Simulation is a process of testing the established mathematical or logical model of the system or decision-making problem to understand the system behavior or help solve the decision-making problem. The main advantage of simulation is that it can test the problems in the trading model or any assumptions of the model system, making it the most flexible tool. The simplest and most reliable way to judge whether the trading model has practical value is to conduct long-term testing in as many markets as possible. In order to reduce the detection cost of trading model, detection begins with simulation. The basic principle of trading model testing is "simulated actual combat", and all conditions should be close to actual combat conditions, so as to make the test results as real as possible, because only in this way can trading models have real use value.

1. Emergency

Unexpected events (including price limit) must be included in the inspection process, because besides the normal operation of the trading model, it is also necessary to have the ability to deal with unexpected events, and the influence of unexpected events cannot be ignored because it is a "small probability" event, and the basic principle of "simulated actual combat" should be followed. A mature trading model should be able to resist the risks brought by emergencies, even if it can't capture the excess profits brought by emergencies.

2. Inspection information and data

For the basic analytical trading model, a perfect information database is needed. With the development of science and technology and the continuous application of the Internet, information collection is much more convenient than before, so it is relatively easy to sort out and improve the information database. As for the trading model of technical analysis, because futures funds operate futures varieties, the data of futures varieties have their own uniqueness, and the data of European and American futures have different characteristics. For example, there is no "fault phenomenon" in the futures data of London metal, so it will be no problem to check it by computer. However, domestic futures data sources attack the futures data of the United States, and different trading contracts will have "data faults" when changing months, which cannot be simply treated as stocks. Therefore, we must first check the data through the transaction model.

Actual contract data: From the actual contract transaction data, the shortcomings are very obvious, because the domestic futures contracts only have a period of 1 year at present, so the data period is too short, the long-term trading volume of the contracts is inactive, and the liquidity is small, which is not representative.

That is, monthly continuous data: connected according to the contract delivery date to form continuous data. The continuous data generated in this way has the advantages of actual transactions, but it will be different in actual transactions. Pre-delivery transactions are inactive and lack of representativeness. Like Shanghai Copper, the fourth and fifth contracts are active after delivery. The disadvantage is that it will produce "fault phenomenon", which will cause great distortion to the test results.

Continuous data of price adjustment: according to certain rules, connect the subsequent contract data within a certain period before delivery. The time parameter x here should be determined according to different varieties. Shanghai copper is larger than Dalian soybean and Zhengzhou wheat. Accumulate the price difference between the two contracts during adjustment, and finally add and subtract the accumulated price difference into the data column to get the final futures data. In particular, it should be noted that the adjusted futures data may be negative, and the corresponding data should be adjusted, but this will not affect the trading results detected by the computer. The advantage is that it can reflect the price change level for a long time; The disadvantage is that the data cannot be directly used in actual transactions and needs to be converted.

Weighted continuous data: connect the subsequent contract data at a fixed time, and calculate the continuous price according to the size of the recent month, the size of the distant month, or the ratio of volume to position. With the passage of time, the weight of the recent month contract is getting smaller and smaller, and the weight of the distant month is getting bigger and bigger. The advantage is that the "fault phenomenon" of data is eliminated, and multiple active months can be selected, which is closer to the actual transaction; The disadvantage is that the data cannot be directly used in actual transactions and needs to be converted.

The above four data processing methods have their own advantages and should be selected according to the user's situation. For short-term users, the actual contract data is better, while for long-term users, continuous data can truly reflect the actual long-term profit and loss situation and conduct computer tests. In the test of trading model, in order to ensure the reliability and stability of the test results, sufficient statistical sample data is needed. According to the requirements of large sample statistics, the number of samples should be more than 30. In the short-term trading model, the data time should not be shorter than the time-sharing data of 1 year, and the daily data should not be less than 3 years. It takes more than one cycle to basically analyze the data of the trading model.