Discussion of article "Exploring Seasonal Patterns of Financial Time Series with Boxplot" - page 3

New comment
 
Igor Makanu:

All right, I'm gonna come at this from the other side, let's drop the whole "trader's rubbish and speculation" thing.

Here we have a stochastic process described by means of a time series, we decided to make a discreteness of 15 minutes - OK, and then we decided that for evaluation we will use MA(25) = cumulative average value for 4 hours, and this evaluation for the previous 4 hours allows to allocate a certain time interval where the stochastic process will have a guaranteed growth?

then we can conclude that this process depends on the time series data for the previous 4 hours?

i.e. what I'm getting at is that MA for estimating time intervals is still an adjustment or the investigated process will definitely have a connection with the previous data - only such conclusions I can personally draw, the conclusion that the process has a connection with the previous 4 hours is not correct based on the results of 2015-2017, only an adjustment according to this methodology remains, imho.

We have that, on average, the deviation is above the MA during 4 hours, for each hour.

then we just terver that on average we are on the plus side if we buy below the MA and close above it.

I don't understand why the indignation? it's just a conveniently visualised statistic that will shatter all your dreams (if any) and plunge you into reality.

treading water ) Just because you can't find a cycle this way in history - so maybe you can look for other cycles that are more survivable. I checked with the classics.

Don't forget that you can combine monthly/daily/hourly.

All your algorithms will crash on the same thing, the lack of cycles. That's the foundation. It's easier to look at it once and see it in a diagram than to write a bunch of algos that crash into the same thing.

 
Igor Makanu:

I don't have resentment, I didn't like the TS to evaluate the methodology, as I wrote above such trade on MA doesn't show anything at all

that's why I gave up looking for regularities, or rather there is only one regularity in the market - daily activity of participants, it is repeated by time intervals, and then the mechanics - search for time interval and then the genetics of the tester puts orders chaotically, according to simple rules, those results that have good runs in the tester I run on the forward, a small part of tests makes sense - they pass the forward confidently, even in the mode of all real ticks. With this approach, it is faster to search than to invent a TS and then test it

ok, I'm leaving the topic, I wanted to discuss it - I discussed it with the author, thanks again for Python - useful revised and clearly presented material!

Why delete? ) catch the update, there are inaccuracies in the article in the grouping of data

suggest another detrend if MA doesn't like it, I'll do it

Files:
Seasonal_exploration.zip  385 kb
 
Igor Makanu:

I don't want to spoil my karma, the material is useful, perhaps many readers needed it, and here I am ..... Dartagnan or Robin Hood? ))))) although I'm probably just a burned-out searcher who's already gone round in a hundred circles ))))

I write also to get feedback, not to brag about something I don't understand.

The bot can be improved and served under some sauce on the Market. Even many intricate solutions will not reach this, from 3 lines.

300p


 
Igor Makanu:

Well, if you really want to master the topic of BP research properly, you need to study something with regularities, as a variant of the same methodology (from the article) to study the temperature schedule of the weather, there is definitely a regularity there and there can be no fitting.

if the methodology works correctly on several temperature data, then you can try to transfer it to chaos (CR), and knowing that the method is working, look for why it does not work in CR - i.e. only then apply some filtering magic to CR (increments, regression, normalisation through logarithms, etc.).

It will work with any regular cycles, it is a classic.

Only Alexander knows how to work with floating ones, even if you put a candle in it...
 

I use approximately the same research (but based on thinned ticks) in my TC.

Of course, I made some different conclusions from this data.... I do not need to know in which hours there is a rise or fall on a particular pair (in this case EURUSD), because on another pair there may be a completely different picture....

I care about the principle - the stability of seasonal, calendar "moustache boxes". Are they always like this on average? This article proves - yes, always. The appearance of these boxplots is always about the same. This is a very important point... If we add tick volumes to these studies, it becomes obvious that the spread of these candlesticks is proportional to them. And the rest is a matter of technique.

 

In short (so as not to bore the eager) - the research in this article can and should be used in forecasting the volatility in the market. Not to look at the "here and now" data, but to know that in the next period of time there will be a widening/shrinking of the dispersion channel. And that this will always be the case.

In fact, the pictures from the article are parts of the general picture of market cyclicality. Yes, it is ugly. But it is what it is in our linear world. In the non-linear world it is beautiful....

[Deleted]  
Seasonality is for commodities, not currencies.
 
Alexander_K:

The principle that matters to me is the sustainability of seasonal, calendar "moustache boxes". Are they always like this on average? This article proves - yes, always. The appearance of these boxplots is always approximately the same.

This is a strange reading. It is specifically shown by the author that it is different.

This is a very important point... If we add tick volumes to these studies, it will become obvious that the spread of these candlesticks is proportional to them. And the rest is a matter of technique.

Some kind of Nechayev's nonsense.

 
Vladimir Baskakov:
Seasonality is for commodities, not currencies.

It can be more pronounced there, in theory. In practice, you have to look at it.

Anyway, I have started a more detailed analysis. If there will be interesting results, I will write.
 

I can't rest on this topic, as they say "hello again"!

here I have sketched my test code to test the hypothesis "Let's allow to open trades only in 0-1 hours, with the assumption that in the next few hours the deal will still be closed in profit,".

input int Hour_Open  = 0;
input int Hour_Close = 1;
input int TakeProfit = 500;
input int StopLoss   = 500;

#include <fxsaber\MT4orders.mqh>

int LastTradeDay;
long LastTicket;
double d_takeprofit, d_stoploss;
//+------------------------------------------------------------------+
int OnInit()
{
//---
   LastTradeDay = -1;
   LastTicket   = -1;
   d_takeprofit = TakeProfit * _Point;
   d_stoploss   = StopLoss * _Point;
//---
   return(INIT_SUCCEEDED);
}
//+------------------------------------------------------------------+
void OnTick()
{
   MqlDateTime  dt;
   TimeCurrent(dt);
   bool tradetime = (dt.hour >= Hour_Open && dt.hour <= Hour_Close);

   if(LastTicket > 0 && !tradetime) {
      if(!OrderSelect(LastTicket, SELECT_BY_TICKET))   LastTicket = -1;
      if(LastTicket > 0 && OrderCloseTime() > 0)       LastTicket = -1;
 // if(LastTicket > 0 && OrderClose(LastTicket, OrderLots(), SymbolInfoDouble(_Symbol, SYMBOL_BID), 30)) LastTicket = -1; //close at non-trading time
   }

   if(tradetime && LastTradeDay != dt.day && LastTicket == -1) {
      double ask = SymbolInfoDouble(_Symbol, SYMBOL_ASK);
      LastTicket = OrderSend(_Symbol, OP_BUY, 0.1, NormalizeDouble(ask, _Digits), 30, NormalizeDouble(ask - d_stoploss, _Digits), NormalizeDouble(ask + d_takeprofit, _Digits), "", 0, 0, INT_MIN);
      if(LastTicket > 0) LastTradeDay = dt.day;
   }

}
//+------------------------------------------------------------------+

I ran in optimiser 01.01.2017-01.0.12019, if to look for beautiful balance charts, the optimiser finds takek 100p and stoploss 4600p

if you look for reasonable take/stoploss, there is no pattern at all.


or still the methodology of searching for a pattern does not work, well, as an option I lost practice and wrote a crooked code - I haven't written in MQL for more than a month.

12345678910...33
New comment