Right now trying luck on this beauty, ID 543435 {image}
Ignored
We've just made some upgrade (fixed a couple of errors) and had to recalculate models, so this ID is no longer connected to the propper model, sorry...)
Service development ideas.
1. Telegram bot that will be able to: - report the appearance of new models; - report on model events (reaching calculated 6 levels, reaching goals levels, etc.).
2. We want to give a communication opportunity for users so that they could exchange forecasts
- vote on whether the model will work or not;
- ability to share comments on models. Also we maybe will create users rating. It will depend on correctly predicted models, so other users will see if the comment or forecast on the model comes from an expert or not.
3. Display of interrelated models for each one.
4. An option to view multiple models on the same chart.
5. An option to hide models from the table using filters. For example to hide clearly not proportional models (which usually doesn't work out).
6. Several pre-prepared options for chart scaling.
7. Notes to (any) models ; - personal; - public ones.
8. Selecting multiple models on the same chart (located side by side).
9. Linking multiple charts for multi-timeframe analysis. To be more specific:
- link models using the ID in the comments;
- make projections of levels 6 and other reference points of linked models on each other's graph;
- specify the location of linked models relative to each other. All these links should be displayed in the form of a tree of links.
10. To use neural networks to determine the working models. I will reveal this point in more detail in the next post.
The idea is to estimate the probability of a reversal from the level of the calculated 6th with the help of neural networks. Models can predict accurate support and resistance levels, and the system works the same way on most financial instruments and timeframes. The rules for building models are strictly formalized and cover all the variety of market situations in several types of graphical models.
The idea is to evaluate the probability of working out the model levels using neural networks.
So here's what we got:
1. 7 types of models (they can be divided into 16 subtypes).
2. At the moment, we have accumulated a couple of hundred parameters that can be used to describe the structure of each model, their mutual position and interaction. Some of them refer to several types of models, while others refer to only one of them. Some are described by discrete values, while others are described by constant values. In any case, everything is formalized.
3. We have a mechanism for collecting statistics on historical data, both for parameters and for working out / not working out models. So far, about 30% of the parameters are implemented in the code, and collecting statistics for 5 million bars takes about a day and gives ready-made statistics for about 1 million models.
Thus, we have ready-made and rather carefully developed approaches to preparing the source data and specific requirements for the output data (there are several options, but the very first in the queue is the question "if the calculated 6th level of a particular model is the level of a trend reversal?").
16.05.2021
Recently, we have made a number of non-striking changes, and starting tomorrow, a slightly modified algorithm will work for all models except EAM. The changes relate to the models detection sub-algorythm in case when new point 4 is found. To briefly describe the essence of the changes, the models will be processed more accurately But in the near future, bugs may be noticeable, because we have not yet had time to recalculate the previously found models in the database
Some news about what is happening with our project
1. This week we plan to make the first public transactions. We do not promise anything supernatural, at the moment we just got satisfactory statistics on the history from a couple of ideas and now we will make an initial run on a real mini-account to "feel how it works".
The format and expectations are as follows:
- trade with pending orders from the calculated 6 points to the 1st goal;
- approximately half of the orders will not work and will be canceled before the trade is opened;
- we expect 1-2 deals per week;
- in total, we plan to conduct 15 transactions at this stage;
- approximately 40% of them, as we hope, will be profitable.
2. The interface has been changed, but the innovations are still in a non-working state
- the "renew table" button has appeared. Soon, when you click it, the table with models will be updated automatically, you will not need to reload the entire site. By default, the button will be enabled. When you select a model, it will turn off and auto-update will be paused until the user clicks the button again.
- the "Transaction Calculation" tab appeared in the menu on the left. When you click on it, you can see a form in which you will soon be able to calculate the parameters of concluding transactions. This will help you observe the maximum risk for a single trade, as well as calculate more accurate entry levels/closing the transaction based on the quotes of your broker / dealer. Detailed instructions for use will be given a little later
3. Also, another innovation will soon appear - when you select a particular model, the chart will display not only the model itself, as it is now, but also the models associated with it in 2 different categories:
- previous models;
- embedded models.
What are these categories we will explain soon These models will be rendered in muted colors not to spoil the perception of the selected model.