Moving Average vs. Exponential Smoothing for Forecasting

Moving average vs exponential smoothing is the first choice most forecasters face. Both methods smooth out noise in a series to estimate its underlying level, and both are simple enough to build in a spreadsheet. The difference lies in how they treat old data. A moving average gives equal weight to the last few periods and none to anything older. Exponential smoothing gives the most weight to the latest period and gradually less to older ones, never dropping them completely. This guide explains the formulas, works through both on the same data, shows how the smoothing constant changes the result and offers guidance on when to use each. It is part of the forecasting methods in budgeting and forecasting and complements time series forecasting for finance teams.

The formulas

Simple moving average

Forecast(t+1) = (A(t) + A(t-1) + ... + A(t-n+1)) / n

where n is the number of periods in the window. A three-month moving average averages the last three actuals.

Simple exponential smoothing

Forecast(t+1) = alpha × A(t) + (1 - alpha) × Forecast(t)

where alpha (the smoothing constant) is between 0 and 1. A high alpha reacts quickly to the latest actual; a low alpha changes slowly.

How the weights differ

Both methods assign weights to past data. For a three-period moving average, the weights are one-third each on the last three periods and zero beyond. For exponential smoothing with alpha = 0.3, the weights are:

The weights sum to 1. Exponential smoothing never fully forgets, but old data fades. A moving average has a sharp cliff: an unusual month drops out of the forecast suddenly when it leaves the window, which can cause a jump.

A worked comparison

Use eight months of sales (in thousands of dollars): 100, 108, 104, 112, 118, 115, 124, 128. The series trends upward. Forecast months 4 to 8 and compare errors.

Three-month moving average

Mean absolute error = 39.7 / 5 = 7.9.

Exponential smoothing, alpha = 0.5

Start with the first forecast equal to the first actual (100) and update each period.

Mean absolute error for months 4 to 8 = (8 + 10 + 2 + 10 + 9) / 5 = 39 / 5 = 7.8.

Exponential smoothing, alpha = 0.9

The same process with alpha = 0.9 gives forecasts for months 4 to 8 of about 104.3, 111.2, 117.3, 115.2 and 123.1, with absolute errors of 7.7, 6.8, 2.3, 8.8 and 4.9. The mean absolute error is about 6.1.

Summary

On a rising series, methods that react faster do better. As alpha approaches 1, exponential smoothing converges on the naive forecast. The simple methods all lag the trend, because they estimate a level and do not model growth. None of them is a good fit for a trending series unless the trend is handled separately.

Choosing alpha

Choose alpha by testing several values on a hold-out period and picking the one with the lowest error. Excel's Solver can do this for you. Do not choose alpha because it produced the best fit to the whole history; test it on data not used to fit it.

Handling trend and seasonality

Simple versions of both methods assume no trend and no seasonality. Extensions:

A common practical approach is to deseasonalize the data with seasonal indexes, smooth the adjusted series and re-apply the indexes to the forecast.

Moving average vs. exponential smoothing: when to use which

For many finance teams, exponential smoothing with a tested alpha is the better default, because it adapts and needs only the last forecast and actual. A moving average is useful for visualizing trends and for removing noise in dashboards.

Practical tips

1. Compare with a naive benchmark every time. See financial forecasting.
2. Clean outliers before smoothing. A one-off order inflates a moving average for n periods and fades slowly in exponential smoothing.
3. Do not chase noise. A high alpha in a noisy series amplifies errors.
4. Track bias. Smoothing methods lag trends, creating consistent under-forecasts in growth or over-forecasts in decline. See forecast bias in financial planning.
5. Review the parameters periodically. An alpha chosen two years ago may no longer fit.

Use in demand planning

In product businesses, exponential smoothing is a standard baseline for item-level demand forecasts. Planners then add events and review exceptions. See demand planning and inventory optimization.

Linking to the economics

A smoother forecast of units does not reflect changes in price or mix. Forecast revenue and margin using units together with unit economics. See unit economics.

Common mistakes

• Choosing the window or alpha based on fit to history only.
• Using the simple methods on strongly trending or seasonal data.
• Failing to adjust outliers.
• Not measuring against a naive forecast.
• Setting alpha once and never testing again.

Related guides

To bring in explanatory variables, read regression analysis for financial forecasting in Excel. For a driver-based alternative, see driver-based forecasting.

FAQs

What is the difference between a moving average and exponential smoothing?

A moving average weights recent periods equally and ignores older ones; exponential smoothing weights recent periods more and older ones progressively less.

What is the smoothing constant alpha?

A number between 0 and 1 that controls how quickly exponential smoothing reacts to new data. Higher values react faster.

Which is more accurate?

It depends on the data. Exponential smoothing is usually more adaptable, but test both against a naive benchmark.

Can they handle seasonality?

Not in their simple forms. Use seasonal indexes or Holt-Winters.

How do I pick alpha in Excel?

Test several values on held-out data, or use Solver to minimize error, and confirm on a later period.