Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 31 views

    STAT721 Test1 2022 Solutions

    Uploaded by

    Jacquiline Isaac
    This document contains a multi-part question regarding time series analysis and modeling of time series data. Several models are fit to different datasets including an ARMA model, linear regression model, and ARIMA models. Diagnostic checks are performed and conclusions are made regarding the adequacy of different model fits.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    STAT721 Test1 2022 Solutions

    Uploaded by

    Jacquiline Isaac
    31 views5 pages
    This document contains a multi-part question regarding time series analysis and modeling of time series data. Several models are fit to different datasets including an ARMA model, linear regression model, and ARIMA models. Diagnostic checks are performed and conclusions are made regarding the adequacy of different model fits.

    Original Description:

    STAT721

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document contains a multi-part question regarding time series analysis and modeling of time series data. Several models are fit to different datasets including an ARMA model, linear regression model, and ARIMA models. Diagnostic checks are performed and conclusions are made regarding the adequacy of different model fits.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    31 views5 pages

    STAT721 Test1 2022 Solutions

    Uploaded by

    Jacquiline Isaac
    This document contains a multi-part question regarding time series analysis and modeling of time series data. Several models are fit to different datasets including an ARMA model, linear regression model, and ARIMA models. Diagnostic checks are performed and conclusions are made regarding the adequacy of different model fits.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 5

    Question 1 4 marks

    Given

    𝑦𝑡 = 0.3 + 𝑢𝑡 + 0.5𝑢𝑡−1 − 0.3𝑢𝑡−2


    𝑢𝑡 ~𝑖𝑖𝑑(0, 𝜎 2 )
    a) Compute by hand 𝑐𝑜𝑟𝑟(𝑦𝑡 , 𝑦𝑡+1 ) and 𝑐𝑜𝑟𝑟(𝑦𝑡 , 𝑦𝑡+2 )

    𝐶𝑜𝑣(𝑦𝑡 , 𝑦𝑡 ) = 𝐶𝑜𝑣(𝑢𝑡 + 0.5𝑢𝑡−1 − 0.3𝑢𝑡−2 ; 𝑢𝑡 + 0.5𝑢𝑡−1 − 0.3𝑢𝑡−2 )


    = 𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡 ) + 0.5𝑐𝑜𝑣 (𝑢𝑡 ; 𝑢𝑡−1 ) − 0.3𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡−2 )

    +0.5𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡 ) + 0.52 𝐶𝑜𝑣(𝑢𝑡−1 ; 𝑢𝑡−1 ) − 0.5(0.3)𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡−2 )

    −0.3𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡 ) − 0.3(0.5)𝑐𝑜𝑣 (𝑢𝑡−2 ; 𝑢𝑡−1 )+(−0.3)2 𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡−2 )


    = (1 + 0.52 + (−0.32 ))𝜎 2 = 1.34𝜎 2

    𝐶𝑜𝑣(𝑦𝑡 , 𝑦𝑡+1 ) = 𝐶𝑜𝑣(𝑢𝑡 + 0.5𝑢𝑡−1 − 0.3𝑢𝑡−2 ; 𝑢𝑡+1 + 0.5𝑢𝑡 − 0.3𝑢𝑡−1 )


    = 𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡+1 ) + 0.5𝑐𝑜𝑣 (𝑢𝑡 ; 𝑢𝑡 ) − 0.3𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡−1 )

    +0.5𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡+1 ) + 0.52 𝐶𝑜𝑣(𝑢𝑡−1 ; 𝑢𝑡 ) − 0.5(0.3)𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡−1 )

    −0.3𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡+1 ) − 0.3(0.5)𝑐𝑜𝑣 (𝑢𝑡−2 ; 𝑢𝑡 )+(−0.3)2 𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡−1 )


    =[0.5 − 0.3(0.5)] 𝜎 2 = 0.35𝜎 2

    𝐶𝑜𝑣(𝑦𝑡 , 𝑦𝑡+2 ) = 𝐶𝑜𝑣(𝑢𝑡 + 0.5𝑢𝑡−1 − 0.3𝑢𝑡−2 ; 𝑢𝑡+2 + 0.5𝑢𝑡+1 − 0.3𝑢𝑡 )


