Exponential moving average

After each time interval, shown as a candle on the chart, Petr records its last price. Instead of adding all the values again, he can take the previous EMA and move it part of the distance towards that price. This is how Fidelity explains the method. TradingView expresses the same calculation as a weighted new price plus a weighted previous EMA. This is called recursion: the next step uses the result of the last one. [1] [2]

A coefficient, a number setting the new price's weight, determines that fraction. TradingView gives the common relationship 2 / (N + 1), where N is the selected length. A higher N gives the new input less weight and slows the adjustment. Petr therefore needs more than the label EMA: he needs the length setting and the coefficient convention. Length controls how the weights are assigned; the example below shows one step in numbers. [3]

At the very beginning, Petr has no previous EMA. Fidelity describes starting with an SMA, the ordinary average of the initial prices. Choosing that starting result is called initialization. If another tool starts differently or loads a different amount of history, the lines can differ even with identical subsequent prices. The difference gradually fades with further calculations; the name EMA alone therefore does not guarantee identical values. [1]

The previous EMA already carries older price information. Each new step reduces its influence again, as TradingView explains. This is why the method is called exponential: with a fixed coefficient, the influence of an original value fades through repeated multiplication by the same factor. The selected length therefore does not define a fixed set of recent prices beyond which everything disappears, as in an SMA. A very small influence is also not exactly zero in the mathematical description. At length 1 under the coefficient convention above, the exception is simple: the EMA equals the new price and the previous result has no weight. [3]

Petr switches from an SMA to an EMA, and with the same period the line usually follows new prices more closely. Fidelity points to both sides of this sensitivity: an earlier response and more frequent short-term changes. In a market without a clear direction, Petr may therefore see a string of quick reversals. A trading rule must also account for the costs of these trades. Proximity to the price alone does not measure the quality of his decisions. [1]

Petr compares his chart with a friend's. In TradingView, Source selects the input, such as Close, the candle's last price. Timeframe sets the indicator's data interval, which can differ from the displayed chart's interval. Before calling a difference an error, they compare the market, data provider, input price, interval and handling of missing values. Comparing the calculation itself makes sense only once the inputs match. [2] [4]

On a live chart, the newest EMA moves with the price. TradingView explains that an open candle's Close is a continually updated value, not its final close. An EMA using it changes too, as can a crossing. Petr therefore needs to distinguish a rule evaluated during the interval from one that waits for its close. A historical test cannot treat the final price as information he already knew earlier. [5]

Petr can calculate an EMA, but that does not yet specify when or how to trade. In its strategy documentation, TradingView separates a signal, order creation and actual filling. Petr also needs exit rules, timing and costs. Watch for overfitting too: tuning rules to the accidents of a past chart. Testing on separate data without further adjustment helps reveal this weakness; even an exact calculation cannot promise a future outcome. [6]