Taking Neural Networks to the next level - page 19

 
@NELODI Yes what you saying is true. Similarly having a model that works to explain the market most of the time is the best you can hope for, the rest requires judgement. If that works in the long run then no problem. I am quite sure that you are not correct about the scaling though. If you have model that works most of the time and gets defeated sometimes then it is still reasonable to scale. Not necessarily after every little loss. This could improve efficiency exponentially it is not martingale
 
Brian Rumbles:
@NELODI Yes what you saying is true. Similarly having a model that works to explain the market most of the time is the best you can hope for, the rest requires judgement. If that works in the long run then no problem. I am quite sure that you are not correct about the scaling though. If you have model that works most of the time and gets defeated sometimes then it is still reasonable to scale. Not necessarily after every little loss. This could improve efficiency exponentially it is not martingale

This increases your risk exponentially. You only have a certain degree of control of losses but not profits.

 

You guys make fair points about the scaling. However I think that if a model actually works then it is positive expectancy. Therefore scaling will help it, I will personally do this on demo and see if I can prove myself wrong. That is trade when your model is valid, Stop when invalid and increase lot size and continue when valid again if it is necessary and reset when profitable. It might be wrong but it might be really efficient, my scaling is not martingale. I will see for myself.


I don't see why losses and profits have any different degree of control, what is the reason for this?

 
Brian Rumbles:

You guys make fair points about the scaling. However I think that if a model actually works then it is positive expectancy. Therefore scaling will help it, I will personally do this on demo and see if I can prove myself wrong. That is trade when your model is valid, Stop when invalid and increase lot size and continue when valid again if it is necessary and reset when profitable. It might be wrong but it might be really efficient, my scaling is not martingale. I will see for myself.


I don't see why losses and profits have any different degree of control, what is the reason for this?

For a system that works, there is no need to scale. With a fixed % of balance you still scale, but keep the risk the same.

All these loss recover methods never pay out and only hides a system that in essence does not work.

The reason you control loss is that you can decide when to cut the loss. Only Elon Musk and Chuck Norris can control profits because they decide what price it is. For us normal beings, the market decides what profit it will bring. 

 

Well I am not really trying to have an argument with you guys. Again, some things you said have some logic such as increasing your exposure may increase your risk, there is no need to scale if a system works, you cant force the market to move in your desired direction, you can control your loss by closing your trades. 

The rest of what you are both saying I can disagree with. I did not say double on loss, instead I said I don't use martingale. I did not say it was necessary, instead I said it can improve efficiency of a positive expectancy model.

You say you can control your losses by closing a trade but you think that you cannot also control your profits by closing out a trade? A profit and a loss are the same phenomenon. infact what you are managing is not profit or loss but exposure to risk.


I think Elon has a lot of trouble controlling losses and even making a profit at all , unless you mean by tweets

I also never said anything about reversing direction. I think there is too much of a failure of communication in this forum, I discuss ideas here but people filter what you say by adding assumptions
 
Yes OK, it is the model that is the risk and if the model breaks down, exposure to that risk is by your lot size. I will stop discussing this here, I think you guys are looking at things too simplistically
 
NELODI:
Any profitable trader will tell you that you should NOT change the direction of your trade, unless something fundamental has changed. And if something fundamental has changed, then you should cut your losses and re-evaluate the situation before making any new trade decisions. By opening new positions with double volume in the opposite direction only because your last trade didn't work out, basically means that you do not trust your system.
Yes Ok

I do not disagree with what you are saying, you just don't read what I said properly. You could be right about the scaling but I will see for myself on demo. This is equivalent to research and a willingness to be wrong and learn

 

NELODI, I disagree - not in general about the fact that there are differences between real trading and demo, but about the relevance of liquidity for the average forex retail trader.

Although what your saying is essentially correct, I would argue that the relevance very much depends on the market and on the stop distance:

- if you're scalping for just a few pips, it's of course more important to get filled at a good price than if your stops (SL/TP) are a hundred pips away

- if you're trading forex majors the liquidity usually is so insanely high, that you won't move the price with a relatively ridiculous single digit lotsize request

Of course liquidity is a real thing, but forex is also the biggest market and not some rare commodity options, so I don't believe that for anybody with a <100k account, there is a liquidity "problem".


Let's get back to machine learning / neural networks ;-)

 

500 lots aren't enough to influe eurusd's price :) : a drop in the ocean. 

and i fully agree with starting always with a low deposit and min lots

2019 Forex Trading Stats for Marketers
2019 Forex Trading Stats for Marketers
  • Finance Magnates Staff
  • www.financemagnates.com
FM HomeThought Leadership2019 Forex Trading Stats for Marketers As marketers in the forex sector know only too well, acquisition is tougher than ever before. There’s so much competition, tighter than ever regulations and limits on leverage and incentives. The battle to catch and retain traders is fierce and survival depends on the ability to...
 

Chris, i love to question this topic ;)

If a model performs best with normal price and tweaking it with specific features couldnt imporve its acurracy reasonably (/not overgfitting) how can one explain the bad result i got from the label utilization i used and explained a few sites ago?

I used the same labels but in a different order: 

Lebel Method 1 (badly working):  Neuron 1= Points in CRV Profitdirection; Neuron 2 Points in CRV Lossdirection; (first as TP wether long or short, second as SL)

Label Method 2 (pretty well):       Neuron1= Points in Long Direction; Neuron 2 Points in ShortDirection;  (TP or SL depending on better CRV)


Prices were exactly the same but in different order. it produced totally different results. the NN needed longer to find out about the pattern and may got distracted to other "hypothesis". therefore it might be worth considering that specific features ( not only prices could be worth thinking about). Other ideas/theorys?