Simple exponential smoothing (SES) forecasts a time series by weighting past observations, with exponentially decreasing influence for older data. Its smoothing parameter α lies between 0 and 1 and controls how quickly older observations lose influence. section 7.4.3 section 7.5
Use SES when the series has a stable level rather than systematic trend or seasonality:
- Choose SES for stationary series centered around a constant mean; use double exponential smoothing when a trend exists. section 7.4.4
- Fit the model to training data with statsmodels; the fitted model emphasizes recent observations and can produce forecasts. section 7.4.6
- Compared with ARIMA, SES generally needs less preprocessing because it does not require differencing or autoregressive and moving-average component selection. section 7.5
Fit SES with statsmodels
from statsmodels.tsa.holtwinters import SimpleExpSmoothing ses_model = SimpleExpSmoothing(train).fit() The resulting fitted model can generate forecasts from the recent patterns represented in the training data. section 7.4.6