    = 𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡+2 ) + 0.5𝑐𝑜𝑣 (𝑢𝑡 ; 𝑢𝑡+1 ) − 0.3𝐶𝑜𝑣(𝑢𝑡 ; 𝑢𝑡 )
    +0.5𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡+2 ) + 0.52 𝐶𝑜𝑣(𝑢𝑡−1 ; 𝑢𝑡+1 ) − 0.5(0.3)𝑐𝑜𝑣 (𝑢𝑡−1 ; 𝑢𝑡 )

    −0.3𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡+2 ) − 0.3(0.5)𝑐𝑜𝑣 (𝑢𝑡−2 ; 𝑢𝑡+1 )+(−0.3)2 𝐶𝑜𝑣(𝑢𝑡−2 ; 𝑢𝑡 )

    =−0.3𝜎 2

    => 𝑐𝑜𝑟𝑟(𝑦𝑡 , 𝑦𝑡+1 ) =0.35/1.34=0.2612

    𝑐𝑜𝑟𝑟(𝑦𝑡 , 𝑦𝑡+2 ) =-0.3/1.34=-0.2238


    b) Now use appropriate R code to compute a partial autocorrelation function for the first 4 lags
    > ARMAacf(ma=c(0.5,-0.3),pacf=TRUE,lag.max=4)
    [1] 0.2611940 -0.3134899 0.1888699 -0.1654122

    Question 2 8 marks
    Use the following code to load a dataset called airmiles into RStudio from the TSA package
    > library(TSA)

    > data(airmiles)
    > class(airmiles)
    [1] "ts"
    > frequency(airmiles)
    [1] 12

    a) Generate a plot of the data making sure that the label on the time axis reads
    Time in months

    > ts.plot(airmiles, xlab="Time in months")

    b) Use appropriate code to fit the following model to this data set
    ln 𝑦𝑡 = µ + 𝑢𝑡 ; 𝑢𝑡 ~𝑖𝑖𝑑(0, 𝜎 2 )
    Clearly list the output that you have obtained along with the appropriate coding that has been
    used.
    > airmiles=log(airmiles)
    > out=lm(airmiles~1)
    > summary(out)

    Coefficients:
    Estimate Std. Error t value Pr(>|t|)
    (Intercept) 17.50735 0.01315 1332 <2e-16 ***
    ---
    Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

    Residual standard error: 0.1398 on 112 degrees of freedom

    ln 𝑦𝑡 = 17.50735 + 𝑢𝑡 ; 𝑢𝑡 ~𝑖𝑖𝑑(0, 𝜎 2 = 0.13982 )

    c) Generate an autocorrelation plot for ln 𝑦𝑡

    > out=acf(airmiles, type="correlation")


    d) Given that there is autocorrelation in the data, what consequence does this have for the
    standard error associated with the estimate of µ that you obtained in (b)
    The standard errors will be incorrect/inconsistent

    e) Use appropriate code to perform a Newey-West adjustment to the results you obtained
    in (i). Clearly show the code that you have used and the results you have obtained
    > coeftest(out,vcov=NeweyWest(out,verbose=T))

    Lag truncation parameter chosen: 2

    t test of coefficients:

    Estimate Std. Error t value Pr(>|t|)


    (Intercept) 17.507347 0.024388 717.86 < 2.2e-16 ***
    ---
    Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

    f) Assume we want to test 𝐻0 : 𝜇 = 17. Use the output in e) to derive an appropriate t-value
    for testing this hypothesis.
    17.507−17
    𝑡= 0.02438
    =20.79

