@ Fat Tails: I will add to your excellent commentary. Correct me if I am wrong.

Applying the Inverse Fisher to a non-normalized indicator is problematic because you are forced to choose a normalization window. Thus, once the signal lasts longer than the normalization window, the normalization changes, even if the original indicator value did not. This essentially double-counts whatever price action the original indicator was designed to capture, and can lead to false signals.

Personally, I have found normalization of non-normalized indicators fruitless.

while I always appreciate a heavily justified mathematical discussion,

but what I did not see clealy stated was how such an indicator is superior to a whole host of other simpler indicators, nor how to use ths inverse fisher transform

for example, one of the most complicated indicators that used to crash Ninja, remains the MACDBB

there are numerous forms of it, some even show a yellow (default colors) dot, just to highlight the ZLC (zero line crossing)

justify please

while I always appreciate a heavily justified mathematical discussion,

but what I did not see clealy stated was how such an indicator is superior to a whole host of other simpler indicators, nor how to use ths inverse fisher transform

for example, one of the most complicated indicators that used to crash Ninja, remains the MACDBB

there are numerous forms of it, some even show a yellow (default colors) dot, just to highlight the ZLC (zero line crossing)

justify please

Thanks again Fat Tails for the reply. Do you use it or think it's good at filtering out false moves or signaling the main part of a move? Well at least you didn't say it was flawed

Consider the PFE as being like a chameleon. It becomes a different indicator for each instrument and timeframe.

I am a new user of this site and while I realize this thread is somewhat dated I was wondering if anyone has attempted to modify this indicator as proposed in the article by William Eckhardt as referenced earlier?

I coded the examples in the article in Excel using the original formula and even though I could not exactly match the values in the series 1 & 2 arrived at by Mr. Eckhardt I was able to verify that the results are very different when a currency conversion is performed, again as indicated by Mr. Eckhardt.

I then coded the formula revision (in Excel) he suggests to allow the algorithm to pass his C-test and the results are exactly the same for both series both before and after the currency conversion which I believe now passes the C-test. I am not a mathematician and am also not stating that even with Mr. Eckhardt's changes that it would make this indicator useful in any way however, I've played with this indicator a bit and if one changes the variables according to the chart's time period (and maybe instrument) it can be made to pick out trends. As noted in previous posts and Mr. Eckhardt's article this is not the earmark of a good indicator as it lacks consistency needed however, Mr. Eckhardt's formula changes seem to rectify that condition. at least in a spreadsheet test using his examples.

Below is the comparison of the original PFE (blue), which shows inconsistent behaviour, and the PFE which I modified following the suggestion of Mr. Eckhardt (red). The modified indicator is attached.

]

@Fat Tails

Thank you for the revision.

I've tested it and have observed some problems with it. I have no idea whether it's the algorithm, the implementation, Ninja, or the way I'm using it. I've picked settings that will indicate a trend change (for now my definition for this situation is a 13EMA & 20SMA crossing) where the PFE (revised and original) cross the zero line. Although these settings might be different for different time periods they are relatively close.

Here's what I have observed so far:

1. Sometimes when switching instruments (futures instruments 6E>>ZW or ZC) the plot will just disappear. This happens on 1 and 2 minute charts. If the time period is changed to a higher value (3 min. and up) the plot will return.

2. The plot does not scale to the zero line which at times completely "disappears". Compressing the Y axis will sometimes make it return, but not always. Other indicators with a zero plot line (like the ZL MACD) the zero line remains in view and the plot scales accordingly.

3. Sometimes the plot will completely disappear and can be made to return by increasing the value of the period by one (12 to 13) with no change to the smooth period (8).

4. The ToS implementation of the original indicator also has a +/- 25 and 50 plot line which helps one visually gauge the current value. The original Ninja version or the revised have neither.

In one of your posts you stated that the indicator was completely useless so it is unclear to me if you revised it to see if the new algorithm worked any better but are still of the opinion that it is a useless indicator which in fact it may be. This is only and experiment for me, so I will defer to your desire or not to address any of the above. Regardless, thank you for your efforts.