Market etiquette or good manners in a minefield - page 94
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
You need to look at the prediction plausibility for the different lengths of the pattrines.
Here's what we get with the plausibility of the expected direction of the open position for the 3 entry classifier (red - CB, blue - kotier):
We can see that the following table of probable states is built for the pattern (Fig. right):
1. 000 -> Buy
2. 001 -> Sell
3. 010 -> Buy
4. 011 -> Sell
5. 100 -> Buy
6. 101 -> Sell
7. 110 -> Buy
8. 111 -> Sell
In other words, don't get wise - the direction of the open one is determined by the direction of the last PT segment and is counter-directional to it. This fact confirms the statement that the optimal TS in the market is a trivial reversal. That is what is described in detail in Pastukhov's work. The only point that deserves special attention is that now. having this classifier in hand, one can confidently single out areas where the market can be characterized as 'efficient' and keep a low profile. I mean those npatterns, for which the relative prediction probability is low (for example number 2 and 7). Or on the contrary, wait for "arbitrage" patterns (000 or 111) to appear and play "for sure". By the way, patterns 4 and 5 are also essentially arbitrage patterns (100 and 011 see fig. right).
Although, I do not rule out the case of a non-trivial prediction. This is when the probability series is not alternating. Here, the reversal TS will not work or will show a low return, but the classifier will be on top. You need to look at the kotir at different horizons of H partitioning. The classifier is by definition the most powerful tool in our hands. No NS will give a more reliable prediction concerning the expected movement, and there is nothing better than NS, because it requires a priori knowledge about the market for model building.
In general, just think, colleagues, Pastuhov in his work thoroughly examined a trivial case for a one-way classifier - "to open every time to the opposite side", and gave some hints for the analysis of a more common case - the patterns. If we try our best, we can obtain a universal classifier of market conditions that can reconstruct the "market mood" with a certain accuracy. And considering the "limitless" possibilities of pattern analysis (the length of patterns used is not limited), we can let our mind run free!
This pleases me immensely! In fact, everything makes me happy!
I'm going for a beer.
In general, think about it, colleagues, Pastukhov in his work thoroughly analyzed a trivial case for one-way classifier - "to open every time to the opposite side" and gave hints for the analysis of a more general case - patterns.
On the background of the general and private joy, imho, it is worth remembering that conclusions in the paper are made for a stable H-volatility.
Hence the task of choosing such H, for which volatility will be stable within the trading period.
Perhaps we should return to the previous question
about a choice of method to determine such H?
Amidst the general and private joy, it is worth remembering, imho, that the conclusions in the paper are made for stable H-volatility.
Hence the task of choosing such H, for which volatility will be stable within the trading period is formulated.
Perhaps we should return to the previous question
about a choice of method to determine such H?
There are only two methods: statistical selection and empirical selection. I tend to think that H should be a multiple of a spread for a certain instrument of a certain brokerage company. Big values of H will "blur the picture" and increase risks, small ones lead to scalping. According to my chart it turns out that using of H above 4 spreads is not reasonable, but the average rebate in this case is equal to 4 spreads. It's not much, but if regular + reasonable MM it should work.
There are only two methods: statistical selection and empirical selection. I tend to think that H should be a multiple of a spread for a certain instrument of a certain brokerage company. Big values of H will "blur the picture" and increase risks, small ones lead to scalping. According to my chart it turns out that using of H above 4 spreads is not reasonable, but the average rebate in this case is equal to 4 spreads. It's not much, but if regular + reasonable MM it should work.
I think you can/should empirically set reasonable boundary conditions, e.g:
And further I would like to use statistical method, but not as a result of TS functioning, but as an analysis of typical instrument movements.
For the beginning we can leave MM and set lot = const.
It is clear that the total profit = CoTrade*CalculatedProfit.
At the same time, CoLTrade == H-inversion.
Hence the question:
What should be the statistical constraints (MO, variance, RMS, etc.) to determine H-inversion ?
to Neutron
Turns out the ticks you sent in weren't for half a year, but only 21 days... that makes a difference. So you can use H more (400+ ticks for 21 days is even a lot).
What should be the statistical limits (MO, variance, RMS, etc.) for determining the H-inversion ?
Yes, of course boundary conditions are important and necessary. It makes no sense to cut the cotier with a threshold of less than one medium (or even two) as there is nothing to catch on such a slicing.
