Market etiquette or good manners in a minefield - page 52
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
And the length of the error vector is something new... we didn't go through -:)
How do you look for the length of the side of a triangle? That's right! - You take the sum of the squares under the root.
Same here, only you normalize it by the length of the data vector. So this is nothing new. It's all scientific.
Here is an example of one of the 'compression' cycles across all weights:
I also have all the weights in the same array and the indexing is about the same
You're predicting the "colour" of the future +1 bar or you're estimating the movement more accurately, using the history of the current bar?
Understood, thank you. Tell me, how do you estimate the optimal result of your NS, i.e. how many percent of bars of the total volume of the experiment will be successful? Well, for example, 99% of 1000 bars predicted correctly. What is your estimation?
So... while you're thinking about the question (and given the time difference with Novosibirsk, you're probably asleep), I decided for the sake of curiosity to try to predict the colour of the bar using the AR method. I tried to predict "lob" and then I tried to predict the bar colour. I tried to predict "directly" direction ("+" or "-") of x[n]-x[n-1] (H+L)/2 without model identification. Similarly, as I expected, it is rubbish, because one cannot do it at once. But I remembered one old idea of series processing and obtained any experimental result (on 15 min of EURUSD 5 000 samples):
The most disappointing thing is that there is no error... But you're right, Serega. Knowing the "direction" in a bar and the mean square "bar breathing" (for some mysticism) we can build a good strategy. So what are your results? You must have estimated how much the system is allowed to lie, right?
Understood, thank you. Tell me, how do you estimate the optimal result of your NS, i.e. how many percent of bars of the total volume of the experiment will be successful? Well, for example, 99% of 1000 bars predicted correctly. What is your estimate or what is the existing result.
If we want to predict the direction of the expected movement only, we should estimate the percentage of correct predictions using the formula: p=n(+)/N, where N is total number of experiments, n(+) is the number of correctly guessed signs.
If we are talking about predicting the expected movement taking into account its amplitude, we can correctly estimate the forecast reliability by building a forecast cloud by the algorithm described in this topic above, and draw a straight line through it using the method of least squares. The tangent of its slope characterizes the prediction accuracy of the selected algorithm, while the dispersion of experimental points shows the risks.
Here is an example of such a plotting for the training sample (red) and for a test sample (that didn't take part in the training):
You are right, Seryoga. Knowing "direction" in a bar, mean square "bar breathing" (for some mysticism) we can build a good strategy. So what are your results? You must have estimated how much the system is allowed to lie, right?
I did. That's what the first part of this thread is about. Read it. According to results of this estimation, if the system gives statistically reliable advantage (>50% of correct entries), then it unambiguously defines optimal parameters for MM, allowing to take the market to the maximum.
P.S. It's amazing how much time it took you to finally understand it! And how many insulting nicknames have I heard from you in the same time? And how many more of them will I have to listen to...
to Neutron
Если говорить о прогнозе ТОЛЬКО направления ожидаемого движения, то оценивается процент правильно угаданных движений по формуле: p=n(+)/N, где N - полное число экспериментов, n(+) - число правильно угаданных знаков.
If we are talking about predicting the expected movement taking into account its amplitude, we should correctly estimate the reliability of the forecast using the algorithm described in this topic above and draw a straight line through it using the method of least squares. The tangent of its slope characterizes the prediction accuracy of the selected algorithm, while the dispersion of experimental points shows the risks.
Here's an example of such drawing for the training sample (the red one) and for the test one (that didn't take part in the training):
Estimated. The first part of this thread is devoted to it. Read. According to results of this estimation, if the system has a statistically significant advantage (>50% of correct entries), then it unambiguously defines optimal parameters for MM, allowing to squeeze the market to the maximum.
Can you tell me clearly what your guessing rate is on your super duper NS system. Besides, you said you were only predicting colour, what's that got to do with amplitude? That's not what the topic was about. You don't have to force me to read everything, especially your numbers might have already changed.
P.S. It's amazing how much time it took you to finally understand it! And how many insulting nicknames have I heard from you in the same time? And how many more of them I'm gonna get...
Seryoga, you have a damaged psyche and inflated ego. Hurry up and don't write such nonsense. Also, you have a short memory and have forgotten how you called normal and basically good people nut jobs. Aye-aye, doHur.
PS: It's amazing how long it will take you to finally understand that pyremy AR models will not give you worse results compared to your NS.