Machine learning in trading: theory, models, practice and algo-trading - page 1480
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
To find out when a trend starts/ends a flat is a challange, then you have to switch between impulse trading and reverse trading, and only the frequency of trading depends on the smoothing parameters.
What the fuck are you talking about... when are you going to get over these stupid notions of trend - flat. If they occur more frequently, then it's like a "flat", if they occur less frequently then it's like a "trend", but the main point is in the market reversals.
When you know where the reversals are, you don't need to know anything else, you know where to open position, where to close it, where to put a stop ...
Knowing the "mode change between trend and flat" you don't know anything else, because nobody can even describe what a trend is and what a flat is, it's totally subjective bullshitMy experiments with scaffolds show the same situation.
Training for 6 months 2017, tes for Oct/Dec 2017 - err on the test/forward 42%. If you go by the valking forward method, when you shift to 2018 (Training for 6 months 2018, tes for Oct-Dec 2018) the error is 55% and a quick drain.
If you look at the annual graph, in 2017 there was a global growth with rollbacks up to 40 days, both on the tray and on the test. And in 2018 there was some more growth and the beginning of a global decline.
So we can conclude that if the 2 month pullback hasn't stopped, it's a new reverse trend. Les didn't realize that and for a third of the training section while there was still growth - trained for that growth, that's why he lost on the test.
That's right, it doesn't work, no data.
What the fuck are you talking about...when are you going to get over these retarded notions of trend - flat. There is no such thing, they are subjective indescribable concepts, there is only a change in trend (reversals), if the reversals occur frequently, then it's like a "flat" if it's rare, then it's like a "trend", but the whole point is in the reversals of the market.
If you know where the reversals will occur, you know where to open position, where to close it, where to put a stop ...
Knowing the "mode change between trend and flat" you don't know anything, since no one can even describe what a trend is and what a flat is, it's completely subjective bullshitThe trend is a positive autocorrelation of first lag retracements in this timeframe, the flat is negative.
All right, it does not work, there is no data.
The trend is a positive autocorrelation of returnees of the first lag, on the given TF, the flat is negative.
But autocorrelation of returnees is also a subjective definition of a trend, TFs are also subjective, they are not
But autocorrelation of returnees is also a subjective definition of a trend, TFs are subjective too, they are not
There can be many definitions of a trend, but the essence is in the autocorrelation, the inverse relationship between neighboring increments, when the autocorrelation is positive, the trend-tracking TCs work and the reversal ones merge, when it is negative, vice versa. Another matter is whether there are regularities in alternations of positive/negative autocorrelations that are more stable than the obsolete price increments. And unfortunately there are none, at least no more than with the increments, if we talk about "normal data", i.e. from particular brokerage companies, particular instruments at frequencies greater than one minute.
There are also indirect factors besides dependencies, dependency may be tempting by its sign, but it is very noisy, which reduces to nothing and even to a hard minus all "temptability" of the dependency.
There can be many definitions of a trend, but the essence is in the autocorrelation, the inverse relationship between neighboring increments,
I see what you mean, but here again we are dealing with a period (TF), which doesn't exist; you're calculating the autocorrelation at some fixed length section
I understand what you mean, but here again there is a tie on the period (TF) which does not exist, you are counting the autocorrelation on some fixed length section
Exactly right, we need to predict the autocorrelation, or if you like Hearst coefficient or similar, for a window in the future, for a series with a given quantization period (minutes, hours, etc.), but I repeat, this is not very common.
Sorry, what does "period (TF) which does not exist" do not quite understand exactly. I can assume that you mean if you take the original order-log, quantization on candlesticks, period value is arbitrary, but if so, I would argue that we are dealing with different patterns on different scales, there is a certain fractality, there may be a trend on one scale, on another a flat and a particular strategy usually works with a particular pattern on a particular scale, and variations on less is considered noise.
I understand what you mean, but here again there is a tie on the period (TF) which does not exist, you are counting the autocorrelation at some fixed length section.
I remember, how in the branch TP, a villager with the nickname bas assured that the autocorrelation coefficient is the key to the market. But, because he was doing it, after eating a lot of potatoes (and I can't stand root vegetables), I constantly vomited.
Sorry, what means "period (TF) that does not exist" I did not quite understand exactly.
I just continued our conversation from the previous page in the context that only bounces (extrema) are unambiguous in their essence, a bounce is either there or not, and the indicators, including autocorrelation functions, require a "period"(a fixed size of a moving window) in their calculations. But the optimal size of a "period", if it exists at all, is different at every moment of time, it is constantly changing.
We need either to search for optimal periods at each point of time (and use this platform to build some rules/system)
or to return to the extremums that are single-valued in their essence
or we are working with non-objective calculations
Karoch. it makes sense to dig in the direction that I voiced a few pages ago, the first experiments and already interesting results...
Listen buddy, do you really think these are cool signals? I feel sorry for you, I don't want to disappoint you, but the signals suck. There's a lot of trades in one trade or I don't get it. ***
Listen buddy, do you really think these are cool signals? I feel sorry for you, I don't want to disappoint you, but the signals suck. There's a lot of trades in one trade. Or I don't get it. ***
Sorry Dear Michael, I did not think that you would come here and see this shame, I'm really very embarrassed, well, since you're here, let me ask you as one of the most experienced and respected inhabitants of the branch of machine learning. What I'm doing wrong? In what direction should I go in order to earn as much money as you do and build the same excellent trading systems as you do?