Market patterns - page 35
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
Not exactly dumb, but it's not the most complete representation, a two-component vector can only be used to check the prediction at a fixed distance from the current bar, otherwise you have to predict N bars ahead with exponentially decreasing probability. But I agree that for a particular TS we need only two numbers, the estimated potential for price shift and its probability, the first component defines the price target and the second one defines the scale of risk.
In general I agree, having such a distribution, by calculating the correlation with the actual bias potential (with hindsight), one can quickly and accurately understand the predictive potential of the pattern, without classical testing.
1.The continuity of the FWR is a forced assumption to make it easier to apply the mathematical apparatus. The nullity of this assumption with sufficient sampling has subsequently been found out.
2. The assumption is that the function D(t) can be either positive or negative and then everything converges to your reasoning about positive and negative feedback.Lucky_teapot what he says, any extrapolator is easily checked by building a fixed step prediction indicator from it (calculated on tests). For example you predict price for N(=const) bars ahead based on current information. That is, for each bar at point t, you use only price data not later than t-N. You build the indicator and it becomes clear where and how you missed.
Not exactly dumb, but it's not the most complete representation, a two-component vector can only be used to check the prediction at a fixed distance from the current bar, otherwise you have to predict N bars ahead with exponentially decreasing probability. But I agree that for a particular TS we need only two numbers, the estimated potential for price shift and its probability, the first component defines the price target and the second one defines the scale of risk.
In general I agree, having such a distribution, by calculating the correlation with the actual bias potential (with hindsight), one can quickly and accurately understand the predictive potential of the pattern, without classical testing.
Lucky_teapot what he says, any extrapolator is easily checked by building a fixed step prediction indicator from it (calculated on tests). For example you predict price for N(=const) bars ahead based on current information. That is, for each bar at time t, you use only price data not later than t-N. You build the indicator and it becomes clear where and how you missed.
Sorry, didn't notice the reply((
This topic has been looming in my mind as a vague concept since the beginning. But so far it's hard to see how to do it practically.
It would be a clear and standardised representation of the output of the prediction system, which is then easy to check for quality on the fly without resorting to testing.
The idea is that the indicator output should be converted into signal(s) or expected price shifts in the future, which are then compared to the retrospective real ones, the closer the better, and the cumulative total is then calculated.