State Space Models and the

Kalman Filter
References:

RLS course notes, Chapter 7.

Brockwell and Davis, Chapter 12.

Harvey, A.C. (1989), Forecasting. Structural Time Series Models

and the Kalman Filter. Cambridge University Press.

A. Pole, M. West, and P.J. Harrison (1994), Applied Bayesian

Forecasting and Time Series Analysis. Chapman-Hall, New
York.

M. West and P.J. Harrison (1997, First edition 1989), Bayesian

Forecasting and Dynamic Models. Springer-Verlag, New York.
1
Basic equations:

Xt = FtSt + vt,
St = GtSt1 + wt,
vt N [0, Vt],
wt N [0, Wt],
where

Xt is (multivariate) observation process

St is unobserved state process

Ft, Gt, Vt, Wt in principle known, though may have to be

estimated in practice.

2
Why consider such models?

All ARMA or ARIMA models may be rewritten as state space

models

Extends to multivariate case automatically

Many nonstationary or seasonal models...

Natural formulation of Bayesian approach

3
Solution by Kalman Filter

Note on terminology:

A state space model is in principle any model that includes an

observation process Xt and a state process St. The equations
may be nonlinear, or non-Gaussian.

The Kalman Filter is a particular algorithm that is used to solve

state space models in the linear case. This was first derived by
Kalman (1960).

Some people refer to Kalman filter models but in my view this

is imprecise terminology.

4
 Bayesian derivation of the Kalman Filter

Follows Meinhold and Singpurwalla (1983).

We use the following fact familiar from multivariate analysis:

If
! " ! !#
Y1 1 11 12
MV N
Y2 2 21 22
then
1 1
h i
Y1 | Y2 = y2 M V N 1 + 1222 (y2 2), 11 1222 21 .

5
Let X t denote all information available up to time t (the -algebra
generated by Xs, s t). Suppose

St1 | X t1 M V N [St1, Pt1].

But St = GtSt1 + wt so

St | X t1 M V N [GtSt1, Rt], Rt = GtPt1GT

t + Wt .
Then
! " ! !#
St GtSt1 Rt RtFtT
| X t1 M V N , .
Xt FtGtSt1 FtRt FtRtFtT + Vt
Hence

St | Xt, X t1
M V N [GtSt1 + RtFtT (FtRtFtT + Vt)1(Xt FtGtSt1),
Rt RtFtT (FtRtFtT + Vt)1FtRt].

6
Thus we have the recursive equations
St = GtSt1 + RtFtT (FtRtFtT + Vt)1(Xt FtGtSt1),
Pt = Rt RtFtT (FtRtFtT + Vt)1FtRt.
Issues:

Initiation of S0, P0

Prediction and smoothing: estimate St given X1, ..., XT . t >

T is prediction problem, 1 t < T is smoothing problem.

Estimation: assume parametric forms, Ft = Ft() etc. De-

fine likelihood function using prediction error decomposition

T
f (Xt | X t1).
Y
f (X1, ..., XT | ) =
t=1
In fact, the modern derivation of exact MLE for ARMA pro-
cesses is based on this approach.

7
 Application to Financial Time Series

Let yt be day-t return. GARCH(1,1) model:

yt = tt, t N [0, 1], t2 = 0 + 1yt1

2 + 2 .
1 t1

Alternatively, the simplest case of a stochastic volatility model

Direct estimation by MLE possible.

yt = teht/2, ht+1 = 0 + 1ht + t, t N [0, Ht].

Solution by Monte Carlo/Bayesian methods (focus of current
Sequential Monte Carlo program at SAMSI).

Also generalized method of moments (GMM) approaches popular

among econometricians.

8
 Long-Range Dependence

References:

Brockwell and Davis, Section 13.2

J. Beran (1994), Statistics for Long-Memory Processes. Chap-

man and Hall, New York.

P. Doukhan, M.S. Taqqu and G. Oppenheim (eds.) (2003),

Long-Range Dependence. Birkhauser, Boston.

9
The covariance function of any causal invertible ARMA process
satisfies

|k | Ark , some r [0, 1).

Long-range dependence is concerned with processes that satisfy

k Ck2d1, some C > 0, d. (1)

 
1 1
Stationary and invertible if d 2 , 2 .

P 2
|k | = though k <
P
Note that

1 where H is Hurst coefficient

Also write H = d + 2

d = 0 or H = 1
2 correspond to standard short-range (including
ARMA processes)
10
 History

Hurst (1951) first proposed this model (with H 0.7) based on

a study of hundreds of years of flood volumes of the River Nile.

Mandelbrot (c. 1968) proposed a model of Fractional Brownian

Motion: a Gaussian process with the self-similarity property

Yct cH Yt, t (, )
for any fixed c > 0. Differences of FBM are called Fractional
Gaussian Noise.

