Evaluating Time-Series Model Performance

A financial analyst is evaluating a time-series model used to forecast monthly sales of a product. The analyst has collected data over the last five years, resulting in a dataset with 60 observations. After fitting an autoregressive integrated moving average (ARIMA) model to the data, the analyst computes various metrics to assess the model’s accuracy. The metrics include the Mean Absolute Error (MAE), the Mean Squared Error (MSE), and the Root Mean Squared Error (RMSE).

At the conclusion of the evaluation, the analyst determines the MAE to be 200 units, the MSE to be 60,000, and the RMSE to be approximately 245 units. The analyst weighs these metrics as follows: MAE is favored because it is easy to interpret, MSE is useful because it emphasizes larger errors due to squaring the differences, and RMSE is useful when he needs a measure in the same unit as the data. The analyst is unsure which metric would provide the most robust indication of model performance for future sales predictions.