Put a 20-period simple moving average on any chart and watch what happens when a large candle drops out of the back of the window. The line jumps, sometimes hard, even though the newest bar barely moved. Nothing happened in the market at that moment. Something happened to the arithmetic. That single behaviour explains most of what people get wrong about moving averages.
What the line is actually computing
A simple moving average adds the closing prices of the last N bars and divides by N. Each bar in the window counts equally. When bar N+1 arrives, the oldest close is removed and the newest is added, so the line responds to two events at once: what just printed, and what just expired.
An exponential moving average takes a different route. It keeps a running value and blends each new close into it with a fixed weight, commonly expressed as 2/(N+1). Recent prices carry more influence and old prices decay away rather than falling off a cliff. There is no drop-out jump, because nothing is ever fully discarded.
Both produce a smoothed version of price. Neither predicts. The average is a summary of what already happened, which is why every moving average signal arrives after the move it describes has started.
| Type | Weighting | Behaviour | Typical use |
|---|---|---|---|
| SMA | Equal across the window | Smoothest line, slowest to turn, jumps when a large old bar expires | Long-horizon trend reference, such as a 200-period line |
| EMA | Exponentially decaying toward the past | Turns earlier, hugs price more closely, more false turns | Intraday trend filters and faster crossover systems |
| WMA | Linearly decaying, oldest bar weight zero | Between the two, with a hard cut-off at the window edge | Less common, used where a defined lookback matters |
Period selection is a trade-off, not an optimisation
Shorter windows react faster and produce more signals. Longer windows react slower and produce fewer. That is the entire relationship, and no period escapes it. The 50 and 200 are popular because they are popular: enough participants watch them that price sometimes reacts around them, which is a self-referential reason rather than a mathematical one, and it is still a real reason.
The failure mode is re-tuning. A trader has a poor month, drops the 21 for the 34, has another poor month, tries the 13. Each change is fitted to noise that has already passed. If a period must be chosen, tie it to the swing length you actually trade on the timeframe you actually trade, write it down, and stop touching it. A parameter you change monthly is a parameter you never tested.
Why crossovers arrive late by construction
A fast average crossing a slow one is a statement that recent prices have moved far enough to overcome the slower line's inertia. By the time that is arithmetically true, a meaningful part of the move has printed. This is not a flaw to be engineered away. It is what a lagging summary does.
Crossovers also fail badly sideways. In a range, the two lines braid around each other and every crossing is a loss. Systems built purely on crossings tend to give back in chop what they earn in trend, which is the well-documented shape of trend following: many small losses, occasional large gains, and long stretches where the equity curve goes nowhere. Anyone running that shape needs the stomach and the risk rules to match, because leveraged trading can lose more than the account holds.
No moving average configuration removes risk. A signal tells you the average has changed direction, nothing about how far price will travel next. Stops and position size decide outcomes; the line only decides when you look.
The uses that hold up better
Three applications survive contact with live markets more often than crossover systems do. The first is direction filtering: take entries in one direction only while price sits on one side of a slow average, and take the other direction only when it sits on the other. That does not generate trades, it removes them, which is usually the more valuable function.
The second is dynamic reference. In a strong trend, pullbacks often stall near a moving average, and traders use it as a moving version of a support or resistance level. It works until it does not, and it works better on instruments and timeframes where the participants watching it are numerous.
The third is as an input to other calculations. The MACD line is nothing more than the difference between two exponential averages, with a third average of that difference drawn on top. Bollinger Bands are an average with a volatility envelope around it. Understanding the underlying average makes those derivatives far less mysterious.
Where averages break
Gaps are the obvious case. A weekend gap or a news gap moves price a long way in zero bars, and the average takes many bars to catch up, so anything derived from it is stale exactly when it matters. Illiquid instruments and thin sessions produce the same effect in miniature.
The subtler break is the source data. An average computed on closing prices from one broker's feed will differ slightly from another's, because CFD feeds are aggregated and marked. On a 200-period daily line the difference is cosmetic. On a 5-period intraday line used to trigger automated entries, it is not, and it is one of the reasons a strategy that tested well on historical data can behave differently live. Averages are exact arithmetic performed on approximate inputs, and the approximation is the part people forget.
"When someone asks me which moving average is best, I ask them what decision the line is supposed to make for them. Half the time there isn't one, and they just wanted the chart to look busier."
— Alex Onta, Executive Director, SINGUARD
Key Takeaways
- An SMA reacts to both the newest bar and the bar leaving its window, which is why it can jump when nothing has happened.
- An EMA weights recent prices more and never discards old ones, so it turns earlier and whipsaws earlier for the same reason.
- Crossovers lag by construction and fail in ranges, so most durable uses are filters and references rather than standalone entry signals.
- Re-tuning the period after every losing stretch fits noise; fix it to the swing length you trade and leave it alone.
Frequently Asked Questions
What is the difference between an SMA and an EMA?
A simple moving average gives every bar in its window the same weight, then drops the oldest bar entirely when a new one arrives. An exponential moving average weights recent bars more heavily and never fully forgets older data. The EMA therefore turns sooner after a change in price, and also whipsaws sooner when the change was noise.
Which moving average period should I use?
There is no correct period, only a trade-off between lag and false signals. Shorter windows react faster and generate more of both good and bad turns. A practical approach is to fix the period to the swing length you actually trade on your chosen timeframe, then leave it alone, rather than re-tuning it after every losing week.
Do moving average crossovers work?
A crossover is a lagging confirmation that two averages have already changed order, so it will never catch a turn near its extreme and it produces frequent false signals in sideways markets. Traders who use crossovers usually treat them as a trend filter for other entries rather than as a standalone signal, and they still need stops and position sizing because trading carries a high risk of loss.