    Question 3 3 marks

    Given

    𝑦𝑡 = 𝑦𝑡−1 + 2 + 𝑢𝑡 ; 𝑢𝑡 ~𝑖𝑖𝑑(0, 𝜎 2 )

    compute appropriate expressions for

    𝑦̂𝑡+𝑘|𝑡 ≡ 𝐸(𝑦𝑡+𝑘 |𝑦𝑡 … 𝑦1 )

    and

    𝑉𝑎𝑟(𝑒̂𝑡+𝑘|𝑡 |𝑦𝑡 … 𝑦1 )

    where 𝑒̂𝑡+𝑘|𝑡 ≡ 𝑦𝑡+𝑘 − 𝐸(𝑦𝑡+𝑘 |𝑦𝑡 … 𝑦1 )

    Note

    𝑦𝑡+𝑘 = 𝑦𝑡 + 2𝑘 + 𝑢𝑡+𝑘 + ⋯ + 𝑢𝑡+1


     𝐸(𝑦𝑡+𝑘 |𝑦𝑡 … 𝑦1 ) = 2𝑘
     𝑉𝑎𝑟(𝑒̂𝑡+𝑘|𝑡 |𝑦𝑡 … 𝑦1 ) = 𝑉𝑎𝑟(𝑢𝑡+𝑘 + ⋯ + 𝑢𝑡+1 ) = 𝑘𝜎 2
    Question 4 10 marks

    Use the following code to load a dataset called co2 into RStudio from the TSA package
    > library(TSA)
    > library(forecast)
    > data(co2)
    > head(co2)
    Jan Feb Mar Apr May Jun
    1994 363.05 364.18 364.87 364.47 364.32 362.13
    > plot(co2)

    a) Clearly list appropriate code that will fit a AR(2) model with a nonzero mean to this data
    set. Write down the model structure that has been estimated
    > library(forecast)
    > fit= Arima(co2,order=c(2,0,0))
    > fit

    Series: co2
    ARIMA(2,0,0) with non-zero mean

    Coefficients:
    ar1 ar2 mean
    1.5193 -0.6874 369.1537
    s.e. 0.0625 0.0627 1.2611

    sigma^2 = 6.139: log likelihood = -307.02


    AIC=622.05 AICc=622.36 BIC=633.58

    𝑦𝑡 = 369.153 + 𝑒𝑡
    𝑒𝑡 = 1.5193𝑒𝑡−1 − 0.6874𝑒𝑡−1 + 𝑢𝑡

    b)Fit a model with an appropriate linear trend to this data. Clearly list the code you have
    used and write down the model structure that has been estimated
    > t=time(co2)
    > head(t)
    Jan Feb Mar Apr May Jun
    1959 1959.000 1959.083 1959.167 1959.250 1959.333 1959.417
    > out=lm(co2~t)
    > out
    Coefficients:
    (Intercept) t
    -3127.241 1.749

    𝑦𝑡 = −3127.24 + 1.749 ∗ 𝑡 + 𝑒𝑡

    c) Use the above fitted model to forecast and outcome for co2 for April 1998

    forecast=-3127.24+1.749*1998.25=367.69
    d)Now fit a random walk model with a non-zero drift term to this dataset. Clearly list the
    code you have used and write down the model structure that has been estimated
    > out=Arima(co2,c(0,1,0),include.drift=TRUE)
    > out
    Series: co2
    ARIMA(0,1,0) with drift

    Coefficients:
    drift
    0.1418
    s.e. 0.3020

    sigma^2 = 12.04: log likelihood = -348.34


    AIC=700.68 AICc=700.77 BIC=706.43

    𝑦𝑡 = 𝑦𝑡−1 + 0.1418 + 𝑒𝑡

    e)Perform an appropriate test to determine if the model that has been fitted in d) is adequate
    > checkresiduals(out)

    Ljung-Box test

    data: Residuals from ARIMA(0,1,0) with drift


    Q* = 573.99, df = 23, p-value < 2.2e-16

    Model df: 1. Total lags used: 24

    Conclude model fit NOT adequate


    f) How could you set about improving the model fits that you have obtained in the
    question
    Need to include an annual seasonal component

    You might also like