2
$\begingroup$

If I have missing values in a time series that has 40 quarters (ten cycles or ten years) of data, what is the best SAS procedure to use to impute the missing values?

Part 2: I have 390 series (40 quarters each) that follow similar patterns -- most have missing data points (2-3 each), how do I make use of the other 390 series to help impute missing values in any one series? What SAS procedure would I use for that? In the end I want a complete set of 15600 data points.

Improve this question
$\endgroup$
1
  • $\begingroup$ I'm not an expert in the topic, but I think some more details about the nature of your data could be necessary to peek proper way of imputation. $\endgroup$
    – user88
    Commented Jun 24, 2011 at 16:30

2 Answers 2

Reset to default
1
$\begingroup$

For each of your 390 series you have 40 readings. Simply automatically identify an ARIMA Model for each series enabling Intervention Detection to provide estimates of the missing values. If there are 3 missing values the software will identify three Pulses which will yield an estimate if the missing values. The problem is that SAS assumes an ARIMA MODEL first and then estimates the Pulses. In truth the identified ARIMA Model may be flawed by the missing values. An alternative procedure which we have used is to identify the missing values first and then iterate to the ARIMA Model. The whole idea is to run BOTH PROCEDURES to determine which one is optimal for each of your 390 series.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Regarding the nature of the data: The data come from crop storage facilities where the annual harvest causes a cyclic spike in December with usage during the year till the next harvest. Observation are quarterly but the cycles are annual. $\endgroup$
    – Jacob
    Commented Jun 27, 2011 at 15:57
0
$\begingroup$

You can use proc expand to impute missing values, but I'm not sure if it can do what you're asking in Part 2.

Improve this answer
$\endgroup$
1
  • $\begingroup$ It seems that the cubic spline used in proc expand is inadequate due to the Jagged nature of the cyclic pattern in the data. With December observations revealing the peak storage levels after harvest, the September observations are usually the low point with steady usage occurring throughout the year. $\endgroup$
    – Jacob
    Commented Jun 27, 2011 at 16:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.