The market is a controlled dynamic system. - page 237
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
True,there is no benefit to such predictions. --- That's right ;)))
A lot of options, but none of them are useful. And if they're useless, why the hell do we need so many options?
In itself this kind of prediction is not necessary, not because of non-stationarity of the market etc., but because the aim is not to predict the price but to make profit. But these rules may well be a part of the strategy.
For example, let the price is now 1.3000. The forecast: the price will reach the level of 1.3100 before 1.2800 with the probability of 0.75. I.e. if we set a Take Profit figure and a Stop Loss of 2, the Take Profit will trigger twice as often. Everything is there in this prediction and even the probability seems to be in our direction. Everything but the profits. The model should be such that we have a statistical advantage and the target of the prediction will be it, not the price. I.e. we predict that our model/system has a statistical advantage and on a series of trades we have a high probability of making a profit.
For example, we have a system with rigid tp=sl. On tests, probability tp=0.55, sl=0.45. MO=0.1. The forecast target is that this MO will remain positive. Although we can say that the probability of tp will remain >0.5
p.s. I don't trade such systems))
There is a probability of precipitation.
Yes, because it is a binary event (either there is or there isn't) and the only information is probability. Forecasters have nowhere to go)) And for temperature, the standard confidence intervals matrix applies. Forecaster chooses probability level of forecast, for example 0.9. Then he uses his forecast model to get probability distribution and selects interval bounds so total probability of getting into this interval is 0.9. We get only upper and lower bounds of the interval. This is the predicted temperature range
Trying.
A forecast for any controlled dynamic system is information for assessing the situation - no more, but no less.
At the same time, the stage of assessing the situation is directly followed by the stage of making a decision, so the forecast is interesting not in the sense of how things will end, but only in the sense of what has begun.
Hence, the consequence: it is meaningless to use the result of the forecast to determine values of TR and SL, but it is stupid not to use it when making decisions about opening or closing positions.
To twist Carl Clausewitz, let me try to quote him: "...Military science should not accompany the commander directly on the battlefield, it should only prepare him for this battle...". Something similar (:
there are no separate predictions at the point of entry, etc. There is a system - an algorithm that can give a statistical advantage. When one enters the market, one makes a prediction/betting that this algorithm continues to make a profit (has a statistical advantage). No separate predictions in other points are needed. They are only a consequence of the above prediction.
For example, an angler knows that a particular bait, method of hooking the fish, etc. gives him an opportunity to catch enough fish to cover the overhead most of the time. And he doesn't need to predict anything in each individual casting of the float.)) He stupidly fishes and changes his methods if they stop working. What does he predict? Nothing. Although he reacts in dynamics, not by the algorithm - throw, wait 2 minutes, pull it out)).
Not necessarily to you. More to the forecaster in general. More specifically, the forecaster who very often, in and out of place, repeats like a mantra "probability-probability-probability-probability...".
You don't reveal the functional relationship of F(). Don't think I'm trying to tease it out -- no. But I suspect, (and something tells me ;)) that it's a regression... and that...
It's a well known pattern. It has been known for a long time. A lot has been said about it since the 1960s, considered up and down. It works in the steady motion sections. Problems arise when the trend changes. Especially in the switching mode. Many works have been devoted to this problem. But there is no satisfactory solution until now. Thus, the limits of its satisfactory work are known.
But there is a nuance. If your model does not operate with probabilities, then there is no ground to speak about probabilities on the basis of your model. Unless you post facto calculate the frequency of the model result in some compact region. But these are, generally speaking, crutches.
I do not see a problem with regression. It is convenient and universal - it is suitable for any function. The problem is not the regression, but the input and output data. If the input and output is price (autoregressive), nothing will work. The price does not depend on itself. That's why there are so many unsuccessful attempts to create market models with a single input: the price and its indicators. You need to select inputs that affect the price, i.e. go beyond the quotes and their indicators. For example, the share price fell on news of a company's losses, i.e. the losses affected the share price and not what the price was doing before. A company's profitability depends on its economic performance. Although individual companies have their own private assumptions about their profitability (contract losses, etc.), the economic sector or the entire economy is fairly well predicted by regression models.
