EC3060 Econometrics 2: Time Series Summer Examinations 2021/22
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
EC3060
Summer Examinations 2021/22
Econometrics 2: Time Series
Section A: Answer ONE question
1. (a) An AR(1) model was estimated on time-series data yt . The residuals from the regression were saved as res 1. Results of a post-estimation specifification test are reported in Table 1. Do the results suggest any specifification problem with the AR(1) model for yt? (10 marks)
(b) Defifine the Partial Autocorrelation Function (PAC). Using the regression results in Table 2 sketch the PAC for yt . (10 marks)
(c) Using the results from parts (a) and (b), what model would you propose for the series yt? What further tests or statistics would you compute to confifirm your intuition? (10 marks)
Table 1: Regressions on res 1
Reg 1 Reg 2 res 1 coeff std error coeff std error |
||||
L. res 1 1.200277 .1533356 |
||||
L2.res 1 -.0575336 .1158754 |
||||
L3.res 1 -.0045583 .0792348 |
||||
L4.res 1 .0328803 .0786091 |
||||
L.y .5823063 .1581734 -.0014483 .0388487 |
||||
const -.0801802 .0585634 -.0157385 .0944114 |
||||
SS Model SS Residual Nobs |
705.00 347.04 345 |
.0042627 1052.0 345 |
||
|
||||
Table 2: yt regressions (standard errors in parentheses) |
||||
yt 1 2 3 4 |
||||
yt_1 .6933 1.259 1.314 1.309 |
||||
( .038764 ) ( .03095) ( .05388) ( .05404) |
||||
yt_2 |
-.8188 -0.9022 -.8893 |
|||
( .03095) (0 .7448) ( .08938) |
||||
yt 3 |
.06648 .05586 |
|||
( .05386) ( .08952) |
||||
yt 4 |
.005294 |
|||
( .05430) |
||||
const -0.04218 0.06997 -.06826 -.07291 |
||||
(.09414) (.05430) (.05445) (.05459) |
||||
SS Model SS Residual Nobs |
986.5 1070.2 349 |
1698 353.0 348 |
1700 350.4 347 |
1693 347.4 346 |
(d) After how many periods would the influence of a unit shock to your proposed process have decayed to remain within ±0y1e where e is the standard deviation of yt ? (20 marks)
2. (a) Consider the generic model for the conditional variance of a process, yt : et = zt ^gt where ∈t = yt - E「yt |●t_1[ and E「∈t(2)|●t_1[ = gt and ●t_1 denotes the information set available at date t - 1.
(i) What assumptions on zt make this a standard GARCH process? (5 marks)
(ii) What features of conditional variance do GARCH models capture well and what are the main drawbacks of the assumptions made above for modelling high-frequency financial data? (5 marks)
(iii) Describe how you would estimate such a model. (15 marks)
(b) Consider an example of the GARCH(2,1) variant:
gt = 0y15 + 1y2∈t(2)_1 + 0y2∈t(2)_2 - 0yYgt_1
(i) Find the ARMA process for the ∈t(2) associated with this model of conditional variance, taking care to define all terms. (10 marks)
(ii) Assume that the ∈t(2) process had been at the unconditional variance for at least two periods, and that the conditional volatility forecast was also equal to its mean prior to the arrival of a shock to conditional volatility in period t + 1 which raised the
volatility 1 unit above the unconditional variance of the process .
Assuming no further shocks, how many periods will it be before the conditional
volatility returns to be less than the unconditional variance + unconditional standard deviation of the process? (15 marks)
Section B: Answer ONE question
3. Consider the process:
yt = c + Φ1yt−1 + Φ2yt−2 + ut (1)
where yt is an n by 1 vector and E[utu 0 t ] = Σ, a symmetric positive-defifinite matrix.
(a) Find the unconditional mean of the process. Under what conditions does this moment exist? For full marks state the condition in terms of a matrix you can derive from (1). (10 marks)
(b) Show the k period ahead forecast of the process can be written as a weighted average of the unconditional mean and the last observations on the process. (15 marks)
(c) Let n = 2. Find conditions on the VAR parameters and a structural interpretation such that you could recover estimates of the parameters in equation (2) from the system:
y1t = b1 + 30y2t + 322y1t_1 + 321y2t_1 + ∈t (2)
Show your reasoning. (15 marks)
(d) Consider equation (1) and imagine yt ~ 1(1). How would you proceed with estimation? (10 marks)
4. (a) Consider the processes:
zt = xt + t
yt = gxt + 6Zt + y]t
zt = 3Zt + t
where xt ~ 1(1), Zt ~ 1(1) and (t . y]t . t ) ~ 1(0). g , 3 , 6 are non-zero parameters. (i) What is the order of integration of yt - gzt ? (5 marks)
(ii) Find a cointegrating relation between the variables and discuss its uniqueness.
(10 marks)
(b) Consider the system:
Yt = ●Yt_1 + ∈t (3)
where Yt ~ 1(1) is an n by 1 vector and ∈t ~ 1(0).
(i) Given these properties of Yt , ∈t , does the system (3) necessarily display cointegration? Give your reasoning. (5 marks)
(ii) Find an expression for the k-step ahead variance of Yt . Assume E「∈t ∈ [ = x for all t. Given the information above, will this expression tend to a limit? (10 marks)
(c) A researcher is interested in the relationship between two I(1) variables ct , it .
(i) Consider the Root Mean Square Errors (RMSEs) reported for VAR(4) and AR(4) models of the two variables in Table 3. Is there any possibility that they cointegrate? (5 marks)
(ii) Do the Johansen test results in Table 4 support the idea of cointegration between ct and it ? (5 marks)
VAR(4) 7(L)it 7(L)ct |
|||
it .01556 .0165 ct .005915 .005949 |
|||
Nobs |
240 |
240 |
240 |
|
Table 4: Tests of co-integrating rank
rank |
params |
Case 1: trend = constant ll eigenvalue trace |
ctitical value |
||
0 1 2 |
14 17 88 |
1572.591 1581.741 1583.531 |
. 0.07341 0.01481 |
21.88 3.5803 |
15.41 3.76 |
rank |
params |
Case 2: trend = rtrend ll eigenvalue trace |
ctitical value |
||
0 1 2 |
14 18 20 |
1572.6 1583.7 1587.0 |
. 0.08833 0.02738 |
28.858 6.6630 |
25.32 12.25 |
(d) Discuss the identification problem in the VECM, and the nature of the restrictions needed to achieve just-identification. In an empirical investigation, a researcher imposes a single over-identifying restriction which returns a o2 value of 3.83. Is this additional restriction a good fit to the data? (10 marks)
2023-06-02