Moving Average

August 25, 2024 Statistics

A moving average smooths a financial time series by recalculating an average over a rolling window, with lag and responsiveness determined by its weights.

On this page

A moving average smooths a time series by recalculating an average as the observation window moves forward. In finance, moving averages are used to summarize prices, returns, volumes, spreads, revenue, cash flows, and other time-ordered data.

Smoothing can make trend or seasonality easier to see, but it does not predict turning points or establish a profitable trading signal. Every moving average trades responsiveness for noise reduction and necessarily uses historical data.

Key Takeaways

A simple moving average gives equal weight to observations inside a rolling window.
An exponential moving average gives progressively more weight to recent observations.
Short windows respond faster but retain more variation; long windows are smoother but lag more.
Results depend on the data field, frequency, adjustment method, window, and alignment date.
A chart crossover is a rule based on past prices, not evidence of future return by itself.
A moving-average smoother is different from the moving-average error model used in ARMA or ARIMA time-series analysis.

Simple Moving Average Formula

For a trailing window of (n) observations, the simple moving average at time (t) is:

\operatorname{SMA}_{t,n}=\frac{1}{n}\sum_{i=0}^{n-1}X_{t-i}

When a new observation arrives, the oldest observation leaves the window. A five-day SMA therefore always uses five observations after the initial window is available.

The first (n-1) trailing values are undefined unless the analyst deliberately uses a shorter startup window. Filling them with zeros or carrying another value changes the statistic.

Exponential Moving Average Formula

An exponential moving average (EMA) updates recursively:

\operatorname{EMA}_t=\alpha X_t+(1-\alpha)\operatorname{EMA}_{t-1},\qquad 0<\alpha\leq1

One common market convention links an (n)-period label to:

\alpha=\frac{2}{n+1}

That convention is not universal. Software can use different initialization rules, adjusted calculations, or direct normalized weights. Analysts should verify the implementation before comparing results.

Worked Example: SMA vs. EMA

Assume five hypothetical closing prices: $100, $102, $101, $105, and $107.

The three-period SMA on day 5 is:

\operatorname{SMA}_{5,3}=\frac{101+105+107}{3}=104.33

For an EMA with (\alpha=0.50), initialized at the first price, the sequence is $100, $101, $101, $103, and $105. The day-5 EMA is:

\operatorname{EMA}_5=0.50(107)+0.50(103)=105

Day	Closing price	3-period SMA	EMA, (\alpha=0.50)
1	$100	Not available	$100.00
2	$102	Not available	$101.00
3	$101	$101.00	$101.00
4	$105	$102.67	$103.00
5	$107	$104.33	$105.00

Five-day price series compared with a three-period simple moving average and an exponential moving average.

The EMA responds faster to the late price increase because it places more weight on recent observations. Neither line proves that day 6 will rise.

Main Moving-Average Choices

Method	Weighting	Best suited to	Main limitation
Simple moving average	Equal within the window	Transparent smoothing and rolling summaries	Drops an observation abruptly when it leaves the window
Exponential moving average	Geometrically declining	Faster response and recursive updating	Depends on smoothing factor and initialization
Linearly weighted moving average	Explicit weights, often larger for recent data	Custom recency emphasis	Weight choice is subjective
Centered moving average	Window placed around the target period	Historical seasonal or trend decomposition	Uses future observations and cannot be a real-time signal at the edge
Moving median	Median within the window	Smoothing with less sensitivity to isolated outliers	Can discard information about magnitude

“Moving average” can refer to different calculations. A simple moving average of prices, an EMA of returns, and a centered average used for seasonal decomposition are not interchangeable.

Window Length and Lag

A longer window averages more observations and usually produces a smoother line. It also reacts more slowly when the underlying process changes. A shorter window adapts faster but may follow temporary noise.

There is no universally correct 10-, 20-, 50-, 100-, or 200-period window. Those labels also depend on frequency: 20 trading days, 20 calendar days, and 20 months describe very different horizons.

