Forecasting MiM Exercises Part3

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Forecasting and predictive analytics – Exercises – part 3

Teacher: Pierre Jacob ([email protected])

Master in Management – ESSEC Business School – 2022

Each of the first two exercises is worth a quarter of the total grade, while the third exercise is worth half of the
total grade. Have fun!

Exercise 3.1. (Ambiguity). We simulate a process as Yt = 0.9Yt−1 +Wt −0.9Wt−1 , where (Wt ) are i.i.d. N (0, 1) and
Y1 = 2.193, for T = 1, 000 steps. Figure 3.1 shows the series, the autocorrelogram and the partial autocorrelogram
of the series. Furthermore, when we try to estimate the parameters by running arima(y, order = c(1,0,1)) in
R, we get:

C a l l : arima ( x = y , o r d e r = c ( 1 , 0 , 1 ) )
Coefficients :
ar1 ma1 intercept
0.5765 −0.5190 0.0048
s.e. 0.2046 0.2132 0.0363
sigma ^2 e s t i m a t e d a s 1 . 0 2 4 : l o g l i k e l i h o o d = −1430.6 , aic = 2869.2

What is surprising about these results? Propose an explanation.

Series y Series y
1.0

0.02 0.04 0.06


4

0.8
2

0.6

Partial ACF
ACF
y

0.4

−0.02
0.2
−2

0.0

−0.06

0 200 400 600 800 1000 0 5 10 15 20 25 30 0 5 10 15 20 25 30

Time Lag Lag

Figure 3.1: Traceplot, autocorrelogram and partial autocorrelogram of (Yt ) in Exercise 3.1.

Exercise 3.2. (State space representation). Consider a four-dimensional process (Xt ). At each time t, the variable
Xt is a column vector with four entries, denoted by Xt,1 , Xt,2 , Xt,3 , Xt,4 . Consider the state equation relating Xt
to Xt−1 :    
η1 1 0
0 t
   
η2
 Xt−1 + ξ1 t  ,
0 1 0  
Xt = 
ξ2 t 
0 0 0 1
  

0 0 0 0 ξ 3 t

1
i.i.d. 
where η1 , η2 , ξ1 , ξ2 , ξ3 are values in R, and (t ) ∼ N 0, σ 2 . We consider a univariate process (Yt ), which relates
to (Xt ) through the observation equation  
Yt = 1 0 0 0 Xt .

Prove that the above state space model is equivalent to an ARMA(p,q) model for (Yt ), for which you will specify
p and q, and express the autoregressive and moving average parameters in terms of η1 , η2 , ξ1 , ξ2 , ξ3 .

Exercise 3.3. (Experts, laypeople and COVID-19). For this exercise, you must first read an article which you can
find on Moodle or here: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0250935,
entitled How well did experts and laypeople forecast the size of the COVID-19 pandemic?, by Gabriel Recchia,
Alexandra L. J. Freeman and David Spiegelhalter. Answer the following questions (briefly!) based on your under-
standing of the article.

1. According to the article, are experts’ predictions more accurate than non-experts’ predictions?

2. Are the predictions well-calibrated, for experts and for non-experts?

3. The authors describe how they turn each of the intervals, say (a, b), into a probability distribution. Denote
by X a random variable following the distribution constructed from the interval (a, b). What is the value of
the probability P(X ∈ (a, b))?

4. The authors write that “probability distributions of different individuals were combined with simple averaging
of the probabilities (i.e. ‘vertical’ combination)” to create the consensus distributions shown in Figures 1.(a)-
(b)-(c)-(d). Why are there spikes in the consensus distributions, particularly apparent in Figure 1.(a)?

5. Which of the following statements are true, according to the article?

• Overconfidence is influenced by the feedback that individuals receive about their predictions.
• Accountability might help to incentivize good calibration among forecasters.
• There is evidence that overconfident experts are invited more often to advise policy-makers and to give
interviews in the media.

You might also like