How can you use exponential smoothing to improve your forecasting?

1 What is exponential smoothing?

Exponential smoothing is a family of methods that use weighted averages of past observations to estimate the current state of a variable. The weights decrease exponentially as the observations get older, giving more importance to the recent data and less to the distant data. Exponential smoothing can handle various patterns of data, such as constant, linear, seasonal, or cyclical. Depending on the complexity of the data, you can use different types of exponential smoothing, such as simple, double, or triple.

2 How does exponential smoothing work?

The basic idea of exponential smoothing is to update the estimate of the current level of the variable by combining the previous estimate and the latest observation. The degree of smoothing is controlled by a parameter called the smoothing constant, which ranges from 0 to 1. A higher smoothing constant gives more weight to the latest observation, while a lower smoothing constant gives more weight to the previous estimate. The formula for simple exponential smoothing is: Level_t = alpha * Observation_t + (1 - alpha) * Level_t-1 where Level_t is the estimate of the current level of the variable, Observation_t is the latest observation, and alpha is the smoothing constant.

3 How can you choose the smoothing constant?

The smoothing constant is a key parameter that affects the performance of exponential smoothing. If the smoothing constant is too high, the estimate will be too sensitive to the latest observation and may overreact to noise or outliers. If the smoothing constant is too low, the estimate will be too slow to adjust to changes in the level or trend of the variable. Therefore, you need to choose the smoothing constant that minimizes the forecasting error, which is the difference between the actual and predicted values. One way to do this is to use a trial-and-error approach, where you test different values of the smoothing constant and compare the results. Another way is to use an optimization algorithm, such as the Solver in Excel, that can find the optimal value automatically.

4 How can you forecast with exponential smoothing?

Once you have estimated the current level of the variable using exponential smoothing, you can use it to forecast the future values. The simplest way to do this is to assume that the level will remain constant and use the same estimate for all future periods. This is called a naive forecast and it works well when the data has no trend or seasonality. However, if the data has a trend or seasonality, you need to use more advanced methods, such as double or triple exponential smoothing, that can capture these patterns and update the estimates of the trend and seasonality along with the level. The formulas for double and triple exponential smoothing are: Trend_t = beta * (Level_t - Level_t-1) + (1 - beta) * Trend_t-1 Seasonality_t = gamma * (Observation_t - Level_t) + (1 - gamma) * Seasonality_t-m Forecast_t+m = (Level_t + m * Trend_t) * Seasonality_t-m+1 where Trend_t is the estimate of the current trend of the variable, Seasonality_t is the estimate of the current seasonality of the variable, Forecast_t+m is the forecast for m periods ahead, beta and gamma are the smoothing constants for the trend and seasonality, and m is the length of the seasonal cycle.

5 What are the benefits and limitations of exponential smoothing?

Exponential smoothing has several advantages and disadvantages that you should be aware of before using it for your forecasting. It is easy to understand and implement, as it only requires a few parameters and calculations, as well as being robust and flexible, able to handle different types of data and adapt to changes in the level, trend, or seasonality. Additionally, it is efficient and scalable, using only the most recent data and not needing to store or revise the entire history of observations. However, it is based on subjective choices when selecting the appropriate type and value of the smoothing constants and the seasonal cycle. It is also sensitive to outliers and noise, distorting the estimates of the level, trend, or seasonality. Moreover, it is not suitable for long-term forecasting due to its assumption that patterns will continue in the same way without accounting for structural breaks or external factors.

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?