View source: R/SpecPrior-generators.R
Damp | R Documentation |
By default, the level term in a local level model and the trend term in a linear trend model are 'damped'. The level term or trend term are pulled towards 0. The amount of damping can be specified by the user, or can be estimated from the data.
Damp(coef = NULL, shape1 = 2, shape2 = 2, min = 0.8, max = 1)
coef |
A number between 0 and 1. |
shape1 |
A positive number. Defaults to 2. |
shape2 |
A positive number. Defaults to 2. |
min |
A number between 0 and 1. |
max |
A number between |
With a prior for a main effect, in a local level model, the level term has the form
level[j] ~ damp * level[j-1] + errorLevel[j]
,
and in a linear trend model, the trend term has the form
trend[j] ~ damp * trend[j-1] + errorTrend[j]
.
With a prior for an interaction, in a local level model, the level term has the form
level[k,l] ~ damp * level[k-1,l] + errorLevel[k,l]
.
and in linear trend model, the trend term has the form
trend[k,l] ~ damp * trend[k-1,l] + errorTrend[k,l]
(See the documentation for function DLM
for
an explanation of the k,l
subscripts.)
Values of for damp
are restricted to the range 0 <= damp <= 1
.
In linear trend models, including a damping term with a value near 1
typically results in more accurate forecasts (Hyndman et al 2008). There
are exceptions, however: for instance, damping of the trend for the time
effect is probably not appropriate in mortality forecasts for developed
countries (Oeppen and Vaupel 2002). Damping is also not necessary
appropriate in local level models.
The user can set the level of damping by providing a value for the
coef
argument. Alternatively, an appropriate value can be inferred
from the data, using a beta prior on the transformed parameter
(damp - min)/(max - min)
. The beta prior is specified
using parameters shape1
and shape1
. The default
values give a boundary-avoiding prior, confined to the range (min, max)
,
with min
defaulting to 0.8 and max
defaulting to 1
.
(See Gelman et al 2014, pp313-318, for a definition
of boundary-avoiding priors.) Setting shape1 = 1
and shape2 = 1
gives a uniform prior on the range (min, max)
.
Setting the damp
argument to NULL
in function
DLM
turns off damping.
An object of class Damp
.
Hyndman, R., Koehler, A. B., Ord, J. K., & Snyder, R. D. (2008). Forecasting with' exponential smoothing: the state space approach. Springer.
Oeppen, J., & Vaupel, J. W. (2002). Broken limits to life expectancy. Science, 296(5570), 1029-1031.
Damp
is used in calls to function DLM
## default
Damp()
## known value
Damp(coef = 0.95)
## estimate, but restrict to values between 0.85 and 0.95
Damp(min = 0.85, max = 0.95)
## uniform prior on the range (0, 1)
Damp(min = 0, max = 1, shape1 = 1, shape2 = 1)
## informative prior favouring high values, but
## not ruling out any value between 0 and 1
Damp(min = 0, max = 1, shape1 = 9, shape2 = 1)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.