Average Directional Index

Peter sees a falling price and rising ADX below it. The chart is not contradicting itself: a downward trend may be strengthening. To understand why, he starts with a candle, which summarises one time interval. High is its highest price and Low its lowest. TradingView calls the current High minus the previous High UpMove; the previous Low minus the current Low is DownMove. +DM takes UpMove only when it is positive and greater than DownMove. −DM uses the opposite comparison. On a tie, both components are zero. The calculation selects the larger positive shift in an extreme price; it does not add both directions at once. [1]

A price shift needs a scale. Fidelity compares it with True Range (TR), a range that also accounts for the distance from the previous closing price. TR is the largest of High–Low and the absolute distances of both extremes from the previous Close, meaning the closing price. Absolute distance measures the size of a difference without its sign. Directional movements and TR are first smoothed using Wilder’s method, so they also carry history. The smoothed upward movement divided by smoothed TR, multiplied by 100, gives +DI; −DI is calculated similarly. The minus in its name denotes downward direction, not a negative number. With a zero denominator, the ratio itself is undefined. [2]

Peter now wants to know how far the lines differ. +DI and −DI produce an intermediate result, DX: their absolute difference is divided by their sum and multiplied by 100. This is the comparison Fidelity describes. Absolute is the crucial word. Swapping the values changes the direction of dominance but leaves DX unchanged. This does not calculate price growth or the chance of success. If the sum of the lines is zero, there is nothing to divide by; any substitute value an application supplies is an implementation convention, not the result of that ratio. [2]

DX describes the current ratio; ADX smooths its history. Fidelity starts with the average DX over the selected number of periods, then incorporates later values into the previous result. For length n, multiply the previous ADX by n − 1, add the current DX and divide the total by n. For example, 14 refers to calculation periods, not automatically days. A single pair of directional values therefore lets Peter calculate DX, but not ADX without its history. When comparing applications, he must also check initial data, lengths and smoothing methods. The result retains history and reacts to market changes with a lag. [2]

The DMI set described by TradingView assigns different jobs to its lines. +DI above −DI means the smoothed upward component is stronger; the opposite order favours the downward component. ADX itself describes the strength of the measured imbalance after smoothing. A rise in ADX is therefore not necessarily good news for a bitcoin holder: it can accompany a deepening decline. Falling ADX does not automatically announce a price reversal either. A trend can continue with a less pronounced imbalance while the average still reflects earlier conditions. None of these lines alone tells us what happens next. [1]

TradingView lists the traditional guide levels of 20 and 25. Below the first, a trend is often described as weak; above the second, as stronger. But Peter would be mistaken to turn a threshold crossing directly into a buying rule: a strong trend may point down. Values depend on the market, time interval and settings; these are not universal boundaries for every chart. Low ADX does not mean a quiet or motionless price, either. Large moves may alternate in both directions, leaving only a small measured imbalance. [3]

Peter also needs to know when he read the number. TradingView notes that High, Low and Close are not final until a candle closes. A new high or low can change directional movement and True Range, affecting the resulting indicator. A crossover seen during the candle may therefore disappear by its close. Testing must distinguish a decision during the candle from one after it closes; the final value cannot be treated as information that was already available earlier. [4]

When the directional lines cross, their order has changed. That alone does not demonstrate a profitable trade. TradingView warns about false signals and separates a signal from actual execution in its testing guidance. If Peter uses ADX as a filter, he needs advance rules for entry, exit, trade size and when the plan ceases to apply. Results must include fees and slippage, the difference between expected and actual execution prices. Repeated tuning on the same historical data leads to overfitting: a rule learns the past but may fail elsewhere. Testing on data that was not used to choose the rule helps address this. [3] [5]