Current applications include environmental (e.g. climatic) data,

finance, internet traffic and many others.

11
 Fractional Differencing

Introduced independently by Granger and Joyeux (1980) and

Hosking (1981).
(I B)dYt ARM A(p, q)
that has a covariance satisfying (1). Here we interpret
( )
d(d 1) 2 d(d 1)(d 2) 3
(I B)dYt = I dB + B B + ... Yt
2! 3!
d(d 1) d(d 1)(d 2)
= Yt dYt1 + Yt2 Yt3 + ...
2! 3!
The case p = q = 0 is fractionally integrated noise. Its spectral
density and autocorrelation function are given by
2 2d 2
 
f () = |1 ei|2d = 2 sin ,
2 2 2
(k + d)(1 d)
k =
(k d + 1)(d)

12
 Estimation

Assume general form of model ARIMA(p, d, q) though some of

the methods are designed to work in more general cases, e.g.
f () = |1 ei|2df0() with f0 a bounded spectral density.

1. Maximum Likelihood

2. Spectral Methods

3. Wavelet Methods

13
Maximum Likelihood Approach (see Brockwell and Davis)

In principle can write down exact likelihood for ARMA(p, d, q)

processes, treating the autoregressive parameter , the moving
average parameter and the fractional differencing parameter d
as unknown parameters = (, , d).

Classical asymptotics of MLE hold in this situation (but its very

hard to prove that!

In practice often use Whittle approximation

( )
X I(j )
log L log f (j ; ) +
j f (j ; )
where {j } are Fourier frequencies, I is periodogram and f is
spectral density.

This has same asymptotic properties as exact MLE.

14
 Spectral Approach

We know f () C2d for near 0. Rewrite as

log f () log C 2d log . (2)

Suggests trying to fit (2) directly.

First approach: GewekePorter-Hudak (1983) proposed an OLS

regression of log I(j ) on log j , first m Fourier frequencies, some
m << T2 .

Second approach: Minimize

m
X I(j )
2d
log(Cj ) + 2d
.
j=1
Cj
Now known as Local Whittle Estimator following seminal paper
of P.M. Robinson (1995).
15
 Wavelets Approach

Wont go into details but this may be the best approach overall.

See several papers of Peter Craigmile (Ohio State), also Vladas

Pipiras of UNC.

16
 Hurricanes and Global Warming

What this is about:

A problem related to climate change that involves some very

challenging applications of time series analysis!

Acknowledgements to Thomas Knutson (NOAA/GFDL) and Evan-

gelos Evangelou (UNC).

17
18
19
20
 Some Basic Facts About Hurricanes

(From a presentation by Kevin Trenberth)

SSTs > 26oC (80oF)

High water vapor

Weak wind shear

Weak static stability

Pre-existing disturbance

Also:

Large variability from year to year

El Nino means more activity in Pacific, less in Atlantic
Large interdecadal variability in Atlantic (AMO)
21
Source: Presentation by Kevin Trenberth

22
 Hurricanes play a key role in climate, but are not in models
and are not parameterized

Competition between thunderstorms and convection, but these

are not resolved and are treated as sub-grid-scale phenomena

Climate models have premature onset of convection

Result: Existing models are likely to underpredict hurricanes

23
This analysis compares two reconstructions of the TC series
(Vecchi and Knutson)

One-encounter: Assumes a single encounter between a mod-

ern storm and a historical ship track is sufficient to count the
storm as one that would have been observed historically

Two-encounter: Similar, but requires two ship storm en-

counters

The two-encounter model applies a stronger correction to the

historical record

24
 Tropical Cyclones by Year
(Data by Vecchi and Knutson)

Oneencounter *
25

Twoencounter
Tropical Cyclones

** *
20

* *
* **
15

* * ** *
*
* * * * *
* * ** * * ***** ** * * * * * * ** *** *
** * ** * * * * ** * ** * * * * *
10

*** * * * * *
* * * * * * *
* * ** *
* * *
*** ** * ** ** ** *
*** **** * * ** ** * * ** ** * **** *
* * * * **
********* *
5

*
*
** * **

1880 1900 1920 1940 1960 1980 2000

Year

25
 Three Reconstructions of SST

Reynolds
Kaplan *
Hadley **
1.0

* ***
* *
SST Anomaly

** ** * * * * * * ***
*
** ** **** ******* ** * * *** ********
0.5

* ** * **** *** ********** ** * *** ** * *

** ** * ** ** * *
** * * * *** ********** ***** * * * ********* *** **** *********
* * *
* * * **
***************** * * ********** **** * *
* **** * * ******* ******** ** *
* * ** *** * * ****** *** ** **** * * * * ***
*
0.0

