Why trend filters like ADX fail at the tick scale

[Amanhasnoname]

Indicators such as ADX or moving average slopes work on time bars with meaningful smoothing windows. Applied to tick-level or sub-minute data, they behave poorly.

At tiny horizons, price movement is dominated by noise and microstructure effects. A trend measured over 14 one-second bars reflects bid-ask bounce and random fluctuations more than any directional flow. The indicator flips constantly, producing signals that carry little information.

Lag is another problem. By the time a smoothed indicator confirms a trend on short bars, the move may be over.

For scalpers, more suitable tools measure activity and imbalance: tick direction counts, spread changes, volatility relative to normal, and distance from short-term reference levels.

If a trend filter is desired, compute it on a higher timeframe, such as five minutes, and use it as a regime filter, not as a trigger.

Testing on tick data reveals how well the filter predicts the next few seconds. Often, the answer is not much better than chance.

Which filters have shown any predictive value at the horizon of the trades?