[R] Impute missing values within a time-series
warener.mike at gmail.com
Mon Apr 16 18:24:09 CEST 2007
I have to deal with a time series data and I'm looking for some
methodological help. The time series starts in January 1992 and until
April 1996 (52 observations) the values are ok (there is some trend
and seasonality in the data, but nothing special). From Mai 1996 to
June 1997 no observations were recorded, thus these 14 observations
are missing (NA). Starting from July 1997 until Dezember 1998
observations were recorded again. Compared to the data before the
missing values, it seems to me that a shift in level (downwards) had
occured. Furthermore, I have two (complete) regression variables that
are (weakly) positive correlated with the target variable (correlation
~ 0.2 and 0.6).
I want to impute the missing values in the middle of the time series
and want to ask for the "standard" approach to do this.
Should I split the time series just before the first missing value,
model the first part of the time series by arima methods and predict
the missing values? If so, how can I model the level-change (which
should be at t=53, the first occurence of a missing value).
Should I split the time series after the last missing value and
somehow "impute the missing values" backwards? Is this possible?
Shouldn't I treat the data as time series but use a robust regression
approach to impute the missing values?
To establish a relationship to R (and not just asking a methodology
question): What would be the preferred packages in R to achieve my
task? Robust regression with lqs? Is there a package for regArima
Any help is greatly appreciated….
More information about the R-help