Discussion of article "Thomas DeMark's Sequential (TD SEQUENTIAL) using artificial intelligence" - page 7
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
Yes, my output is looking ahead.
It is clear, I described above that it should not just look ahead but not overlap with the signs, the model is a simple Knn (nearest neighbours method), and everything is still under a big question due to lack of data and unknown algorithm for obtaining the target.
Target is simple, if the previous signal has a positive profit taking into account the spread, we mark 1 if negative, then 0. The work goes from signal to signal, and it does not matter what will be the next signal, to buy or sell, the main thing that the signal had a profit, if we were trading purely by strategy on the blue bought on the red sold.
Targeting is simple, if the previous signal has a positive profit including spread, we mark 1 if negative, we mark 0. The work goes from signal to signal, and it does not matter what will be the next signal, to buy or sell, the main thing that the signal had a profit, if we were trading purely by strategy on blue bought on red sold.
Well, in general on such a small sample Knn with ten nearest neighbours, gives almost 58% accuracy, on the test in non-sampling)))). So either the target has a little bit of past retourn or chips, or it's a grail!
What do you mean past retourned? I don't understand a bit.... How is this done?
When a signal appears, the input indicator variables are written to the window size, i.e. we look into the past. The output variable is taken from the future. That is, we don't know the result of the last signal until the next one appears.
Just the input information is used in such a way that it is the reason for the price, so there are fish here.
What do you mean past retourned? I'm a little confused.... how is it done?
When a signal appears, the input indicator variables are written to the window size, i.e. we look into the past. The output variable is taken from the future. That is, we don't know the result of the last signal until the next one appears.
Just the input information is used in such a way that it is the reason for the price, so there are fish here.
Depending on how you count the return or price increment, you can deceive yourself, it is necessary that returns do not "cross", for example, if you count the return by internal parameters of the candle (Close-Open)/Open, then two neighbouring returns are independent and you can for example N returns as fixtures, and N+1sign as a target, but if returns are counted between candles (Close(t-1)- Close(t))/Close(t-1), then you can no longer use N+1 return as a target, because it intersects with the N-th, you need to take N+2nd.
Well, if there are intersections, you can get anti-real scor. You should check it on a non-dependent target, so that the past is not interfered with, then the result will be clean.
data.zip
What is this dataset, if it's not a secret? How did you build it? I got 67% on the test, is that good or bad?
Well, I guess over 50 is good. But the dataset is skewed. When it is balanced in output (equal number of ones and zeros) then the results of the tester will coincide with your checks.
Depending on how you count the return or price increment, you can fool yourself, it is necessary that returns do not "overlap", for example, if you count returns according to the internal parameters of the candle (Close-Open)/Open, then two neighbouring returns are independent and you can for example N returns as fixtures, and N+1sign as a target, but if returns are counted between candles (Close(t-1)- Close(t))/Close(t-1), then you can no longer use N+1 return as a target, because it intersects with the N-th, you need to take N+2nd.
Well, if there are intersections, you can get anti-real scor. You should check it on a non-dependent target, so that the past is not interfered with, then the result will be clean.
That's a bit of a stretch. I don't have any interference. I make sure the data collection is clean. Your example applies to forecasting systems in the first place. But for the sake of argument, I'd say your remark is quite appropriate. It refers to the section about the purity of data collection. If you make a small mistake, the results will be radically different and will not correspond to reality.