Thoughts on the random
The point is that the market is not a pseudo-random sequence or a random sequence. There are patterns in the market. And if there are patterns, it is no longer a random sequence.
A clear example of this is trends and flat patterns. These are regularities.
Therefore, it is useless to generate something in the subject of randomness and compare it with the market....)))
1) No :https://www.mql5.com/ru/code/8790
2) Possibly somewhere yes : https://c.mql5.com/mql4/forum/2012/11/predict.gif
Good afternoon!
I'm writing this and wondering how not to offend or provoke anyone into flubbing. I hope to be constructive, and, I'm just asking (not proving, not refuting, just wanting a dialogue).
If you take a series of quotes for many years and create on their basis a file of zeros and ones: zero - if the next price is greater than the previous one; one if vice versa - you get a pseudo-random sequence. Carefully call it with prefix "pseudo" for the time being.
Further, we generate ideal trades based on the pseudo-random sequence: if 1, we buy and exit on the next bar, if 0, we sell and exit on the next bar. The resulting equity chart is almost a flat line directed upwards (including the spread).
Now, a question: if we try to repeat our pseudo-random sequence using the Monte Carlo simulation expecting to reach the same result as in the initial step, i.e. ideal entries, what will we get? Let's calculate: there are 60,000 hour bars, hence there are 2^60,000 (!) different possible rows of zeros/units. Only one of them perfectly describes the inputs. It's kind of clear that even if we load the comp with 100 trillion generations, we probably won't get the desired result. Each time, our resulting equity will resemble a drain at the rate of spread. And it (the result) is there in nature! We have it on our history. In other words, we pant, count and smoke, we find nothing and say: "Ok, the problem is not solved, I'm going to bed. Doesn't it remind you of the problem of the probability of life in our universe? It seems to have comparable probability values in number of orders of magnitude.
I've laid out the general context, there's a lot to think about. What class of problems, so to speak, does my idea belong to?
Somehow I had a similar idea myself. Imagining a quote as a binary series, is it possible to decode the process that generates it? Technically, a pseudo-random sequence is generated by a shift register with linear feedback (RSLOS). So our decoding task is to find the LCLOS that generated our pseudo-random sequence. Such problem is solved by Burlecamp-Massey algorithm. I tried to decode a price quote using this algorithm but it did not work, although it didn't take much time. Interestingly, if you don't replace analog values with binary ones and try to decode the generating process of our analog pseudorandom price series, you can use the sameBurlecamp-Massey algorithm.In this case, the generating process will be Prony's autoregressive model x[n] = SUM a[k]*x[n-k]. Apart fromthe Burlecamp-Massey algorithm, the Levinson-Durbin algorithm would be more robust. The problem with Prony's analog AR model is that it is unstable unlike the binary RSLOS and its predictions can quickly go to infinity. We can overcome the instability by assuming that our pseudo-random quote has noise. Then instead of an AR model reproducing all historical data with zero error, we may use an approximate AR model solved for example by Bourg's method. This is an econometric problem. It is interesting to note that finding the exact Prony model is equivalent to fitting the exponential sum SUM C[k]*EXP(B[k]*k) into our series, where B[k] can have both negative and positive real part (positive part leads to instabilities). Burg's approximated AR model solves the same problem by fitting damped exponents. In short, by taking the path of decoding a price series, we arrive at econometric AR models.
will take off at the same speed as the track
Velocity relative to what?
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Good afternoon!
I'm writing this and wondering how not to offend anyone or provoke a flood. I hope to be constructive, and, I'm just asking (not proving, not refuting, just wanting a dialogue).
If you take a series of quotes for many years and create on their basis a file of zeros and ones: zero if the next price is greater than the previous one; one if vice versa - you get a pseudo-random sequence. Carefully call it with prefix "pseudo" for the time being.
Further, we generate ideal trades based on the pseudo-random sequence: if 1, we buy and exit on the next bar, if 0, we sell and exit on the next bar. The resulting equity chart is almost a flat line directed upwards (including the spread).
Now, a question: if we try to repeat our pseudo-random sequence using the Monte Carlo simulation expecting to reach the same result as in the initial step, i.e. ideal entries, what will we get? Let's calculate: there are 60,000 hour bars, hence there are 2^60,000 (!) different possible rows of zeros/units. Only one of them perfectly describes the inputs. It's kind of clear that even if we load the comp with 100 trillion generations, we probably won't get the desired result. Each time, our resulting equity will resemble a drain at the rate of spread. And it (the result) is there in nature! We have it on our history. In other words, we pant, count and smoke, we find nothing and say: "Ok, the problem is not solved, I'm going to bed. Doesn't it remind you of the problem of the probability of life in our universe? It seems to have comparable probability values in number of orders of magnitude.
I've laid out the general context, there's a lot to think about. What class of problems, so to speak, does my idea belong to?