MetaTrader does not reflect reality ! How can I fight this ? - page 19
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
Generally speaking, the most pressing question for me is how to assess the future performance of the TS.......
It is not possible to answer this question accurately.... We can assume that the TS will work in the future.... The only thing left for us is to monitor its performance according to various criteria.... Again, the key parameter is the maximum drawdown... Why?
It is the only parameter that does not change with every new trade... If it has changed, then there is a reason to think about what to do next...
I would phrase the question differently: After optimization what are the most workable variants?
In general, for me the most pressing question is how to assess the performance of the TS in the future....... There is no problem with TSs, the only problem is how to understand how this TS will work in the future....))))
It is necessary that during the optimization of each parameter the optimization graph changes smoothly from value to value. It must be clear that the system is working, while the optimization is not adjusting, but, as follows from the name of this process, the selection of optimal parameters for an already working system. In other words, if the market changes, the system will not start losing on these parameters, but simply brings non-optimal profit.
I.e. all you need to do is to write a system that is profitable, and through optimization choose the optimal parameters for each trading instrument. And, of course, the less such parameters - the better.
It should be possible for each parameter to be optimised so that the optimisation graph changes smoothly from value to value. It should be clear that the system is working, while the optimization is not adjusting, but, as its name implies, selecting the optimum parameters for an already working system. I.e. if the market changes, the system does not start losing on these parameters, but simply brings non-optimal for itself profit.
That is, all you need to do is to write a system which is profitable, and through optimization to select the optimum parameters for each trading instrument. And of course, the less of these parameters the better.
I agree. A practical proof of this is a wide optimal zone and its relative stability for all parts of the history.
Isn't half a year of TC not enough? On an hourly timeframe that's about 3168 bars.
It's probably not about time. For me, the most important parameter by which we can more or less confidently judge the quality of our judgments about the TS is the number of deals in the testing area (or OOS). It is rarely paid attention to, but it is very important, because statistics rule in trading.
The lower is the number of trades, the higher should be the profit factor to conclude that the system will work in the future. If a longer plot with a number of trades of the order of a thousand will show a profitability of 1.65, it seems to be already a serious bid. I am doing research in this area and am going to post an article on the subject soon. About 70% of it is written.
P.S. Leonid, I am closely following your PAMM account. Everything is going great for you.
It should be possible for each parameter to be optimised so that the optimisation graph changes smoothly from value to value. It should be clear that the system is working, while the optimization is not adjusting, but, as its name implies, selecting the optimum parameters for an already working system. I.e. if the market changes, the system does not start losing on these parameters, but simply brings non-optimal profit.
In other words, all that is needed is to write a system that is profitable, and by optimizing it to find the best parameters for each trading instrument. And of course the less of these parameters - the better.
With a significant change in any macroeconomic parameters (such as interest rates), the system "without any hint" can go to the drain, of course, if it indirectly exploited the links with any macroeconomic parameter.
I'm not arguing. I'm supplementing.
It's probably not about time. For me, the most important parameter by which one can judge a TS more or less confidently is the number of trades in the testing area (or OOS). It is rarely paid attention to, but it is very important because statistics rule in trading.
The lower the number of trades, the higher the profit factor should be to conclude that the system will work in the future. If a longer plot with a number of trades of the order of a thousand will show a profitability of 1.65, it seems to be already a serious bid. I am doing research in this area and am going to post an article on the subject soon. About 70% of it is written.
When working with my system I don't use OOS at all. I base my estimation of the optimization results exactly on the number of deals. Hundreds and thousands of non-intersecting in time trades, the profit exceeding the maximal drawdown by an order or two. MO or profit factor is not a very important parameter in case of a solid statistics. Therefore values 1.5 and less are quite satisfactory. The main thing is a reasonable statistical advantage. Optimization can also be performed by increasing/decreasing the spread.
But again, we must understand why the system operates. And what is necessary to stop it from working. I.e. it makes sense to force the system to "bend" by varying trading conditions.
Seems to me: if you see the result on the balance change chart at a thousand trades for a few months as a straight up, then you need to dig for the reasons. Perhaps, you were lucky enough to come across such a great system of your period. Then analyze and reanalyze the reasons why it happened. And the analysis should not be reduced to the fact that there was a global trend or some flat, but the analysis should be presented in the certain evaluation of that period, expressed in numbers.
Therefore it is ideal when the system is activated when these indicators are consistent with its robustness, rather than when it has shown a good performance in the last weeks to months.
If there is a significant change in any macroeconomic parameter (e.g. interest rate), the system can "without a hint" go into the drain, of course if it has indirectly exploited the link to any macroeconomic parameter.
>> I'm not arguing, I'm adding.
I can not generalize to all systems. I will give you an example of my system. My system does not care about flux-trends, I only switch them off when there is news. But here is what I have observed:
In 2008 the system was changing its optimal parameters on the 4th day of each month for half a year. I did not manage to find out why it was the 4th day, was it a coincidence or not.
After a strong news, the system would drain two weeks (out of the year) (until the nearest 4th day). I could not ascertain the reasons. But I have found out what corresponds to them. Therefore, the system recognizes such periods. But the period was not repeated.
After a certain date, the system starts working on a dozen of trading symbols, improving its indicators from month to month. I do not know what this date corresponds to.
The system started losing constantly on one trading instrument only once, before that the average profit was zero for a week, which did not fit the statistics in any way. After stopping, the system started losing on the tester.
The system does not work on so many trading instruments and the reasons are known.
stupid indicators in an abnormal market give false signals to drain
the thing i'm proud of is the property of my indicators (working on bars) that after a gap they go in themselves... as seen last week, two days of no signal by the end of friday
i think it's good that i have enough intelligence to create such a system, but what about the people who use simple indicators that do not work on tick history?
This is probably not a question of time. For me, the most important parameter by which we can more or less confidently judge the quality of our TS judgments is the number of trades in the testing area (or OOS). It is rarely paid attention to, but it is very important, because statistics rule in trading.
The lower is the number of trades, the higher should be the profit factor to be more confident in concluding that the system will work in the future. If a longer plot with a number of trades of the order of a thousand will show a profitability of 1.65, it seems to be already a serious bid. I am doing research in this area and am going to post an article on the subject soon. About 70% of it is written.
P.S. Leonid, I am closely following your PAMM account. Everything is going great for you.
Alexey, it's not about the PAMM. The whole point is to understand that the TS is not eternal - everything ends at some point. And there is one problem - the more we do testing, the more we watch and analyze working of TS on demo-real, the faster it becomes obsolete for real trade, so when we put it into trades it will not last long. Therefore I would like to make correct decisions much faster than after 300 trades on REAL, because in practice after 300 trades on data, which TS has not seen, it usually ends..... I haven't seen an TS that would work for more than half a year on intraday bars(from an hour or less). Forex is a very volatile market )))))
P.S. And the article would be interesting to read....
It should be possible for each parameter to be optimised so that the optimisation graph changes smoothly from value to value. It should be clear that the system is working, while the optimization is not adjusting, but, as the name of this process implies, selecting the optimal parameters for an already working system. I.e. if the market changes, the system does not start losing on these parameters, but simply brings non-optimal for itself profit.
That is, all you need to do is to write a system which is profitable, and through optimization to select the optimum parameters for each trading instrument. And, of course, the fewer such parameters - the better.
This all is clear, but it's on the level of feelings. I would like to have some figures, some mathematics to formalize these feelings somehow.....)))))