I thought it would obviously have to be time invariant. Doesn't TA only depend on the data in the chart? I don't know much about TA, but Wikipedia and Investopedia say that TA is based on the study of past market data, primarily price and volume. This seems to implicate that if the same pattern emerges in the charts in April or June, the conclusions should be the same. Same for a market moving slow or fast. If all trades are done at half the speed, shouldn't the conclusions on the same pattern (but stretched twice as long in time) be the same? And isn't the model describing a bear trap and a bull trap essentially the same, just applied inverted?
I know there's a kind of "law" in (amateur, perhaps also professional) algorithmic trading that your strategy should be time invariant. Some even go as far as saying it should be
market invariant. In reality, I think it's more a heuristic to find out if the "pattern" you think you detected is just noise, or if you're onto something.
This is completely 'finger in the air', but I sometimes think that certain patterns that seem to be a good predictor emerge across different time scales (hence, motivation for the above "law"), but they're not equally important/active across all scales.
EDIT: The previous paragraph would still be compatible with "TA is based on the study of past market data, primarily price and volume. This seems to implicate that if the same pattern emerges in the charts in April or June, the conclusions should be the same" I would argue: you could simply be missing one variable, say, something like "total volume necessary until pattern x becomes active/has a strong enough influence on the market".
To come to your counter question, I cannot exclude that a signal is useless (usable?) because it lags. Even more, since this signal is interpreted by humans with neural networks in their heads, they can very well learn how to interpret them without understanding them (as we do with so many things). Simply by training. One doesn't need to have an explicit model that makes sense. And when many people start acting on a certain signal (even if it's not based on a proper model and it doesn't make sense), the very fact that a significant part of the actors take that signal into account creates a self fulfilling prophecy. So yes, that may very well constitute an exploitable pattern.
That was the point I was trying to make. I believe "chartism"/TA is more than just a self-fulfilling prophecy, but there is some self-fulfilling aspect to it, mainly in the form of the exact points of resistance/support.
What I mean is: Some combination of signals tells traders "market is bearish, even though we're currently going up" (the non circular part of TA). Next question, where exactly to go short? "Hm, let's say DSMA50. Seems like a good place as any." After the first few sell at exactly that point, it becomes a "point of heavy resistance" (the circular part of TA).
/my 2 crackpot cents
I am not a TA expert (I just stare at charts without a clearcut method), but my theoretical take on TA is that in order to be successful it has to describe/utilize human valuation psychology (and the bugs and quirks therein).
If that's the case, given that the human brain has a memory function, and thus has been 'trained' by prior price action, it makes sense that your price prediction signals could vary over time as well, no?
