Time Series Forecasting in SAS
• What's main difference between time series and classic regression analysis?
In time series analysis, the target/dependent variable is usually a function of own past. If we want to apply proc reg,
then we need to apply lag function, that is, one of the independent variable would be : lag_y=lag(y)
Sometimes those lag effects get complex since we may need to lag different units,
e.g. lag 7 days to get weekly seasonality, lag 12 to get monthly seasonality; or even more complicated
when the difference (Y_t-Y_t-1) is also an independent variable.
To find out the appropriate lag effects and potential difference variables, we need to apply some procedures that is particular useful for time series,
including: Proc ARIMA, Proc Timeseries, Proc AutoReg, Proc FORECAST, Proc ESM(Exponential smoothing models), Proc SPECTRA(Spectral analysis)
• Preparing Data for Forecasting
◊ Read the raw transactional data.
◊ Convert the transactional data to time series data by
accumulating the data to equally spaced time points.
◊ Plot the data.
-- Identify data pathologies.
-- Suggest forecasting approaches.
◊ Address data pathologies.
-- Transform skewed data.
-- Impute missing values.
-- Detect unusual observations (outliers).
: We are using some stock data to demonstrate the data diagnostic process.
• Time Series in SAS
Slides presentation for
"Time Series in SAS"
• Partial correlation network analysis & Bootstrap Resampling
Pdf slides presentation for
"Partial correlation network analysis & Bootstrap Resampling in SAS"
• Tip to Success In Study
I hear, and I forget.
I see, and I remember.
I do, and I understand.
- Confucius (551 - 479 BC)
Back to Regression tutorial home
Back to Survival Analysis tutorial home
Back to Case Study tutorial home
Back to SAS tutorial home
Back to Homepage