I have spent months developing a program for back-testing trading methods. The program uses tick data directly from the CME. My computer spent 7 hours working last night only for me to wake to awful results. The strategy is a very simple momentum play that only takes short positions. It tests both NQ and ES from 9:25 to 4:05. It has 2 variables. VARIABLE A: When to enter. Entrance is determined by first taking the highest tick either since 9:25 or the last position exited. Every tick is evaluated to see if it is the high tick. The entry is determined by taking the high tick subtract VARIABLE A. When the target is triggered the ordered is sent. VARIABLE B: When to exit. Exit is determined by first taking the lowest tick since the position was taken. Every tick is evaluated to see if it is the low tick. The exit is determined by taking the low tick plus VARIABLE B. When the target is triggered the ordered is sent. Each variable is tested in a range from $.25 to $10.00. Sixteen hundred different variations per day are put into an Excel matrix. The results are terrible. I have tested this not only on the short side, but the long side also. I am using market orders with a time delay of 3 seconds and a cost of $2.00 per side. I feel my problem may be the use of tick data. When I chart the data I notice a large amounts of volatility in short increments. This volatility is normally in a small amount of contracts. Every thread regarding back testing seems to reference minute or volume bars. Does anyone have any suggestions regarding the using the use of bars and/or different strategys?
I also use CME tick data when backtesting. I'm not quite sure what you mean by the results being "terrible". Have you done other forms of testing which would indicate a bug in programming? If so, you will have to change your definitions of entry and exit.
The terrible results have been 99/100 losing money. I have only tested 10 days worth of data so far and not one set has shown a profit of more than 3 basis points. I have spent hours searching for a bug. Do you break your data down into bars rather than ticks? Maybe you use a certain time frame in which the ticks must occur.
I use tick data and do not break the data down into bars, nor do I impose any other constraints. It looks like it's time for some more serious debugging.