How to Avoid Excessive EA Optimization

It is often mentioned that the MT4 historical test is useless. In the past, there were indeed cases where the historical test did not match the real test. With the multiple upgrades of mt, the comparison between the real test and the historical test can now be said to be more accurate. The optimization of the historical test encounters more problems. It is the problem of excessive optimization. This has nothing to do with the mt historical test. No matter how accurate the test is, it will also encounter this problem. Then again, it is impossible for you to spend a lot of time on simulation testing every time you do an ea. Besides, short tests may not be able to reflect the whole picture. Historical testing is an essential tool. It depends on how you use it. How to avoid excessive optimization of the EA?

Let me tell you two stories.

◎ Someone practiced archery, but he was always inaccurate. When he saw the hearts of other people's arrows, he couldn't help but be jealous. Later, he thought of a good way to shoot the arrows first, and then run to the target where the arrows were shot and draw the "arrow target". In this way, it seems that it is also the hearts of arrows and arrows;

◎ It was also said that the British football lottery atmosphere prevailed, and a certain gentleman also wanted to make a fortune through the football lottery, but he couldn't guess, so he thought of an idea: before the weekend started, he selected 3,000 people through the yellow pages as the target, and wrote letters separately, claiming that he could accurately predict the results of the weekend A team against the B team. For 1,000 of them, he predicted that the A team would win, and for the other 1,000 people, he predicted that the B team would win, and for the remaining 1,000 people, he predicted that the two teams would fight together. The next weekend, he divided the 1000 people who predicted the accuracy of the previous week into three wholesale letters, making three predictions respectively. When there were 33 people left, he wrote to them separately, stating that if he wanted to get the prediction of this weekend's match result, he had to pay him 1,000 pounds first. As a result, most of these people had already fallen in love with this "discerning" expert, and a certain lord finally made a lot of money.

The so-called "over-optimization" is to use the historical data matching system to write an EA for the relationship between a historical market and indicators and data. In order to make the EA appear to have a good performance, the parameters are continuously adjusted and optimized, and the filtering conditions are set to make the EA and the historical data are completely consistent. The results are certain that this EA performs well in the historical data test, will buy in time before the big rise, and sell in time before the big fall, but will the EA send a signal in time before the next big rise and big fall? I'm afraid not, because this EA is written for the past situation, and it may not be applicable to the future. From my own experience, the more conditions the EA sets, the more complex the structure, and the more serious the "over-optimization" situation. Then again, writing EA cannot leave the historical data, otherwise we will become blind and unable to start, how to avoid excessive "over-optimization", I do this, first prepare long enough historical data, divide it into two parts, the former for observation, the latter try not to look at it, EA comes out first use the previous data for testing, adjustment, until the EA meets the target requirements and then use the latter historical data to test it, if the performance of EA is similar to the previous data, then I think this EA is basically qualified.