* * * * *
***** ** **** *** * ** *** ************ ******* ** * ** * * * * ******** * *** ****
** * * * *
* * ** * ************************ * *** **
* * * * *
* *** *** * *** **
0.5

* *
* ** * **

1850 1900 1950 2000

Year

26
 Trends of TC Counts and SSTs

TC (1enc)
4

TC (2enc)
SST (Reynolds)
Normalized Trend

SST (Kaplan)
SST (Hadley)
AMO
2
1
0
2

1850 1900 1950 2000

Year

27
Trends in hurricane counts are similar to those in TCs but the
data are much sparser

US landfalling hurricanes show no trend or a slight decrease but

this can be explained as a sampling effect

28
Linear and Nonlinear Trends in Three Series
29
 One analysis is a simple bivariate time series analysis of TCs
and SSTs

Apply square root transformation to TCs to normalize vari-

ances

After prewhitening both series, lag-0 cross-correlation is 0.27

with a p-value of about .002

30
 Series 1 Series 1 x Series 2

1.0

1.0
0.5

0.5
0.0

0.0
0.5

0.5
1.0

1.0
0 5 10 15 20 0 5 10 15 20

Series 2 x Series 1 Series 2

1.0

1.0
0.5

0.5
0.0

0.0
0.5

0.5
1.0

1.0

0 5 10 15 20 0 5 10 15 20

31
 Series 1 Series 1 x Series 2

1.0

1.0
0.5

0.5
0.0

0.0
0.5

0.5
1.0

1.0
0 5 10 15 20 0 5 10 15 20

Series 2 x Series 1 Series 2

1.0

1.0
0.5

0.5
0.0

0.0
0.5

0.5
1.0

1.0

0 5 10 15 20 0 5 10 15 20

32
 Why Linear Trends?

The gold standard for this kind of analysis is detection and attri-
bution analysis (D&A), which is a statistical technique devised
by climatologists that is used to apportion an observed climate
signal as a combination of different forcing factors.

33
 Outline of D&A Method

Use a climate model to generate signals under different

forcing factors. The main forcing factors used in practice are
(a) greenhouse gases, (b) atmospheric particles (aerosols),
(c) solar fluctuations, (d) volcanoes.

Perform a multiple regression analysis to decompose the ob-

served climate signal as a linear combination of the model
signals. The details of this analysis are quite complicated,
because the signals have dimension 5000.

If the coefficient due to greenhouse gases is statistically sig-

nificant, we say the greenhouse gas signal is detected. If
this condition is satisfied, the regression coefficients are then
interpreted as an attribution of the observed signal over
different forcing factors.

34
This method has been applied many times to temperature se-
ries, and more recently to rainfall. However there are difficulties
applying it to hurricane data, because hurricanes are not well
reproduced by climate models.

Conflicting evidence from climate models, e.g. the GFDL

model suggests a levelling off of hurricane activity as SST
increases (Knutson) but the NCAR model suggests the op-
posite conclusions (Trenberth)

In the absence of an agreed definition of the greenhouse

gas signal, we use a linear trend as an approximation.

The idea is to decompose observed SST and hurricane/storm

counts as a combination of linear trend and long-term fluc-
tuations (AMO)

35
 Trends Fitted to Tropical Cyclones

Period ARMA Trend SE Trend/SE p-value

18782006 (0,0) 0.018 0.0094 1.94 0.055
18782006 (9,2) 0.022 0.022 0.97 0.33
19002006 (0,0) 0.049 0.012 4.11 8 105
19002006 (9,4) 0.050 0.020 2.54 0.011

36
 Tropical Cyclones with Trends
OLS and ARMA Regression
*
25

* *
20
TC Count

* *
* **
15

* * * *
*
* * * * * *
** * * * * **** ** * * * * * * * * * *
** * ** * * * ** * ** * * * * *
10

* ** * * ** *
* * * * * * *
* * *
** **
** * * ** ** * * * * ** * ** ** * **** *
*
* ** ** * * ** * **
5

* **** * *
* * *

1880 1900 1920 1940 1960 1980 2000

Year

37
 Bivariate Time Series Model

xt: SST average in year t

yt: square root TC count in year t

Model:
! ! ! !
xt 0,0 0,1 wt
= + t+ ,
yt 1,0 1,1 zt
! p ! ! !
wt X a0,0,j a0,1,j wtj t,0
= + ,
zt a1,0,j a1,1,j ztj t,1
j=1
! " ! !#
t,0 0 0,0 0,1
N , .
t,1 0 1,0 1,1

Calculate 1,1 and its standard error.

