Again and again I read posts here that say, "I always seem to enter too early" and "I took profit way too soon". While every trader will have to master his own emotions and impulses, there is one factor implicit in the TMA indicator that could be contributing to this. This post is an explanation of that factor and advice in how to correctly interpret the TMA.

Fundamental Properties of the TMA

The TMA is a nondeterministic or noncausal indicator. That means that in order to compute the average at point X, it makes references to values that are in the future relative to X. Well it would be dead easy to make a killing with any indicator that could truly access the future, but this only works when computing historical values. When the TMA gets closer and closer to the present, it needs access to values that haven't been determined yet (like tomorrow's closing prices). To be exact, for a TMA with the Length set to 56, the most recent known value of the TMA is for 57 bars back.

To provide values for the TMA plot closer than 57 bars back, the best we can do is estimate what the future values are going to be. The TMA indicator uses an unbiased estimate. That means it neither assumes that prices will go up in the future nor that they will go down in the future -- if a trend (a bias) is present in the data that trend will not be considered. In estimating what the future closing prices are going to be, it takes the most recent closing price available and extends that value straight across as far as necessary. (Missing graph in original post)

The vertical line is the present, separating past from future. As the TMA approaches this line it has to refer to more and more of the unbiased estimate of the future, and you can see that if it is plotted far enough out to the future it converges to that horizontal line. But on your charts you will only see the part that is to the left of the vertical line. But the important point is that as the TMA approaches the present bar on the chart, it has a characteristic curl on the end, which curls away from the trend that is in the data, becoming more horizontal (diminishing the slope). This curl is always present.

This curl makes it appear as though the TMA is turning and the trend is about to end. If you follow this indicator blindly, what happens? You enter and exit trades too early!

A few pages back Al mentioned that when prices are trending that he enters trades at the center line of the TMA, which ends up being the top/bottom of the channel when it finishes repainting. This is precisely what happens because of this phenomenon.

I developed and tested a few algorithms that attempted to remove that curl by projecting the trend into the future (i.e. a biased estimate). They work great when the trend is straight, but when the trend turns a corner they overshoot wildly. So far, the best estimate is an unbiased estimate. However, I'm beginning to believe that no estimate is even better than an unbiased estimate.

Indicator Blindness

Whenever I use indicators on a chart I have a tendency to look at the indicator and not directly at the price candles. I presume this is common to most traders. The indicator has a tendency to create a kind of blindness to observing what is going on in the price bars. When trading Extreme TMA, what you need to do is examine recent price action to form an opinion of what the TMA is going to look like in the future after all of its repainting is finished. If you read Al's posts you will see that is precisely what he does. If all you do is take a cursory glance at the price action and look mainly at the plot of the TMA estimate, it will tend to deceive you into thinking that the trend is starting to turn when there is in fact no sign of that in the bar chart at all.

For this reason I'm seriously considering eliminating the estimate part of the TMA entirely. Yes, that means that the plot will end 57 bars back from the current bar and you will have nothing but your mental projection of those lines to go on. (Several of you are reacting in alarm already -- don't worry, if I make this change in the stock indicator it will have an option to switch it on or off.) But I believe this is a good thing.

1. It will no longer have that characteristic curl at the end of the line;
2. It will never repaint;
3. It forces you to examine the price chart manually and form your own estimate of how well the trend is holding up.

And your own well thought out estimate of the continuation of the TMA lines will always be superior to what the machine can do. This is a fundamentally different approach than using TMATrue, which isn't a TMA at all, but an LWMA. TMATrue plots only the estimate part of the TMA and makes no reference to the actual TMA at all. The majority of opinions that I've read on this thread say that TMATrue is crap and it doesn't work for this trading method. (Please don't go posting that you disagree that TMATrue is crap and you use it all the time etc etc. You are entitled to your opinion and to use TMATrue all you want. But Al said he has tested TMATrue and could not get results with it, and I'm just saying that these findings support that conclusion.) So if the estimate part of the TMA can't be used to trade successfully, then I'm inclined to conclude that it contributes nothing and only distorts our view of the TMA, so it should be eliminated.

But changing the FastTMALine to eliminate the estimate part would have consequences upon the TMA slope limits (-0.4 to +0.4) because those values are taken from the TMA estimate and not from the actual TMA. And the other software tools used here that compute the TMA may still be using the TMA estimates. It would also affect the alerts features, which some people rely upon for their trading signals. For these reasons I probably will not make these changes to the official released version of FastTMALine but just my own private version (but opinions on this are invited).

Summary
The TMA cannot generate a value for the current bar or the bars near it, so an estimate must be done. The best available estimation algorithm results in a flattening of the slope of the TMA as it approaches the current bar. This flattening effect can trick the user into assuming that the trend is slowing down or turning when it has not. To not be fooled into entering or exiting a trade too early, the user needs to discount what the estimate is showing and mentally project what the TMA will actually appear like after it has finished repainting. Complete elimination of the estimate part of the TMA plot is the most direct way of ensuring that would be done, but it would have consequences on the way that the indicator is used. A change to the released version of FastTMALine isn't currently planned, unless the readers of this thread demand it.

nonlinear wrote:If anyone is close enough to the math, I would be interested in a high level explanation of the difference between TMAtrue and the MACD histogram.