Scribd
Upload
27 views

Midterm Cheatsheet

MAD is the mean absolute deviation and MSE is the mean squared error, which are used to measure error in predictions. A time series is stationary if its mean and variance are constant over time and the covariance between two times is dependent only on the lag between the times. The autocorrelation function (ACF) measures the correlation between values at different times in a time series. A MA(q) process is stationary and its ACF cuts off after lag q, while an AR(p) process is stationary if all roots of its polynomial are outside the unit circle and its partial autocorrelation function (PACF) cuts off after lag p. An ARIMA(p,d,q) combines AR and MA components

Uploaded by

Joel Tan Yi Jie
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
27 views

Midterm Cheatsheet

MAD is the mean absolute deviation and MSE is the mean squared error, which are used to measure error in predictions. A time series is stationary if its mean and variance are constant over time and the covariance between two times is dependent only on the lag between the times. The autocorrelation function (ACF) measures the correlation between values at different times in a time series. A MA(q) process is stationary and its ACF cuts off after lag q, while an AR(p) process is stationary if all roots of its polynomial are outside the unit circle and its partial autocorrelation function (PACF) cuts off after lag p. An ARIMA(p,d,q) combines AR and MA components

Uploaded by

Joel Tan Yi Jie
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

MAD and MSE

MAD : sum of abs(actual – predicted) divided by total count


MSE : sum of abs(actual - predicted)2 divided by total count

Stationary
1) Mean must be constant (not dependent on time t)
X t =exp ( t 2 +t S t + Zt ) , St =St −12 , Z t=WN ( 0 ,1 )
E [ X t ]=exp ( t 2 +t S t )∗E [ Z t ]
X t is not stationary

2) Variance must be constant (not dependent on time t) Difference Operator ∇


X t =Z t +c ( Z t −1+ Z t −2 +…+ Z 1 ) ∇ X t =X t −X t −1=( 1−B ) X t
2
Var ( X t )=Var ( Z t ) +c ( t−1 ) Var ( Z t ) B X t =X t −1 ; Bi X t= X t −i ; ∇ i X t =∇ ( ∇i−1 X t )
E.g. Transforming X t =exp ( t +t S t + Zt )to a stationary series
2
When c =0 => it is stationary, when c ≠ 0 => not stationary
If 1 and 2 does not hold, conclude X t not weakly stationary. Verify with 3 2
Y t =ln X t =t + t St + Z t
3) Covariance γ ( t , k )=cov ( X t , X t +k ) =0 => X t is stationary ∇ 12 Y t =Y t −Y t −12=t 2 +t St + Z t −[ ( t−12 )2 + ( t−12 ) S t−12+ Z t −12 ]
γ k =cov ( X t , X t +k ) =E [ ( X t −μ ) ( X t+ k −μ ) ] [THIS IS THE ACVF AT LAG k] ¿−144+12 S t −12 +24 t +Z t −Z t−12
∇ 12 Y t =−144+ 12 St −12+24 t+ Z t−Z t−12−[−14 4+ 12 St −12+24 ( t−12 )+ Z t −12−Z t−24 ]
2
(ACVF is finding the covariance function)
Strictly stationary if for any n ∈ Z+ and all integers h, (X1, ...,Xn) and (X1+h, · · ¿ 288+ Z t−2 Z t−12+ Z t −24
· ,Xn+h), s ∈ Z have the same distributions. 2
The appropriate transformation is ∇ 12 ln ( X t )

Autocorrelation functions (ACF)


MA(q)
cov ( X t , X t +k ) γk MA(q) models are stationary
ρk = =
√ var ( X t ) var ( X t +k ) γ 0
( ) ( )
q q
E [ X t ]=0 since E [ Zt ]=0 ; Var ( X t )=Var Z t + ∑ θ j Z t − j =σ 2 1+ ∑ θ2j
γ 0=var ( X t ) ; ρ0=1 j=1 j =1
2
ACVF=Cov ( X t−k , X t )=( θ k + θ1 θ k+1 +…+θ q−k θq ) σ
e.g. Derive ACFs
X t =Z t−θ 1 Z t−1−θ2 Z t −3
Cov ( X t−k , X t )=Cov ( Z t− k −θ1 Z t− k−1−θ2 Z t−k−3 , Z t −θ1 Z t −1−θ2 Z t−3 )
** expand each term like ( Z t−k −θ 1 Z t−k−1−θ 2 Zt −k−3 ) ( Z t −θ1 Z t −1−θ2 Z t −3 ) but
take Cov of them and put coeffs outside** ACF
2
E.g Cov ( Z t−k , Z t )−θ1 Cov ( Z t−k , Z t −1 ) + …+θ2 Cov ( Zt −k−3 , Z t−3 )
2
NOTE : Cov ( Z t , Z t )=Var ( Zt )=σ ** match with same subscript**
MA(2) because ACF cuts off after lag 2 (1st graph)
AR(p)
AR(p) is stationary if all roots of polynomial ϕ ( z ) are outside unit circle (>1)

AR(2) because PACF cuts off after lag 2 (2nd graph)


ARIMA(p,d,q)

Stationary?

E.g. X t = X t−1−0.25 X t−2 +Z t


Find p,d,q

Invertible?

[
Calculate E ∇ X t
2
]

You might also like