Shall we discuss an expert based on coding? - page 2
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
In principle, the Expert has already been implemented, but I would like to discuss the idea and the model, I would like to hear criticism and weaknesses of the model.
If you are ready to show the tester's report for a year, for example... Let's have a look.
I would like to discuss, maybe someone also wants to implement a similar, we are still excited by the results, but experience suggests that somewhere there is a pitfall, that's just where?
In fact, coding has nothing to do with it, it's just a transition from analogue pattern to digital one. You can get by without it, e.g. use expert by K-neighbour method and run it. The effect will be the same. I used both digital and analog patterns; I tried it both ways for a long time. I don't have any luck with trading, though my device may work well in some periods. It may work one whole month with profit, in fact without losing trades. Then in few trades it will sell out in 3-4 days.
The conclusion is as follows: market movements can be divided into two types - normal and abnormal. Normal movement makes up to 70% of the entire movement of the tool. It can be easily forecasted and is suitable for earning money using various advanced TA tools and similar statistical methods as well. Abnormal movement makes up to 30% of movement and everything works contrary to any statistics. We earn 70% and lose 30%, so we are in good shape? But it is not so. Abnormal market movements turn out to be more volatile and easily eat up 70% of the normal movement. I still do not understand how to timely learn to detect these types of market movements.
Working with this tool I have to solve several problems, each of which is good even separately. For some period of time we should get statistics (per day, per month, per 10 years), how detailed should be AB or ABCDEFGH pattern, how many patterns should be found on history for getting meaningful statistics, and the pattern itself (you have an indicator, or you can just code combinations of candles etc), and in principle how to interpret the pattern statistics (sometimes it is better to trade in the opposite direction)
Anyway, searching for answers to these questions led me to NS and now statistics is only an auxiliary tool.
If you're ready to show the tester's report for the year, for example... Let's see.
70% we earn, 30% we drain, are we happy?
If you're ready to show the tester's report for the year, for example... Let's have a look.
Alas, I don't trust the tester, I've been convinced of that many times, although I may not understand it, I trust Excel, it's all visible and understandable, I'm a fish in water with it
This is tantamount to flipping a coin. Why such a crude example? Because indicators have never predicted or even helped to predict the future.
If you read the topic carefully from the beginning, you will notice that prediction was never even intended. It was about analysis by applying a large number of different tools to a pattern of several candlesticks. In fact it is a variant of the trend following trading strategy in its original design. And it is not about any prediction, especially about 49%.
You should be more careful before expressing your opinion on the subject.
If you read the thread from the beginning you will see that prediction wasn't even a part of the plan. It was about analysis by applying a lot of different tools to a pattern of several candlesticks. In fact it is a variant of the trend following trading strategy in its original design. And it is not about any prediction, especially about 49%.
You should be more careful before expressing your opinion on the subject.
Well, not exactly, we do assume that the probability of the codes will remain for some time, so it turns out that we predict that after code X there should be a rise (fall), although this may not be true or even not at all, there is no guarantee that the probabilities will remain
I don't see anything trend following in this system, just a stupid indicator pattern with positive statistics, no guarantees, no trend, we have no idea what the trend is at each current moment of code arrival
Excel analysis does not take the spread into account. And an expanding spread can fundamentally change the results of a strategy.
Excel analysis does not take the spread into account. And an expanding spread can fundamentally change the results of a strategy.
Why? You can take the spread into account. Simply, we will need to record the deal history in Excel formulas, in which any spread can be calculated. In addition, the value can be generated randomly. With any spread. By the way, I should try it. It is possible to do the calculation immediately in MT in case of importing the history of trades into the file, but then it will not be possible to view different variants simply by changing the value in Excel.
The idea is good, I've tried it, except that the probability of winning with enough history on a long enough period will always be 49% (+-2%). Consider rigidly adjusting the period of statistic selection, but this would lead to complex patterns being caught rarely and paying attention to 5-30 past pattern repetitions makes no sense, too little for statistics. At the same time I would advise to open in the opposite direction from the signal, because the balance should be restored and if at the moment the signal says buy and 65% of history says it was so, the statistics should come to 49% and 65% is a deviation from the normal value. This is tantamount to flipping a coin. Why such a crude example? because indicators have never predicted or even helped predict the future.