Choose a window based on the decision horizon, data frequency, seasonality, and validation evidence. Selecting the window that would have produced the best historical trade creates data-mining risk.

Finance Uses

Trend Description

A moving average can show whether a price, margin, spread, or operating metric has generally risen or fallen over the selected window. It describes the selected historical horizon rather than an intrinsic trend.

Monitoring and Control

Treasury, finance, and risk teams can smooth daily cash balances, transaction volumes, settlement failures, or pricing exceptions to identify sustained changes. A smoother can also delay recognition of a sudden problem, so raw observations should remain visible.

Forecast Baseline

A trailing average can serve as a simple benchmark forecast. More complex models should demonstrate improved performance over such a baseline on later data.

Technical Indicators

Moving averages are inputs to indicators such as Bollinger Bands and moving-average crossovers. A crossover can be precisely defined and back-tested, but historical results must address costs, look-ahead bias, parameter selection, market regimes, and out-of-sample performance.

Smoother vs. Moving-Average Time-Series Model

A rolling SMA averages observed values. In an MA((q)) time-series model, the current value depends on current and lagged innovations or error terms. The shared name causes confusion, but the calculations and purposes differ.

Term	Uses	Example
Rolling moving average	Observed values in a moving window	Average of the last 20 closing prices
Exponential smoothing	Current value and prior smoothed estimate	Recursive estimate of current level
MA((q)) model	Current and lagged unobserved innovations	Time-series model for serial error structure

Common Mistakes

Using unadjusted prices across stock splits or distributions.
Mixing closing prices from different time zones or market sessions.
Treating a centered moving average as if it were available in real time.
Comparing indicators with different initialization or missing-value rules.
Selecting a window after observing which one generated the best historical result.
Assuming a smoothed line removes risk rather than hides short-term variation.
Treating a crossover as causal evidence or a guaranteed signal.
Confusing price smoothing with an ARIMA moving-average error term.
Ignoring trading costs, taxes, spreads, and turnover in a strategy test.

How to Evaluate a Moving Average

Identify the exact data field, adjustment method, frequency, and timestamp.
Confirm whether the calculation is trailing, centered, or forward-looking.
Document the window length, weights, smoothing factor, and initialization.
Compare the smoothed series with raw observations.
Test sensitivity to reasonable windows and market regimes.
Preserve chronological order and avoid future-data leakage.
For forecasting or trading claims, compare with a simple benchmark after costs on later data.
Explain what decision changes when the moving average changes.

Authoritative Sources

NIST: Moving Average and Smoothing Techniques: explanation of averaging and exponential-smoothing methods.
NIST: Centered Moving Average: alignment of odd and even windows and limitations when trend is present.
NIST: Common Univariate Time-Series Approaches: distinction between autoregressive and moving-average time-series models.

Time Series Analysis: Analysis of chronological dependence, trend, seasonality, volatility, and forecasting.
Aggregation: Combination of financial records or exposures under defined grouping rules.
Bollinger Bands: Indicator combining a moving average with volatility-based bands.
Trend Line: Chart line used to describe directional price structure without rolling averaging.
Regression Analysis: Estimation of conditional relationships between an outcome and explanatory variables.

FAQs

What is a moving average used for?

It smooths a time series so an analyst can summarize recent level or direction, monitor changes, or establish a simple forecast benchmark. It does not predict a turning point by itself.

What is the difference between an SMA and an EMA?

An SMA gives equal weight to observations in a fixed window. An EMA updates recursively and gives more weight to recent observations, so it generally reacts faster to new data.

Does a moving-average crossover predict returns?

Not reliably by definition. A crossover is a historical-price rule whose usefulness depends on evidence after parameter selection, costs, timing, and out-of-sample testing. It cannot guarantee a future return.

This article provides general financial and statistical education. It does not provide a trading signal, market forecast, or personalized investment advice.

Decile