fkf {FKF} | R Documentation |
Fast Kalman filter
Description
This function allows for fast and flexible Kalman filtering. Both, the
measurement and transition equation may be multivariate and parameters
are allowed to be time-varying. In addition “NA”-values in the
observations are supported. fkf
wraps the C
-function
FKF
which fully relies on linear algebra subroutines contained
in BLAS and LAPACK.
Usage
fkf(a0, P0, dt, ct, Tt, Zt, HHt, GGt, yt)
Arguments
a0 |
A |
P0 |
A |
dt |
A |
ct |
A |
Tt |
An |
Zt |
An |
HHt |
An |
GGt |
An |
yt |
A |
Details
State space form
The following notation is closest to the one of Koopman et al.
The state space model is represented by the transition equation and
the measurement equation. Let be the dimension of the state
variable,
be the dimension of the observations, and
the number of observations. The transition equation and the
measurement equation are given by
where and
are iid
and iid
,
respectively, and
denotes the state
variable. The parameters admit the following dimensions:
|
|
|
|
| |
|
|
|
|
|
Note that fkf
takes as input HHt
and GGt
which
corresponds to and
.
Iteration:
The filter iterations are implemented using the expected values
and variances
of the state in the following way
(for the case of no NA's):
Initialisation: Set with
and
Updating equations:
Prediction equations:
Next iteration: Set and goto “Updating equations”.
NA-values:
NA-values in the observation matrix yt
are supported. If
particular observations yt[,i]
contain NAs, the NA-values are
removed and the measurement equation is adjusted accordingly. When
the full vector yt[,i]
is missing the Kalman filter reduces to
a prediction step.
Parameters:
The parameters can either be constant or deterministic
time-varying. Assume the number of observations is
(i.e.
). Then, the parameters admit the following
classes and dimensions:
dt | either a (time-varying) or a (constant) matrix. |
Tt | either a or a array. |
HHt | either a or a array. |
ct | either a or a matrix. |
Zt | either a or a array. |
GGt | either a or a array. |
yt | a matrix.
|
BLAS and LAPACK routines used:
The R function fkf
basically wraps the C
-function
FKF
, which entirely relies on linear algebra subroutines
provided by BLAS and LAPACK. The following functions are used:
BLAS: | dcopy , dgemm , daxpy . |
LAPACK: | dpotri , dpotrf .
|
FKF
is called through the .Call
interface. Internally,
FKF
extracts the dimensions, allocates memory, and initializes
the R-objects to be returned. FKF
subsequently calls
cfkf
which performs the Kalman filtering.
The only critical part is to compute the inverse of
and the determinant of
. If the inverse can not be
computed, the filter stops and returns the corresponding message in
status
(see Value). If the computation of the
determinant fails, the filter will continue, but the log-likelihood
(element logLik
) will be “NA”.
The inverse is computed in two steps:
First, the Cholesky factorization of is
calculated by
dpotrf
. Second, dpotri
calculates the
inverse based on the output of dpotrf
.
The determinant of is computed using again the
Cholesky decomposition.
The first element of both at
and Pt
is filled with the
function arguments a0
and P0
, and the last, i.e. the (n +
1)-th, element of at
and Pt
contains the predictions for the next time step.
Value
An S3-object of class “fkf”, which is a list with the following elements:
att | A -matrix containing the filtered state variables, i.e. att[,t] = . |
at | A -matrix containing the predicted state variables, i.e. at[,t] = . |
Ptt | A -array containing the variance of att , i.e. Ptt[,,t] = . |
Pt | A -array containing the variances of at , i.e. Pt[,,t] = . |
vt | A -matrix of the prediction errors i.e. vt[,t] = . |
Ft | A -array which contains the variances of vt , i.e. Ft[,,t] = . |
Kt | A -array containing the “Kalman gain” i.e. Kt[,,t] = . |
logLik | The log-likelihood. |
status | A vector which contains the status of LAPACK's dpotri and dpotrf . means successful exit. |
sys.time | The time elapsed as an object of class “proc_time”. |
References
Harvey, Andrew C. (1990). Forecasting, Structural Time Series Models and the Kalman Filter. Cambridge University Press.
Hamilton, James D. (1994). Time Series Analysis. Princeton University Press.
Koopman, S. J., Shephard, N., Doornik, J. A. (1999). Statistical algorithms for models in state space using SsfPack 2.2. Econometrics Journal, Royal Economic Society, vol. 2(1), pages 107-160.
See Also
plot
to visualize and analyze fkf
-objects, KalmanRun
from the stats package, function dlmFilter
from package dlm
.
Examples
## <--------------------------------------------------------------------------->
## Example: Local level model for the Nile's annual flow.
## <--------------------------------------------------------------------------->
## Transition equation:
## alpha[t+1] = alpha[t] + eta[t], eta[t] ~ N(0, HHt)
## Measurement equation:
## y[t] = alpha[t] + eps[t], eps[t] ~ N(0, GGt)
y <- Nile
y[c(3, 10)] <- NA # NA values can be handled
## Set constant parameters:
dt <- ct <- matrix(0)
Zt <- Tt <- matrix(1)
a0 <- y[1] # Estimation of the first year flow
P0 <- matrix(100) # Variance of 'a0'
## Estimate parameters:
fit.fkf <- optim(c(HHt = var(y, na.rm = TRUE) * .5,
GGt = var(y, na.rm = TRUE) * .5),
fn = function(par, ...)
-fkf(HHt = matrix(par[1]), GGt = matrix(par[2]), ...)$logLik,
yt = rbind(y), a0 = a0, P0 = P0, dt = dt, ct = ct,
Zt = Zt, Tt = Tt)
## Filter Nile data with estimated parameters:
fkf.obj <- fkf(a0, P0, dt, ct, Tt, Zt, HHt = matrix(fit.fkf$par[1]),
GGt = matrix(fit.fkf$par[2]), yt = rbind(y))
## Compare with the stats' structural time series implementation:
fit.stats <- StructTS(y, type = "level")
fit.fkf$par
fit.stats$coef
## Plot the flow data together with fitted local levels:
plot(y, main = "Nile flow")
lines(fitted(fit.stats), col = "green")
lines(ts(fkf.obj$att[1, ], start = start(y), frequency = frequency(y)), col = "blue")
legend("top", c("Nile flow data", "Local level (StructTS)", "Local level (fkf)"),
col = c("black", "green", "blue"), lty = 1)