EMA: The Exponential Moving Average and What It Costs

The exponential moving average is built so that recent prices influence it more than older ones. That earned it a reputation as the fast average — deserved, but with a caveat that usually goes unmentioned.

EMA: how the exponential moving average is calculated

Each new value combines the current price with weight k and the previous average with weight (1 − k), where k = 2 / (N + 1). Because of this recursion, old bars fade out gradually rather than dropping out abruptly as they do from a simple average's window. That removes the characteristic jolt an SMA makes when an old extreme leaves the lookback.

How often it signals on real data

We ran the rule "price above EMA(50)" over daily bars for 12 currency pairs — 518 bars each, about two years. We counted frequency rather than profit: how often the state changes and what share of days the indicator spends in signal.

• State changes over the period: 38 (median across pairs)
• Annualised: 20.1
• Share of days in signal: 54.2%
• Sample: 12 pairs, 518 daily bars

The measurement is reproducible: the script and data are in the site's repository and the conditions are described in the methodology. Frequency is not return: it says only how often you will have to make a decision.

What 'faster' actually means

At a turn, EMA reverses before an SMA of the same period — visible on any chart. But the measurement shows something less obvious: the number of crossings with price is the same for both, roughly twenty a year. EMA does not produce more signals; it produces the same signals slightly earlier. The difference is in timing, not in count.

Where the speed helps and where it hurts

It helps in trends: an earlier turn means less of the move given away. It hurts in ranges: the same sensitivity drags the line after every spike, and a crossover entry happens on noise. This is precisely why the simple moving average is not obsolete — it is steadier where steadiness matters more than speed.

Why the first EMA value comes from an SMA

The formula is recursive: today's value needs yesterday's. The very first bar has no predecessor, so something has to stand in — usually a simple average of the first N bars. That is a convention, not mathematics, and platforms implement it differently: some seed with an SMA, others with the first close. The discrepancy decays over three or four periods and is invisible on long history, but on short samples two platforms will show different values for the same EMA. When a backtest disagrees with a terminal, this is the first thing to check.

Choosing between EMA and SMA

By what costs your system more: a missed start or a false entry. A trend-following approach with rare trades benefits from EMA, because each trade is expensive and lateness eats the result. A system with frequent entries benefits from SMA, because noise is the costlier error. There is no universal answer, and searching for a period by brute force is a direct route to overfitting.

This material is for educational purposes and is not individual investment advice.