Trend and Moving Averages

The short answer

A simple moving average adds the last N closing prices and divides by N. An exponential moving average does the same with more weight on recent closes. Both are plotted at the current bar even though they describe a window that ends there.

That last point is the whole story. A moving average tells you about the past window. It is drawn where the present is. Everything people find confusing about moving averages comes from that gap.

Smoothing and lag are one property

Averaging removes short-term variation. That is what makes the line readable. But averaging also means the line cannot respond to a change until enough new data has entered the window to move the mean.

You cannot have one without the other. A shorter period reacts faster and smooths less — more responsive, more noise. A longer period smooths more and reacts slower. There is no setting that gives you responsiveness without noise, because they are the same dial.

This matters because a great deal of effort gets spent looking for the "right" period, as though a correct answer exists and has simply not been found yet. It does not. There is only a trade-off you have to choose a point on, deliberately, in advance.

What a moving average can tell you

Three legitimate uses:

Direction of the recent window. If the average is rising, closes have been rising over that window. This is a description, and a reliable one.

A reference for "extended". Distance from a moving average measures how far price has travelled from its recent mean. Unusually large distances are unusual. They are not obliged to close, and identifying an extreme tells you nothing about timing.

A mechanical, consistent trend filter. "Only take long setups while price is above the 200-period average" is arbitrary, but it is consistently arbitrary. Consistency is worth something on its own, because it makes results interpretable — you know what rule produced them.

What it cannot tell you

It cannot know a trend has ended, because it is computed from data that ended before the turn. Every trend-following tool gives back part of the move at the end. That is not a flaw in the tool; it is the price of not needing to predict.

It cannot work in a range. When price oscillates around the mean, the average sits in the middle of the oscillation and every crossing is a false signal. A trend-following tool applied to a ranging market produces a stream of small losses. This is the tool behaving correctly in the wrong conditions.

Crossovers

When a shorter average crosses a longer one, the recent window has moved faster than the longer one. That is all a crossover states — and it states it late, since both inputs are already lagging.

Crossover systems can be profitable, in trending conditions, with strict risk management and the acceptance of many small losses between the few large winners. They are not profitable because the crossover predicts anything. They work, when they work, because trends persist long enough for a few large wins to outweigh many small losses.

That is a claim about market behaviour, not about the indicator — and it is testable, which is what makes it worth stating in that form.

Parameter shopping

Test twenty period combinations and one will look excellent. This will happen with random data.

If the best-performing combination is dramatically better than its neighbours, be suspicious rather than pleased. A parameter that captures something real tends to degrade gracefully: 48 and 52 should work nearly as well as 50. A sharp peak at exactly one setting is the signature of a fit to the sample.

The backtesting lesson covers how to test in a way that makes this failure visible rather than invisible.