Estimation of the spectral density function
The spectral density function, f( ) The spectral density function, f(x), is a symmetric function defined on the interval [- , ] satisfying and The spectral density function, f(x), can be calculated from the autocovariance function and vice versa.
Some complex number results: Use
Expectations of Linear and Quadratic forms of a weakly stationary Time Series
Expectations, Variances and Covariances of Linear forms
TheoremLet {x t :t T} be a weakly stationary time series. Let Then and where andS r = {1,2,..., T-r}, if r ≥ 0, S r = {1- r, 2 - r,..., T} if r ≤ 0.
Proof
Also since Q.E.D.
TheoremLet {x t :t T} be a weakly stationary time series. Let and
Then where and Also S r = {1,2,..., T-r}, if r ≥ 0, S r = {1- r, 2 - r,..., T} if r ≤ 0.
Expectations, Variances and Covariances of Quadratic forms
TheoremLet {x t :t T} be a weakly stationary time series. Let Then
and
and S r = {1,2,..., T-r}, if r ≥ 0, S r = {1- r, 2 - r,..., T} if r ≤ 0, (h,r,s) = the fourth order cumulant = E[(x t - )(x t+h - )(x t+r - )(x t+s - )] - [ (h) (r-s)+ (r) (h-s)+ (s) (h-r)] Note (h,r,s) = 0 if {x t :t T}is Normal.
TheoremLet {x t :t T} be a weakly stationary time series. Let Then
and where
Examples The sample mean
and Thus
Also
and where
Thus Compare with
Basic Property of the Fejer kernel: If g() is a continuous function then : Thus
The sample autocovariance function The sample autocovariance function is defined by:
where or if is known
where or if is known
TheoremAssume is known and the time series is normal, then: E(C x (h))= (h),
and
Proof Assume is known and the the time series is normal, then: and
where
since
hence
Thus
and Finally
Where
Thus
Estimation of the spectral density function
The Discrete Fourier Transform
Let x 1,x 2,x 3,...x T denote T observations on a univariate one-dimensional time series with zero mean (If the series has non-zero mean one uses in place of x t ). Also assume that T = 2m +1 is odd. Then
where with k = 2 k/T and k = 0, 1, 2,..., m.
The Discrete Fourier transform: k = 0, 1,2,...,m.
Note:
Since
Thus
Summary: The Discrete Fourier transform k = 0, 1,2,...,m.
Theorem with k k/T) E[X k ] = 0 with k k/T) and h h/T)
where
Proof Note Thus
where
Thus Also
with =2 (k/T)+ with =2 (h/T)+
Thus and
Defn: The Periodogram: k = 0,1,2,..., m with k = 2 k/T and k = 0, 1, 2,..., m.
Periodogram for the sunspot data
note:
Theorem
In addition: If k ≠ 0 If k ≠ h
Proof Note Let
Consistent Estimation of the Spectral Density function f( )
Smoothed Periodogram Estimators
Defn: The Periodogram: k = 0,1,2,..., m
Properties: If k ≠ 0 If k ≠ h
Spectral density Estimator
Properties: If k ≠ 0 The second properties states that: is not a consistent estimator of f( ):
Examples of using packages SPSS, Statistica
Example 1 – Sunspot data
Using SPSS Open the Data
Select Graphs- > Time Series - > Spectral
The following window appears Select the variable
Select the Window Choose the periodogram and/or spectral density Choose whether to plot by frequency or period
Smoothed Estimators of the spectral density
The Daniell Estimator
Properties
Now let T ∞, d ∞ such that d/T 0. Then we obtain asymptotically unbiased and consistent estimators, that is
Choosing the Daniell option in SPSS
k = 5
k = 9
k = 5
Other smoothed estimators
More generally consider the Smoothed Periodogram and where
Theorem (Asymptotic behaviour of Smoothed periodogram Estimators ) and Let where {u t } are independent random variables with mean 0 and variance 2 with Let d T be an increasing sequence such that
and Then Proof (See Fuller Page 292)
Weighted Covariance Estimators Recall that where
The Weighted Covariance Estimator where {w m (h): h = 0, ±1,±2,...} are a sequence of weights such that: i) 0 ≤ w m (h) ≤ w m (0) = 1 ii) w m (-h) = w m (h) iii) w m (h) = 0 for |h| > m
The Spectral Window for this estimator is defined by: i) W m ( ) = W m (- ) ii) Properties :
also (Using a Reimann-Sum Approximation) = the Smoothed Periodogram Estimator Note:
1. Asymptotic behaviour for large T 2. 3.
1. Bartlett Examples w m (h) = w(h/m) Note:
2. Parzen w(x) = 1 -2 a + 2a cos( x) 3. Blackman-Tukey with a = 0.23 (Hamming), a = 0.25 (Hanning)
DaniellTukey Parzen Bartlett
1. Approximate Distribution and Consistency 2. 3.
1. Note: If W m ( ) is concentrated in a "peak" about = 0 and f( ) is nearly constant over its width, then 2. and
Confidence Limits in Spectral Density Estimation
1. Satterthwaites Approximation: 2. where c and r are chosen so that
Thus = The equivalent df (EDF)
and Now Thus
Then a [1- 100 % confidence interval for f( ) is: Confidence Limits for The Spectral Density function f( ) : Let and denote the upper and lower critical values for the Chi-square distribution with r d.f. i.e.