1, test time: A good programming must stand the test of time period. If a program, the result is beautiful, but the cycle is only one or two months, which is not credible;
2. Use of funds: Many people post beautiful test results, and the use of funds is often 80% or other percentages, but these are unreasonable choices, because fund management in financial markets is very important. When the market is good, the higher the use of funds, the greater the income, and when the market is bad, the higher the use of funds, the greater the loss, but we can't judge what the next market will be, so it is unreasonable to use the percentage opening method for the results of historical tests, which is also unreasonable.
3. Test method: The opening and closing price tests are unreasonable, and the trend model generally takes the trend reversal point as the opening signal, so the order price appears more accurately.
Analysis of test results:
A. Total number of instructions: that is, the number of signals. If it is too high, it means that the shock market is not well filtered; If it is too low, the risk is high; How to judge whether the number of signals is reasonable? Then there is only a comparison of different models in the same period. There is also the simplest method, which is to take the total number of orders/effective trading day as an example. Generally, the average number of signals in a valid trading day is between 2 and 5 (this data is for reference only);
B. Profit rate: You don't need to look at the total profit, just look at the result after deducting the maximum profit, which must be positive. The longer the test period, the greater the profit rate. In many models, short-term testing is good and long-term testing is bad, so try to measure the longest period that can be measured. (Of course, because market relations may also occur, the long-term profit rate is lower than the short-term profit rate, but generally speaking, the longer the term, the higher the profit rate, which is the test result of a good model. )
C. correct rate: other things being equal, the higher the correct rate, the better, but you don't have to be moved to see the model with high correct rate, and you don't have to worry about the low correct rate of your model. Generally, it is good that the correct rate can be around 45%, because the original intention of programming is to lose money, and the correct rate will naturally be low when it oscillates;
D. Maximum loss rate: If you choose a fixed number of lots, such as 10, your maximum loss rate cannot exceed 10%. Of course, if more lots are selected, the maximum loss rate may be increased. If you choose 80% capital utilization rate, the loss may be even greater, and of course there will be test results with little loss.
E. Short-selling time: Take short-selling days as an example. If the short time is not too high, it is bound to miss the big market. Of course, this item is not the most important. Short time is long, the profit is high, you will miss it. It is not a mistake to miss it, and there is no risk of losing money.