estimate missing data of numerous parallel correlated time series

Hello, I have a problem and a long time ago couldn't find any method that could automatically solve it.

Shortly: Given 500 time series of an environmental phenomena, that more or less correlating to each other. The beginning and ending of the observations is different, and there are a lot of missing data even in the continuous measurements. Since the data in timeseries are monthly averages it shows yearly periodicity that I also want to handle.

Is there any algorithm in Matlab which could handle all time series together and estimate missing data automatically?

Thank you in advance,

Zsolt

[Simon Woodward]

Hi Zsolt

I have a similar problem - I have about 60 time series of well water levels. The levels are highly correlated but there are missing sections in each series. Also, some of the series are only sample weekly or fortnightly, while others are daily. I think some kind of regression approach is suitable. There is a field of research called "imputation" which is used by social scientists to fill in missing social data, but it seems that all of the codes are in statistical packages (R, SAS etc) , and I have not been able to find a Matlab code.

[Simon Woodward]

The work of Gary King at Harvard seems to be close:

http://dash.harvard.edu/bitstream/handle/1/4100248/Honaker_MissingValues.pdf?sequence=2