I made one of these things once ... - page 11

New comment
 

Real experience will add, it is there. I wrote above that there is a filter that will weed out many false breakdowns. But that's not it, it's a study of the TS, not the strength (statistical significance) of the round level. Anything additional to the algorithm for gathering statistics takes us away. The same zigzag, the zigzag has parameters and break points depend on these parameters. We're trying to investigate the level, that's what we're investigating.

If we want to study statistics, how often the zigzag will break near the whole level, it will be other studies, other criteria...

Z.S. I generally believe that the zigzag and this "notorious" fractality of the market has nothing to do with it, it's not what we're investigating...

 
Prival:

But that's not it, it's an investigation of the TS, not the strength (statistical significance) of the round level. Anything additional to the statistical collection algorithm takes us further away.

The strength of the level is not enough. I would define the strength of a level simply in terms of the number of crossings by the price; a strong level is bound to be trampled by the price. I thought you were talking about reversal probability from the beginning.


P.S. For circular levels fractality is not involved, but if we talk about others, it may well be "involved".

 

Tried NormalizeDouble. The results are not exactly clear-cut. It was a bit slower than the two-action variant via an integer variable. But not as much as I expected. That is, you can potentially use it in algorithms intended for speed.

But not for calculating "rounded" levels, because it does not just cut off extra digits, but rounds.

 

Ok, here is a simple script, it calculates the number of crossings of "round" levels plus Delta points. I used it on EURUSD, GBPUSD and USDCSD 10:55 of 16.06.2004. The result is unexpected and interesting.

Comments both on the text of the script and on the question are welcome :)

// TestLevels.mq4
#property show_inputs

extern int Delta = 0;
//+------------------------------------------------------------------+
int start() {
  int pos;
  int ILvl;
  double RLvl;
  int Cnt = 0;
  for (pos=Bars-1;pos>0;pos--) {
    ILvl = High[pos]*100;
    RLvl = NormalizeDouble(0.01*ILvl+Delta*Point,Digits);
    if (High[pos] >= RLvl && Low[pos] <= RLvl) Cnt++;
  }
  Comment("Количество пересечений для Delta = ",Delta,"  составило ",Cnt,", Level: ",RLvl);
  return(0);
}


P.S. For large Delta the script lies, but it should not cancel the unexpectedness of the results

 

OK, one more post and until there are no more answers on this topic :)

I made again a histogram drawing indicator and this is what I got for the mentioned three pairs (horizontally - increment to the round level in old pips, the beginning is zero, the end is 99)

this is EURUSD and GBPUSD

it is USDCAD


Files:
testlevels.mq4  3 kb
 
Candid:

I would not count simple circular level crossings, but the number of matching zigzag vertices with different periods for each step between circular levels. Just a quick thought. That is, not how much price stays near a level, but how much the level is the target for price movements. Or maybe zigzag lines crossed with different periods.
 

It's taking me a long time. I made an indicator in MQL-5

it shows one when the level is broken (by the logic I described earlier)

For the level of 1.29 here are the statistics

2010.07.18 21:20:45 TestLevel (EURUSD,M1) Amount=1113
2010.07.18 21:20:45 TestLevel (EURUSD,M1) Amount=1
2010.07.18 21:20:45 TestLevel (EURUSD,M1) Amount of bars 4039582
2010.07.18 21:20:45 TestLevel (EURUSD,M1) Start date 1993.05.13 00:00:00

i.e. it has broken the level 1113 times.

I attached the indicator.

If I understand correctly from your statistics. 50 level is ok, but it's not what I suggested, I'll try to write my own statistics collector

Files:
testlevel.mq5  3 kb
 
Candid:
Well the first degree is still rough, but at least I'm glad we both respect splitting in half.


It's only rude to assume that the split is only done once. What I meant was that by dividing in half we can get lines of the first sublevel. Dividing in half the intervals between them is the lines of the second sublevel. And so on. That is what is implemented in the Murray indicator. It looks quite plausible.

Candid:
I think Items 1 and 2 can be reduced to one, by selecting the horizon we can obtain both base and interval between levels.


It's not clear how selecting the horizon can determine the base. By base I mean the absolute price value from which all levels of the grid are deferred according to the interval.

Candid:

The result is unexpected and curious.

I don't seem to understand what is so unexpected and interesting. :-(

 
gip:

I would not count simple circular level crossings, but the number of matching zigzag vertices with different periods for each step between circular levels. So, a thought on the top of my head.
The zigzag was here a little earlier, isn't that the statistic it shows?
 
Yurixx:


It's only rude to assume that the split is only done once. What I meant was that by dividing in half we can get lines of the first sublevel. Dividing in half the intervals between them is the lines of the second sublevel. And so on. That is what is implemented in the Murray indicator. It looks very plausible.

Got it at last :). You could have explained it to me much simpler, I wrote right before about division by powers of two, you could have written that you agree with me, and that's all. Then I wouldn't have looked for semantic differences in your words.


It is not clear how the choice of horizon can determine the base. By base, I mean the absolute price value from which all levels of the grid are drawn in accordance with the interval.

I think that the base is either in the chosen horizon or in the senior one. That said, mind you, I'm not saying I know the algorithm for defining it.
I don't understand the surprise and curiosity. :-(

So you don't see any statistically significant effects here? I would expect there won't be any, but it looks like there are. Although not too strong, the grail is unlikely to work.
1...4567891011121314151617
New comment
Reason: