Econometrics for Finance Handout Set II

ECONOMETRICS FOR FINANCE

MSc in Finance, UC3M
Handout Set II, Course 2022-2023

Total Score: 100 points.

Due: December 30th, 23:59.
Work in established groups.

Please read the following instructions before starting

1. This assignment has two exercises to be solved with Eviews. There are
hyperlinks (blue boxes, accessible if you open the PDF with PDF Reader)
that will help you to clarify concepts or direct you to useful information in
each of these exercises. Unless otherwise indicated, conduct inference at the
5% significance level against a two-sided alternative.
2. For each exercise, you are expected to make an Eviews file file named
Exercisenum.wf1 with all the operations involved (e.g., Exercise1.wf1).
Your responses must be included in a single Word/PDF file named Sum-
mary.doc. In this file, you can include detailed answers, comments, figures,
tables, and any other additional element you may consider important to en-
rich your responses.
3. At the beginning of the file Summary.doc, you must state the full names
of the students involved (format: surname(s), name), sorted in alphabetical
order by the first surname. The first student on this list is responsible to
upload a zipped file on the course’s website. This file must be named Hand-
outSet2 Surname.zip, with surname being the submitter’s surname (e.g.,
HandoutSet1 Smith.zip).
4. You are expected to submit the zipped file before the deadline expires.
You may still submit later on, but in this case a cumulative penalty of 10%
(first author responsible of submission) and 5% (remaining team members)
on the final mark will be applied for every hour, or fraction of hour, overdue.

The exactitude of the answers and the quality and clarity in the
resolution of the exercises will be taken as major references for
grading.

-1-
Econometrics for Finance Handout Set II

Exercise I. Fixed Income. Total: 55 points.

The yield to maturity (YTM) of a bond is the constant annual interest

rate that equals the present value of the bond’s expected cash flows to its
market value, i.e., it is the internal rate of return of the bond. It is
a measure of the (expected) profitability implied in the bond’s purchase.
Bonds at different maturities typically exhibit increasing YTMs. The plot
of the YTM of a sequence of zero-coupon bonds with different maturities
issued by a default-free institution against the maturities involved is called
the yield curve. The following table reports YTMs (in percentage) of the
U.S. bonds at different maturities (in years) ranging from 1 month (1/12
year) to 30 years, recorded on 1/12/2022:

Maturity (years) YTM (%)

1/12 4.04
2/12 4.24
3/12 4.33
4/12 4.52
6/12 4.65
1 4.66
2 4.25
3 3.98
5 3.68
7 3.62
10 3.53
20 3.85
30 3.64
Source: U.S. Department of the Treasury.
A problem when generally representing the yield curve is that YTMs are
available at certain maturities only. In the U.S., for example, the Treasury
reports YTMs at the 3- and 5-year maturities, but not, for instance, at the
4-year maturity. The most popular method to construct general estimates of
yield curve is the so-called Nelson-Siegel (NS) model. This model builds
on the following (non-linear) equation:
1 − exp(−λn) 1 − exp(−λn)
" # " #
y(n) = β1 + β2 + β3 − exp(−λn) + ε(n)
λn λn

-2-
Econometrics for Finance Handout Set II

where y(n) is the YTM of a bond with maturity in n years, (β1 , β2 , β3 , λ)0
is a vector of unknown parameters to be estimated, and ε(n) is an error
term obeying usual assumptions. The NS equation is non-linear on the
λ parameter and, therefore, cannot be estimated by means of OLS. If λ
were known, the remaining parameters (β1 , β2 , β3 )0 could be estimated by
OLS because the equation is linear on these parameters.
h iIn this case, we
could simply define the “regressors” x2 (n, λ) = 1−exp(−λn)
λn
and x3 (n, λ) =
h i
− exp(−λn) given the value of λ, and then regress y(n) on
1−exp(−λn)
λn
a constant and x2 (n, λ) and x3 (n, λ) using OLS. We will implement this
approach in this exercise.

a) [5 points] Create a new workfile in Eviews and load the maturities

and the corresponding yields reported in the previous table in Eviews.
Name these variables n and YTM, respectively. Alternatively, create
the variables assigning the values directly.

b) [5 points] Given the preliminary estimate λ b = 0.15, construct the

variables x2 (n, λ) and x3 (n, λ). To this end, define a scalar variable in
b b
Eviews named lambda with the suggested value, and then generate the
variables x2 (n, λ)
b and x (n, λ),
3
b naming them x2 and x3, respectively.
Report the correlation matrix of x2 and x3.

