You are on page 1of 3

Exam Advanced Time Series. Part II.

June 25th 2021

We want to develop a coincident indicator for weekly activity in the Eu-


roarea. There are three weekly variables, but we understand that they can be
very volatile and we want to combine them with one reliable monthly variable.
In particular, the weekly variables are:
Y1;t is the level of weekly google mobility indicator (an indicator of the
movements of people, which is related to economic activity, more movements,
more growth)
Y2;t is the weekly level of the number of restaurant and hotel bookings, a
good indicator of service activity.
Y3;t is the weekly electricity consumption in industry, a good indicator of
the secondary sector.
Y4;t is the number of employees. Monthly and related to contemporaneous
activity.

You decide to write the model in growth rates at weekly frequency. You are
going to use in the model:
y1;t , the weekly growth rate of Y1;t
y2t , the weekly growth rate of Y2;t
y3;t , the weekly growth rate of Y3;t
y4;t , the monthly growth rate of Y4;t

Obviously, we know that the relation between weekly and monthly variables
is not constant for all months because some months contain 4 and other contain
5 weeks. Assume that all the months contain 4 weeks (there is a trick used in
the literature to compute that). Therefore, for the purpose of this exam, what
you have is 3 indicators of weekly activity, 1 indicator of monthly activity and
ALWAYS A MONTH CONTAINS 4 WEEKS.
The model that you have in mind is that all the variables at weekly frequency
are a function of a factor + idiosyncratic part
Just to remind you a little of the relation between variables at di¤erent fre-
quencies, you have at the end of the exam the proof of the relation between
quarterly and monthly variables. YOU DO NOT NEED TO PROVE the re-
lation between monthly and weekly activity, but based on that proof, you can
…gure it out what would be.
1) Write the state space representation. (3 points), assuming that the factor
and the idiosincratic shocks follow an AR(1).
2) Suppose that you have data until period T;June 25th 2021 (weeks end on
Friday). Suppose that the …rst element of …lter (according to the notation seen
in class in the matlab code Homework2) is -0.12. How would you interpret that
number? Why?
3) Obviously the numbers in …lter have no units because you had to stan-
darize all the variables. Suppose that you want to re-escale the factor to describe
quarterly GDP activity (a quarter contains 12 weeks). How would you do that?

1
4) (4 points) Suppose that you want to forecast y1;T +1 ; y2;T +1 ; y3;T +1 ; y4;T +1
(the week ending on July 2nd): You use the trick seen in class of adding 99999
to do this forecasting exercise. Therefore, you estimate the model until July
2nd, being the last row of data, all 99999.
4.1 What is the value of the last Kalman gain matrix (the one in T+1) ?
Why?
4.2 If you forecast (after restandarized) for y4;T +1 a value of 0.3. What is the
economic interpretation of this forecast. Is it weekly growth rate of employment?
Monthly growth rate?
4.3 If we observe y1;T +1 ; y2;T +1 ; y3;T +1 and we forecast the following week
(T + 2), (ending on July 9th), with the usual 99999 trick, how would be the
Kalman gain matrix for observation T + 1. Any columns or row would be equal
to 0? If so, which one?
4.4 If we forecast in period T all the variables for T + 1 and T + 2 and all the
variables released in T + 1 coincide with the forecasted value. Do you think that
the forecast of all the variables for T + 2 after observing the variables released
in T + 1 should be equal, bigger or smaller? Why?

5. This question has nothing to do with the previous ones. Reset your mind.
Suppose that the relation between yt and xt is non-linear following a Markov
Switching speci…cation:

yt = St xt + t

where St = 1 or St = 0
And Pr ob(St = i=St 1 = j; t 1 ) = Pr ob(St = i=St 1 = j)
with t 1 being the information set up to period t-1
Given that you have an observed variable,St you decide to estimate this
model using the Kalman …lter.
Can you do that? How? Why?
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
HELP:Relation between quarterly growth and monthly growth
rate in ‡ow variables.
This relation is important when we talk about activity in months, let´s say
March, Februrary and January and how this activity is related to quarterly
GDP
The idea is the following. Be Q1 is the production in quarter 1 and be the
the variable that has the month name the production in that month.
Q1 = M arch + F ebruary + January
Q1 = (M arch + F ebruary + January)=3
The aritmethic mean can be aproximated by the geometric mean
Q1 = 3 (M arch F ebruary January)1=3
Ln(Q1) = 1=3 (Ln(march) + Ln(f eb) + Ln(Janu)) + Ln(3)
For the same argument
Ln(Q4) = 1=3 (Ln(Dec) + Ln(N ov) + Ln(Oct)) + Ln(3)

2
Ln(Q1) Ln(Q4) = 1=3 (Ln(march) + Ln(f eb) + Ln(Janu)) + Ln(3)
(1=3 (Ln(Dec) + Ln(N ov) + Ln(Oct)) + Ln(3)) =
= 1=3(Ln(M arch) Ln(Dec)) + 1=3(Ln(f eb) Ln(N ov)) + 1=3(Ln(Jan)
Ln(Oct)) =
1=3(Ln(M arch) Ln(f eb) + Ln(f eb) Ln(Jan) + Ln(Jan) Ln(Dec))+
1=3(Ln(f eb) Ln(Jan) + Ln(Jan) Ln(Dec) + Ln(Dec) Ln(N ov))+
1=3(Ln(Jan) Ln(Dec) + Ln(Dec) Ln(N ov) + Ln(N ov) Ln(Oct)) =
= 1=3(ymarch + yf eb + yjan) + 1=3(yf eb + yjan + ydec) + 1=3(yjan + ydec +
ynov) =
= 1=3ymarc + 2=3yf eb + yjan + 2=3ydec + 1=3ynov

where ymonth represent the growth rate on that given month

You might also like