How do you choose between exponential smoothing and ARIMA models?

1 What is exponential smoothing?

Exponential smoothing is a technique that assigns different weights to past observations, giving more weight to the most recent ones. It produces a smoothed series that captures the general trend and level of the data, while ignoring random fluctuations. Exponential smoothing is easy to implement, interpret, and update, and can handle missing values and outliers. However, it has some limitations, such as assuming a constant variance and error distribution, and requiring a large number of parameters for complex models.

2 What is ARIMA?

ARIMA stands for AutoRegressive Integrated Moving Average, and it is a class of models that describe the correlations and patterns in a time series. ARIMA models can capture various features of the data, such as seasonality, cycles, trends, and shocks, by using different combinations of autoregressive, differencing, and moving average terms. ARIMA models are flexible, robust, and widely applicable, and can provide forecasts and confidence intervals. However, they can also be difficult to specify, estimate, and diagnose, and may require transformations and adjustments to fit the data.

3 How to compare exponential smoothing and ARIMA?

Comparing exponential smoothing and ARIMA models requires criteria that measure their performance and suitability for the data and problem. Common criteria include accuracy, complexity, parsimony, and stability. Accuracy can be evaluated with metrics such as MAE, RMSE, or MAPE; complexity with AIC, BIC, or AICc; parsimony with Occam's razor; and stability with cross-validation, sensitivity analysis, or bootstrap. With these criteria, you can choose the model that best fits the data and minimizes information loss while being robust and reliable.

4 How to choose between exponential smoothing and ARIMA?

When choosing between exponential smoothing and ARIMA models, there is no definitive answer, as it depends on the data characteristics, the purpose of forecasting, and the analyst's preferences. Generally speaking, if the data is noisy, irregular, or has missing values or outliers, exponential smoothing may be more suitable. Additionally, if the data has a clear and stable trend, level, and seasonality, exponential smoothing may be a good choice. On the other hand, if the data is complex, non-stationary, or has multiple patterns and features, ARIMA may be more suitable. Moreover, ARIMA may be more appropriate if data requires transformations, differencing, or adjustments. Lastly, if accuracy, complexity, parsimony and stability criteria are similar or inconclusive, you may use other factors such as ease of implementation or software availability to make a decision.

5 How to apply exponential smoothing and ARIMA in practice?

To apply exponential smoothing and ARIMA models in practice, you need to begin by exploring and preparing the data. This means checking the quality, frequency, and distribution of the data, and performing any necessary cleaning, transformation, or adjustment. After that, you need to select the appropriate models and parameters for exponential smoothing and ARIMA and estimate them using historical data. You can use methods such as grid search, auto-selection, or trial and error to find the best models. Then you need to evaluate and compare the models using criteria such as in-sample fit, out-of-sample forecast, or forecast error. Finally, you can select the best model based on the results of the evaluation and comparison and use it to generate forecasts and confidence intervals for the future periods. Additionally, you should update the model periodically with new data and information.

6 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?