1. Perform a decomposition analysis of your time series, try both multiplicative and additive model structures. Models applied should be selected based on your time series characteristics. Don't just use available options on the decomposition menu, but also use other tools, like trend analysis, other smoothing techniques, and forecast techniques when necessary.
2. Calculate forecast errors (residuals) for each model. Find MAD, MSE, RMSE, MAPE, and MPE for each model and discuss model accuracies.
3. Analyze the residuals using appropriate plots. (i.e. create a time series plot, graphical summary, box plot, etc. to check the statistical assumptions for them, and analyze its correlogram). What do the residual autocorrelations suggest for these data? Explain.
4. Choose the best forecast model among your alternatives. Use it to forecast the value of your variable for the next four periods into the future.