c) [5 points] Given λ,b estimate the remaining NS parameters (β , β , β )0

1 2 3
from an OLS regression of YTM on a constant, x2 and x3. Name the
equation as NS and report regression statistics. How much variability
of the yield curve can be explained with the Nelson-Siegel factors? Are
the coefficients of this regression (individually) significant?

d) [5 points] Report 95% confidence intervals for each of the estimated

parameters. Use the corresponding interval to test each of the following
hypotheses individually: d.1) H0 : β1 = 5; d.2) H0 : β2 = 2; d.3)
H0 : β3 = −1.

e) [5 points] Plot the YTM given the actual observations and the NS
estimated values.

f) [5 points] Test the individual hypothesis H0 : β1 = 4 against a one-

sided upper alternative at the 99% confidence level using a test of
significance.

-3-
Econometrics for Finance Handout Set II

g) [5 points] Test the composite hypothesis H0 : β2 = 0, β3 = 0 using a

Wald test.

h) [5 points] Test the composite hypothesis H0 : β2 = β3 = 0 against

a two-sided alternative and a 10% significance level with a χ test,
estimating the restricted and unrestricted models.

i) [5 points] We may alternatively try to characterize the Yield Curve

using polynomial fitting. Using OLS, estimate the cubic polynomial
model y(n) = α + δ1 n + δ2 n2 + δ3 n3 + ε(n), with (α, δ1 , δ2 , δ3 )0 denoting
the unknown parameters and ε(n) the error term of the model. Name
this equation as cubic. To compare the suitability of the cubic model
vs the NS model, use the Akaike Information Criterion (AIC). This
criterion is valid to compare non-nested models (i.e., models that do
not share a common functional form). The model with the smaller AIC
provides a better fitting on the data. Which model accommodates the
data more parsimoniously?

j) [5 points] Report a 80% confidence interval for the parameter δ2 in

the previous model and use it to test the null hypothesis H0 : δ2 = 0
against a two-sided alternative.
Hint: To determine the correspondig critical value use the Eviews
command @qnorm(p), intended to compute the inverse CDF of a
standard normal distribution at the probability p.

In practice, one of the main purposes of yield-curve fitting is to use the

resultant values to determine the theoretical price or fair value of any fixed-
income security. In particular, we use the yield curve of government bonds
to discount the expected cash flows of other securities. The theoretical value
is the present value of the discounted cash flows. Suppose a bond that pays
off a fixed annual coupon C once a year until maturity, and redeems a face
value N after n years. Then, the theoretical price of this bond, PB , would
be:
C C C C +N
PB = + 2 + 3 + ... +
(1 + Y T M1 ) (1 + Y T M2 ) (1 + Y T M3 ) (1 + Y T Mn )n

where Y T Mi denotes the YTM of the yield curve at the i-year maturity.
Since YTM represents the required rate of return under risk-free conditions,
corporate bonds typically add some risk premium γ > 0 in the valuation (the

-4-
Econometrics for Finance Handout Set II

greater the probability of default, the greater δ). Let Y T Mi∗ = Y T Mi + γ

denote the risk-adjusted discounting rate. Then, the bond’s theoretical value
is given by:
C C C +N
PB∗ = + 2 + ... +
(1 + Y T M1 ) (1 + Y T M2 )
∗ ∗ (1 + Y T Mn∗ )n

k) [5 points] The Accounting Department of a bank needs to determine

the fair value of certain bonds at the end of December 2022. These
bonds were issued by a private U.S. company 6 year ago. Each bond
has face value $6,000, pay a single annual coupon of $40 at the end of
each year, and will be redeemed at the end of 2039. Using the YTM
curve implied by the Nelson-Siegel model in the previous exercise and
a default risk premium γ = 1.5%, determine the fair value of a single
bond at the end of 2022, i.e., compute:
40 40 6040
PB∗ = + 2 + ... +
(1 + Y T M1 ) (1 + Y T M2 )
∗ ∗
(1 + Y T M10 )
∗ 17

Technical note The four parameters of the Nelson-Siegel model can

be estimated simultaneously using non-linear least squares (NLLS), an
estimation procedure beyond the scope of this course and for that reason
not required in this exercise. Given some initial values of the parameters
(β1 , β2 , β3 , λ)0 , we can readily obtain the residuals:

1 − exp(−λn) 1 − exp(−λn)
" # " #
ε(n) = y(n) − β1 − β2 − β3 − exp(−λn)
λn λn

Then, NLLS estimates parameter by minimizing the sum of squared residuals

ε(n)2 across maturities, i.e., the same objective function as in the
P

standard LS setting. However, the non-linear dependence on λ prevents the

optimization problem from admitting a close-form solution. As a result, the
NLLS estimator must be determined using numerical optimization building
on iterative algorithms.

