Why EMA Values Differ Across Chart Resolutions

Summary

This document raises a question about why a five-day exponential moving average can differ when calculated from hourly closes versus minute closes. The finer series contains more observations over the same calendar span, and the author reports seeing substantial discrepancies near the end of the plotted period. It asks whether the extra observations make the higher-resolution estimate more faithful to the underlying market signal.

No answer or calculation is included, so the document does not establish which resolution is preferable or quantify the source of the difference. The issue highlights that an EMA’s behavior depends on how its smoothing period is defined and on the sampling frequency: using the same nominal duration across different bar sizes may imply different per-observation decay rates. Close prices also summarize each bar and can omit intrabar movements. The discussion leaves open how to align the EMA parameters and what signal the intended strategy needs; higher resolution alone does not demonstrate greater usefulness.

Key ideas

• The same calendar lookback can contain very different numbers of observations at different bar resolutions.
• EMA outputs can differ when sampling frequency changes, even when the stated lookback duration is the same.
• A bar close omits price movements that occurred within the bar.
• The document poses the comparison but provides no resolution, evidence, or recommendation.

Tags

# EMA with different resolutions


# EMA with different resolutions












I am trying to understand something:

If I calculate an EMA over 5 days, using the hourly close, I have to go over 5 * 24 points.

If I calculate an EMA over 5 days, using the minutes close, I have to go over 5 * 60 * 24 points.

Obviously, the 'close' value is very arbitrary and the hourly data may miss a lot of of things.

What surprised me is that when I plot EMAs, in TradingView, over the same time period, but with a very different resolution as described above, the output is very different, up to 20% in the last few days.

Is it the extra data, in the minutes resolution, responsible for this?

In that light, shouldn't the high resolution data be more useful as more 'true' to the real signal? (with the assumption that using 'close' values is arbitrary and may miss a lot of important moments.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.