Machine learning in trading: theory, models, practice and algo-trading - page 508
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
It seems to me that you idealize too much and for some reason are trying to generalize to "all traders" who fits what, just like some kind of guru, you do not sell trading courses, like a groom?(just kidding)
There is an optimum for risk (lot) given profit expectation. It does not depend on psychology and it's not a "who gets what" talk. For example, the maximum dose of risk is half of the deposit, but in proportion to the risk there will be profit, for example, if you take a small risk (<10% of the deposit per year), it's like trading 10% of capital, while 90% will lie in the money, it is suitable only as a conservation strategy, to prevent inflation eating away at it, but it does not earn, you have to invest 90% in other strategies and assets, diversify, etc. And all the same as a result the risk will be almost proportional to the profit with small smoothing effects of diversification, it is impossible to significantly remove risk and increase profit by means of money management.
Most traders need a relatively risk-free system with very low drawdowns and large profits, otherwise they won't survive in the market, that's what I mean :) because they don't have a cushion and can't wait for a small % increase. So, for example, by this logic, as you have, if you are supposed to do 100% a month, the average drawdown should be 50%, but it's like killing, because 50% drawdown is a point of no return for the deposit, as a rule. And less than 100% a month the majority will not be interested in forex forums, you make a poll, ask who wants how much profit, of these professionals with a good cushion will be 1% ... well, there's more at this forum
How else can you rise in the market from zero, for example?
I'm not a guru, I just communicate a lot with traders for the sake of interest :)Eh eh...
I don't know if it's subtle trolling or if you're so seriously deluded.
You said "it's all clear", why did you start again with small stops and large profits? I told you a diagram, to the extent of my artistic ability, where black and white it is obvious that you CANNOT significantly reduce risk without reducing profits, that is small stops - small profit (on average, of course).
This is exactly what the algotrader is doing, looking for inefficiencies. And if you can't, then most people have nothing to do on the market at all, that's the tale. We're not talking about money management or anything else, we're talking about pure trading. Where did you get such a ratio of profit to risk in the figure, it's an illusion, isn't it? The man has $ 1k, ie by your logic, he has nothing on the market with 5-50% per year.
You may have a mathematical mind, but mathematics does not always equal a sober way of looking at things. I can take some formula and pass it off as truth, for example Pareto's principle or profit/loss ratio like yours, and it doesn't say anything in relation to reality.
All the rest has nothing to do with reality and is nothing but speculation. What does the stableman have to do with it, don't put me in the same camp with him, he has half of the forum of imaginary students here :)
This is generally an argument about some obvious things, like whether a random forest is capable of extrapolating... Obviously not, but we need to argue, pick on the concept of extrapolation and something else :))
Once again, with some advantage in the forecast, for example53-55%, there is an optimal risk management strategy, deviations from it - will give a decrease in profits, on average. There are no significant differences in strategies for those who have 10M$ and those who have 100$, at least in Forex, where 6T$ are traded per day.
There is a huge difference when trading different amounts of forex, and it is very noticeable when you start trading, especially HFT. A la the unwritten rules of brokers. There's no such liquidity there because it's decentralized. And, often, the strategies that work for small fry don't work for big fry. Cardinally different approaches when trading different amounts.
For example, to set High and Low, tick and real volumes - a total of 4 inputs per bar.
For the network to understand the curve shape it must be fed for example with 100 bars - total 400 inputs for the neural network.
I have a training history of approximately 50 000 bars at M1 for 3 months.
What do you think of this approach?
How many internal layers to make? Apparently it should be a lot too, e.g. 400-100-25-1.
I believe such a network will take a very long time to learn. And probably it will find not the most optimal parameters.
And if you make 1000 or 2000 inputs? Will it be unrealistic to achieve something?
1) Exactly! YOU CAN'T DO THAT! Why do they forbid it in developed countries? Why did they invent "qualified investors"?
What exactly are they forbidden to do, trade? Hundred percent a month is quite achievable, as in any business and in any trade, as long as inefficiencies exist they are actively exploited, the main thing is to talk less about them. They are always temporary, i.e. will not work steadily for life. That's why I wrote above that there should be adequate criteria for stopping, as soon as possible to understand whether the model is no longer working. And you just decided to give up on everything. The difference between 5-50% a year and 2-25% drawdown is just that.
Maybe I'm wrong, but it seems to me that it is better to train the network on pure price data, rather than on indicators that usually average, i.e. introduce lag.
For example, to set High and Low, tick and real volumes - a total of 4 inputs per bar.
For a network to understand a curve shape I need to feed it e.g. 100 bars - total 400 inputs for the neural network.
I have a training history of 50000 bars for 3 months on М1.
What do you think of this approach?
How many internal layers to make? Apparently it should be a lot too, e.g. 400-100-25-1
I believe such a network will take a very long time to learn. And probably it will find not the most optimal parameters.
And if you make 1000 or 2000 inputs? Would it be impossible to achieve anything at all?
They do it on recurrent networks, feed prices, I don't know details but they say it works, it takes a long time to train on gpu, mostly on python.
MO + MO = )))
Maybe I'm wrong, but it seems to me that it is better to train the network on pure price data, rather than on indicators that usually average, i.e. introduce lag.
For example, it is better to set High and Low, tick and real volumes - 4 entries in total per bar.
For the network to understand the curve shape it must be fed for example with 100 bars - total 400 inputs for the neural network.
I have a training history of approximately 50 000 bars at M1 for 3 months.
What do you think of this approach?
How many internal layers to make? Apparently it should be a lot too, e.g. 400-100-25-1
I believe such a network will take a very long time to learn. And probably it will find not the most optimal parameters.
And if you make 1000 or 2000 inputs? Will it be unrealistic to achieve something?
I need a formula for L2 normalization. I can't find it. Maybe someone can help.
Network [64 GRU + 32 GRU + 2 Dense] in a classification problem using OHLC -> Buy/Sell model (7000 bar) on ~24 training run gives 0.9 - 0.8 accuracy. And all this in about 30 seconds.