38
 Selection of AR order p

p AIC
0 219.7
1 240.8
2 236.8
3 230.2
4 222.5
5 228.5
6 223.8
7 219.8
8 217.2
9 211.9
10 218.4

Conclusion: p = 1 seems best

39
 Trend Coefficient 1,1 and p-value

Start One-encounter Two-encounter

Year Trend p-value Trend p-value
1878 .20 .25 .14 .40
1890 .51 .006 .43 .02
1900 .71 .0005 .61 .002
1910 .99 .00003 .88 .0002
1920 .86 .003 .78 .007
1930 .43 .17 .35 .27
1940 .45 .20 .35 .32
1950 .78 .08 .74 .10
1960 1.58 .006 1.55 .007
1970 2.22 .008 2.22 .008
1980 4.21 .002 4.21 .002

Still get different results corresponding to the start time

40
 Questions About This Approach

There are clear signs of unit root behavior. I tried fitting mul-
tivariate ARMA models (in SAS - PROC VARMAX) but these
generally dont converge, because of unit root difficulties.

For the hurricane series on its own, ARMA models fit much
better than AR, but still difficulties with AIC and similar measures
(indicative of non-stationarity?)

I also tried fractionally differenced processes but this doesnt

seem to resolve the issues either.

41
 Current Approach Conditional ARMA

Recall
xt: SST in year t
yt: square root TC in year t

Also let Tt denote some modeled trend (initially linear, but later
we consider alternatives)

Model:

xt = 0 + 1Tt + ut (ut ARMA)

ut = xt 0 1Tt, (residuals)
yt = 0 + 1Tt + 2ut + 3ut1 + 4ut2 + vt (vt ARMA)

42
 First attempt: Tt linear

Use TC1 dataset (1-encounter), Hadley SST

ARMA(1,1) for ut
ARMA(7,2) for vt

Focus on coefficient of trend in yt (1 parameter), various start

years ending in 2005

Start Year 1 SE t ratio p value

1880 1.18 0.93 1.28 0.20
1890 1.66 0.87 1.92 0.05
1900 2.19 0.84 2.62 0.01
1910 3.14 0.93 3.36 0.00
1920 2.68 0.99 2.71 0.01
1930 1.71 1.08 1.59 0.11
1940 2.87 1.97 1.46 0.14
1950 3.38 2.23 1.52 0.13
1960 6.41 2.51 2.55 0.01
1970 8.28 3.00 2.76 0.01
43
 Conclusions from this analysis

The results are not a big advance on fitting linear trends without
involving SST (cp. Vecchi-Knutson 2008)

We find significant linear trends from starting times near the

bottom of the AMO (e.g. 1900, 1960), but not from others
such as 1880, 1940.

No clear-cut conclusion of an anthropogenic trend.

44
 Second attempt: Tt based on a GCM

We use 20th Century runs from seven climate models (N.B. one
run from each, though in several cases there are multiple model
runs available)

Compute mean temperature over 060 oN and 7.575 oW as

proxy for North Atlantic SST

Apply spline smoother to remove local variation

Also plot mean of all 7 models

45
 SST 20th Century Model Projections

CCC MPI
295 CSIRO CCSM3
GFDL HADCM3
294 GISS Mean
293
Temp (K)

292

291

290

289

288
1880 1900 1920 1940 1960 1980 2000

Year

46
We fit same model using smoothed mean of seven GCMs as Tt

This time use AR(1) for ut, AR(5) for vt.

Start Year 1 SE t ratio p value

1880 1.39 0.50 2.78 0.005
1890 1.70 0.47 3.66 0.0003
1900 1.82 0.47 3.85 0.0001
1910 1.96 0.50 3.90 0.0001
1920 1.81 0.54 3.34 0.0008
1930 1.37 0.50 2.74 0.006
1940 1.98 0.65 3.05 0.002
1950 2.08 0.64 3.26 0.001
1960 2.43 0.64 3.81 0.0001
1970 2.56 0.52 4.95 106

Conclusion: a statistically significant trend (p < 0.01) from all

starting times
47
 Other Statistical Issues

This analysis assumes TCs and SSTs are normal after taking
square route transformations of SSTs ignores the fact that
TCs are counts

Possible alternative model based on Poisson counts

Existing literature (Elsner, Tsonis, Jagger) has used Bayesian

MCMC methods to create hierarchical model

We are looking at an alternative approach extending Daviss

and Dunsmuirs GLARMA modeling approach (Evangelou, in
progress)

48
 Conclusions

Simple bivariate time series analysis shows strong cross-correlation

between TCs and SSTs, but this could be a by-product of
AMO not directly indicative of anthropogenic trend

Simple linear trend analysis shows conflicting conclusions

results highly sensitive to start time, analysis ignores inter-
dependence between TCs and SSTs

Preferred analysis takes both time series and trend effects

into consideration

Correlating observed time series with trends from climate

models seems most effective approach. However, there are
still arguments about the data.

49