Paper SD16 Missing Data Values: Analyzing their Effects on Rainfall Forecasts Using PROC EXPAND and the SAS Time-Series Forecasting System

Missing values are a common problem faced in the analysis of time-series data. Many SAS® time-series procedures (PROC ARIMA, PROC VARMAX, etc.) are intolerant of missing values, particularly when these missing values are embedded in the time-series, rather than occurring at the beginning or end of the series. PROC EXPAND is designed to convert time series from one sampling interval or frequency to another and to interpolate missing values in the time series Daily rainfall data from the National Climatic Data Center (http://www.ncdc.noaa.gov/oa/ncdc.html) and the DBHYDRO database of the South Florida Water Management District (http://sonar.sfwmd.gov:7777/pls/dbhydro_pro_plsql/sh ow_dbkey_info.main_page) contain rainfall (and other) data with irregular patterns of missing values. This rainfall data has a marked seasonal pattern, with less prominent trend and cyclical components. The sensitivity of long-range and short-range climate forecasts, generated using the Time Series Forecasting System, to alternative options within PROC EXPAND to impute missing values will be examined. Forecasts generated using different imputation methods will be compared.