-5-
Econometrics for Finance Handout Set II

Exercise II. Banking. Total: 45 points.

The file BankPanelData.xls contains an unbalanced panel with quarterly

data on 930 publicly-traded banks over the period 2001Q1-2017Q4. Firm-
specific data refer to: (1) bank’s profitability (or management’s quality), as
measured by ROA (ROA) ratio; (2) risk exposures in the real estate market
(REEXP), defined as the ratio of real estate loans to total loans; (3) book to
market (BTM ) ratio, a measure of growing opportunities; (4) Tier 1 capital
(TIER1 ), a measure of solvency; (5) size, measured by the natural logarithm
of total assets (SIZE); (6) systemic interconnectedness, as measured by the
ratio of short-term wholesale funding (short-term funding obtained in the
interbank market) to total assets (STWSF ); (7) stock market performance,
as measured by quarterly stock returns (RET ); (8) idiosyncratic volatility
(IV ) and (9) idiosyncratic skewness (IS ), defined as the sample standard
deviation and sample skewness of the residuals of the 3-factor Fama-French
regression model estimated with daily data on each quarter, respectively.
Market-wide environmental variables are captured by (10) unemployment
rate (UNEMP) and (11) real GDP growth (RGDPG) in the state in which
the banks is headquartered. In this exercise, we will analyze if the ROA ratio
can be predicted on the basis of the remaining variables.

a) [6 points] Import the data, proceeding as follows: (1) Select all the
data columns in the Excel file, copy, and Paste as new file into Eviews.
The Clipboard Read window pops up. Then, (2) in Step 1 of this
window, leave empty the field “Text Representing NA” and click OK.
Next, in Step 2, choose Dated Panel in the Basic structure menu, set
the variable ENTITY in Cross Section ID series and DATE in Date
Series, and click on the Finish icon. Finally, reply “No” to the last
question Link imported series and alpha object(s) to external source?.
The data (unbalanced panel with 31,646 observations) will be imported
correctly into Eviews.

b) [6 points] Estimate the following model with pooled OLS and ordinary
standard errors:

ROAit = α + β1 ST W SFit−1 + β2 SIZEit−1 + β3 T IER1it−1 + β4 RETit−1

+β5 IVit−1 + β6 ISit−1 + β7 BT Mit−1 + β8 REEXPit−1
+β9 U N EM Pit−1 + β10 RGDP Git−1 + uit

-6-
Econometrics for Finance Handout Set II

aiming to predict future values of ROA given lagged values of the

remaining variables.
Hint: You can make Eviews lag variables by adding (-1) to the variables
in the Specification window of the Equation Estimation Window, e.g.,
specifying Y C X(-1) will run the model Yt = α + βXt−1 + ut .

c) [5 points] Suppose that someone claims that a bank with a business

model characterized by a greater concentration in real estate lending
will be more profitable than a bank diversified in other activities. Given
the estimates of the previous model and the sample available, will you
agree with that view? Justify formally your answer.

d) [5 points] Firms with low values of the BTM ratio are considered
“growth”-type companies (i.e., having growth opportunities), whereas
firms with large BTM ratios are “value”-type firms. For banks with
value profile, everything else constant, will we expect a lower or a
greater ROA in the next quarter? Justify formally your answer.

e) [5 points] Suppose that some unexpected bad news specific to a given

bank is released this quarter, and that the consequence of this news a
massive fall in the bank’s stock price. Which variables (yes, plural) in
the model would reflect this effect? Given the corresponding coefficient
estimates, which effect should we expect on next quarter’s ROA?

f) [6 points] Test the joint restriction that all parameters, except the
intercept, are zero using a Wald test.

g) [6 points] Estimate the model with individual fixed effects and

ordinary standard errors and test the suitability of the fixed effect
estimator against the simplest pooled OLS.

h) [6 points] Report the estimates of a model with individual fixed effects

and robust standard errors against arbitrary heteroskedasticity.

-7-