The statistical advantage is only gained in a series of trades. For that there must be clear starting conditions for opening and closing a position on time, no matter whether it is profitable or unprofitable. Anyone who has played poker in tournaments will understand me. Over the distance mathematics plays a major role. Even a series of small losses are overridden by big wins with significant profit accumulation. It's the same in the market. It is only important to choose the right price range in which you can operate. The TF is not important. The price range and price analysis for entry and exit are important. And the discipline of course. At the beginning it was not so easy to close losing trades. Loss trades are inevitable in any TS and they should be treated as an element of this TS.
You don't reveal the functional relationship of F(). Don't think I'm trying to tease it out -- no. But I suspect (and something tells me ;)) that it's a regression... and that...
Why would it be regression? Curvafitting again...
Yes, because it is a binary event (either is or is not) and the only information is probability.
Why this regression? Curvafitting again...
What prevents us from making a TS that will operate only with probability? I know it's a bit cumbersome, but that doesn't make it not work.It's okay if it works.
Avals:
Perhaps only for fixed stop and take systems can we talk about frequencies and prediction. Most systems do not know in advance where the exit will be and therefore bet on a different market event each time.
I am trying it out.
A forecast for any controlled dynamic system is information for situation assessment - no more, but no less.
At the same time, the stage of situation assessment is directly followed by the stage of decision-making, so the forecast is interesting not in the sense of how things will end, but only in the sense of what has begun.
Hence, the consequence: it is meaningless to use the result of the forecast to determine values of TR and SL, but it is stupid not to use it when making decisions about opening or closing positions.
To twist Carl Clausewitz, let me try to quote him: "...Military science should not accompany the commander directly on the battlefield, it should only prepare him for this battle...". Something similar (:
Well, in general, putting my question, I expected to hear an answer not for UDF, but for UDF opponents, or rather, not even opponents of UDF, but just supporters of the common view of the market as a random phenomenon, subject to the probabilistic concept.
.
In my approach [Market - UDS], a tracking system is the optimal tool for detecting the structure of a movement. Thus, already in the process of identifying the structure, a vector of UDS state is formed. TS in relation to UDF is a superstructure, and here it is quite natural and legitimate to use the vector of UDF state to form the TS output.
I understand that a clear explanatory picture, compactly demonstrating the essence, would be appropriate here.
I don't see a problem with regression. It is convenient and versatile - suitable for any function. The problem is not the regression, but the input and output data. If the input and output is price (autoregressive), nothing will work. The price does not depend on itself. That's why there are so many unsuccessful attempts to create market models with a single input: the price and its indicators. You need to select inputs that affect the price, i.e. go beyond the quotes and their indicators. For example, the share price fell on news of a company's losses, i.e. the losses affected the share price and not what the price was doing before. A company's profitability depends on its economic performance. Although individual companies have their own private assumptions about their profitability (contract losses, etc.), the economic sector or the entire economy is fairly well predicted by regression models.
This scheme is more complicated than autoregressive one, but it is quite suitable for traditional testing and optimization, so the results will be quite clear.
In itself, such a forecast is not necessary, but not because of the non-stationarity of the market, etc., but because the purpose is not to predict the price, but to make a profit. As part of the strategy, these rules may well be in place.
For example, let the price is now 1.3000. Forecast: the price will reach 1.3100 before reaching 1.2800 with probability 0.75. I.e. if we set a take profit figure and a stop loss of 2, then the take profit will trigger twice as often. Everything is there in this prediction and even the probability seems to be in our direction. Everything but the profits. The model should be such that we have a statistical advantage and the target of the prediction will be it, not the price. I.e. we predict that our model/system has a statistical advantage and on a series of trades we have a high probability of making a profit.
For example, we have a system with rigid tp=sl. On tests, probability tp=0.55, sl=0.45. MO=0.1. The forecast target is that this MO will remain positive. Although we can say that the probability of tp will remain >0.5
p.s. I don't trade such systems))
You often hear that from analysts ("...price will reach the level of 1.3100 before the level of 1.2800with the probability 0.75...").
But who, how, when calculated these probabilities ??? No one ever makes such probability calculations !!! In such cases it's appropriate to speak not about probabilities of price movements, but about the analyst's expectations or targeted influence on the audience. That is, the substitution of notions is made here - either out of misunderstanding, or to deliberately mislead the public.
etc.