi )= var ⎜ i | xi ⎟ = σ 2 xi = σ 2
⎜ x ⎟ xi
. Errors are not homoscedastic: ⎝ i ⎠
Level: Hard – Evaluation
AACSB: Analytic
Chapter 10 (25%)
4. (10%) Consider the following regression:
(3) yi = β1 + β2 xi + ei
Suppose both xi and yi are random, and that endogenity exists, i.e., cov(x, e) ≠ 0.
Recall that β2 can be expressed as
cov(x, y) cov(x, e)
β2 = − .
var (x) var (x)
Recall that the LSE for β2 is
∑ (xi − x̄)(yi − ȳ) ̂
cov(x, y)
b2 = =
∑ (xi − x̄) 2 ̂
var(x)
(a) (5%) What does this LSE converge to in probability?
(b) (5%) Is the LSE consistent or not? Why or why not?
Ans:
cov(x, e)
(a) β2 +
var (x)
(b) No. Because it does not converge to β2 in probaiblity/
5. (15%) Assume that both xi and yi are random variables and xi and ei are uncorrleated. Let zi
denote a good instrumental variable (IV).
(a) (5%) What are good properties for a good instrumental variable?
(b) (5%) Provide two moment conditions with the IV zi, that allows us to derive consistent
moments (MM) estimators for the parameters β1 and β2.
(c) (5%) Following (b), Derive the metheod of moments (MM) estimators for the regression
model
Ans
(a)
(b) and (c)
Chapter 11 (25%)
1. (5%) What does it mean by
saying that a model or an equation
is identified?
Ans: When the estimation of an equation’s parameter is possible, then the equation is said to
be identified, and its parameters can be estimated consistently.
2. (20%) Consider the Fulton’s fish market example. The demand equation for this market as:
ln(QUA Nt ) = α1 + α2 ln(PR ICEt ) + α3 MONt + α4T UEt
+α5WEDt + α6T HUt + etd ,
where QUA Nt is the quantity sold in pounds, PR ICEt is the average daily price per pound, and α2
is the price elasticity of the demand.
The supply equation is
ln(QUA Nt ) = β1 + β2 ln(PR ICEt ) + β3STOR M Yt + est ,
where β2 is the price elasticity of supply.
The demand equation shifts daily, while the supply remains fixed, since the supply equation does
not contain the dily indicator variable.
In this example, the number of equations is M=2. We have endogenous variablesL ln(PRICE) and
ln(QUAN). We have exogenous variables: MON, TUE, WED, THU, and STORMY
(a) (10%) Are the demand and supply model identified? Check the necessary condition for
these two equations separately, and explain if the they are identified.
(b) (10%) Ouputs of the reduced-form equations (or, the first-stage equations) are given in
Table 11.4a and 11.4 b. Are the demand and supply equations identified? Why or why not?
Ans: (a) In a system of M simultaneous equations, which jointly determine the values of M
endogenous variables, at least (M − 1) variables must be absent from an equation for estimation for
its parameters to be possible.
(b)
Chapter 15 (25%)
1. (12%) In a panel data model,
yit = β0 + β1x1it + α w1i + ui + eit
Assume T = 2 and N = 2. Let us focus on the respons variable yit. Suppose it has the value provide
the following transformation for the difference transformation Δyi, the within transformation ỹit⋅,
and the random effects transformation y* it
(set α = 0.5) for the response variable yit
i T yit ȳ⋅t Δyi ỹit⋅ y*
it
1 1 2.4 (e) (i)
2 1.6 (a) (c) (f) (j)
2 1 3.5 (g) (k)
2 0.5 (b) (d) (h) (l)
Ans
i T yit ȳ⋅t Δyi ỹit⋅ y*
it
1 1 2.4 0.4 1.4
2 1.6 2 0.8 -0.4 0.6
2 1 3.5 1.5 2.5
2 0.5 2 3.0 -1.5 -0.5
2. (5%) Consider the general model (15.16) with K = 4 and M = 2,
(4) yit = β1 + β2 x2it + β3 x3it + β4 x4it + α1w1i + α2 w2u + ui + eit .
Consider
(5) ui = γ1 + γ2 x̄2i⋅ + γ3 x̄3i⋅ + γ4 x̄4i⋅ + ci .
The Mundlack approach test endogeniety for regression by consider (4) and express H0: no
endogeneity using an equivalent expression with parameters in (4) or (5)
Ans: H0 : γ2 = γ3 = γ4 = 0
5. (5%) In this question, we are interested in testing random effects. Recall the random effects
model: We write the model by
yit = β1 + β2 x2it + α1w1t + (ui + eit )
= (β1 + ui ) + β2 x2it + α1w1t + eit
= (β1i ) + β2 x2it + α1w1t + eit .
Here, are the random effects, ui ∼ N(0,σu2 ) and eit ∼ N(0,σe2 ).
Express
{H1 : Random effects exist
H0 : No random effects v.s.
using model parameters.
Ans: H0 : σu2 = 0 v.s. H1 : σu2 > 0.
6. (3%) The fixed effects model derived yit = β0 + β1 x1it + α w1i + ui + eit is
yit = β1 + β1x1it + eit. Why do we need to remove β0, w1i, ui from the original model?
Ans: To avoid exact collinearity.