I take the liberty of putting forward another assumption for the consilium to determine the upper boundary condition:
Let's consider a simple question: how many transactions per day (maximum) should a wise trading robot perform? The question is not idle and can be found out statistically, but at first glance - maximum 2. That allows us to "bind" H from the bottom to the spread, and from the top to the statistics of the amplitude of daily average movements. By experiments with one-layer NS the significance of predictions on 23+1 inputs (for hourlies) has been checked.
Thus, for the data which I have at the moment, the acceptable size of H is 27 pips, which gives an average of 2 transactions per day during one month at quite non-pips payoffs.
I once did a "strict" estimation of the left boundary for H (I solved it analytically) and it turned out that Hmin=2*Spread. Maximum yield of Cagi-strategy lies in the region of 3-5 Spreads, then there is a monotonous decrease in the yield defined as the average yield per unit time (human). The maximum can only be determined experimentally (it depends on the predictability of the instrument) and is not a stable characteristic of the H partition horizon.
I once did a "strict" estimation of the left boundary for H (I solved it analytically) and it turned out that Hmin=2*Spread. Maximum yield of Cagi-strategy lies in the region of 3-5 Spreads, then there is a monotonous decrease in the yield defined as the average yield per unit time (human). The maximum can only be determined experimentally (it depends on the predictability of the instrument) and it is not a stable characteristic of the H partition horizon.
The maximum can only be determined experimentally (it depends on the predictability of the instrument) and is not a stable characteristic of the horizon of a split H. I have recently come home from work and read a bit of the thread. Now for the topic.
====
Question paralocus: is it finally time to abandon the uncorrected attempt to measure spreads?
The correctness rebuttal is simple, on the contrary. If the unit of measurement is correct, then from
Neutron's estimations implies - "let's all run to trade USDZAR!!!" (spread is 100). Why. 2 trades per day = 1000 pips.
Does it make sense? Do you believe it? Easily verified by your tools. Good luck. I already have a queue in case of it.
// Maybe I missed something in terms of the correctness of the counter-example, like "the market is wrong"?
I mean that the subject is promising, so I want to avoid unnecessary gluconautics. It is not necessary.
Question paralocus: Isn't it finally time to abandon the uncorrected attempt to measure spreads?
The rebuttal to correctness is simple, from the contrary. If the unit of measurement is correct, then from Neutron's
Neutron's estimations implies - "let's all run to trade USDZAR!!!" (spread is 100). Why. 2 trades per day = 1000 pips.
Does it make sense? Do you believe it? Easily verified by your tools. Good luck. I already have a queue in case of it.
// Maybe I missed something in terms of the correctness of the counter-example, like "the market is wrong"?
I mean that the subject is promising, so I want to avoid unnecessary gluconautics. It is of no use.
You've got to use your head, mate. You should think with your head. What's the point of trading with a tool that few people use? For them other regularities will work, and for mass currencies - you can check what brokerage companies trade (I hope you know you're not trading with the market ...). And all skepticism will disappear at once (and maybe something else... different things sometimes disappear...). I know how much spreads and what currency my brokerage company sells, don't you?
1. You've got to use your head, mate.
2. What is the point of trading a tool that few people use?
3. If you have decided to trade with an open market, you may start with an open trade. If you have decided to buy an open trade, you may start with an open market.
4. Create a simple distribution function in Matcadet, to find out the value of the most popular EUR movement (for example, you should verify the prices of a brokerage company for which you are calculating). And all skepticism will disappear at once (and maybe something else... different things sometimes disappear...).
5. For example, I know how much spreads and in what currency my brokerage company sells, and you?
1. Yes, I think so. Or are you hinting that you have to think? :)
2. What's the point of speculating how much she's used if we've hacked her? :)
3. Have you checked that other patterns work there? Or is it just the kags that don't work there?
4. I don't need it. the most popular move = 1 pip. I see it every day on my indicator.
5. I have no idea. What a horror!!! :) How much?
And another question of a personal nature. :) Can you describe the scheme of ticks in your terminal?
Where do you think they come from?
Then we'll discuss it. All very interesting, but it's necessary to urgently and thoroughly deal with the unit.
// You see, you offer others to prove theorems and you yourself want to take a ride on a ponce with no proof. :)