All traders trade mat expectation (game with an advantage), and the trader's task is to build a trading system where mat expectation will always be on his side. Not everyone will do it in a plus, but if the checkmate is built honestly, extrapolates into the future and the trader does not deceive himself (does not hope for a chance), then at a distance he will always take money from the market. Always. In algorithmic trading, when creating automatic trading systems, the issue of the lifetime of trading algorithms is very important, i.e. the ability to extrapolate the found expectation mat into the future. In a constantly changing market, sooner or later there comes a moment when even the most perfect and profitable algorithm starts to make losses, so we ourselves will determine the lifetime of the algorithm in one week and will create a system that will give a positive result with weekly optimization.
We will optimize by a complete enumeration of all parameters, since it is not known how many optimization extremes there will be, and we will test not a single instrument or time frame, but all working time frames and all trading instruments. The task is not easy, but quite doable. Forward test It would seem that the parameters have been optimized and you can start trading. However, this is not the end of the research process. The theory and personal experience show that it is necessary to additionally check the results obtained for forward testing, on an additional historical sample, and the ranges of historical samples are formed in such a way that the optimization period and the period for the forward test follow one another and the optimization period is greater than the forward period. test. As a rule, after checking the forward test, most of the trading strategies no longer look as attractive as after optimization.
Therefore, there is no need to test forward strategies that did not show a positive result even during optimization. Based on this, it is appropriate and reasonable to use the top best outcomes obtained during optimization for the forward test. If there is a system that, according to some algorithm, can mathematically find the best trading parameters and trade on the next time period, we will get a reliable result of work that will not differ from real-time trading by a robot. So we can get the result (profitability graph) of how the robot would work in real life if we chose a certain optimization method. As a result, we optimize only the optimization method (derivative), the entry and exit models themselves are not important to us, they are simple and do not contain any optimization parameters. Reservations At the same time, in order to achieve the identity of the strategy development in the tester with trading in real life, there are some reservations that must be taken into account: this is the fact that the entry and exit points cannot be located inside the bar, since it is often impossible to obtain reliable tick data for a long period of time, and optimization with taking into account tick data is practically impossible due to the necessary computing power, since we are not talking about one test, but about a numerous combination of albeit simple models of entry and exit points. .