How much you produce vs backtesting profits: From 0% to 100%?
http://www.elitetrader.com/vb/showthread.php?s=&postid=1238824#post1238824
http://www.elitetrader.com/vb/showthread.php?s=&postid=1238824#post1238824
How much you produce vs backtesting profits?
- Thread starter OddTrader
- Start date
"Leap of Faith? - How much you've attained (so far) vs backtesting profits?"
http://www.elitetrader.com/vb/showthread.php?s=&threadid=79206
http://www.elitetrader.com/vb/showthread.php?s=&threadid=79206
Because only part of my trading system(method )can be be backtested by computer, I was backtesting manually and averaged 3.8 points per contract/day which would translate to about 650% per 5k/contract account. Splendid, but in reality my system averages 1.5 points per contract/day which is about 200% net. Daytrading eminis.
Could it be that there is no logical necessity nor rationale to expect the future to replicate the past? But isn't every trader projecting some form of the past, -- whether via backtesting, pattern recognition, "gut" feel accumulated over 20 years, etc -- into the future, when he opens a trade? Therein lies the rub.
Backtesting either visually or by computer assumes all trades are fully filled at every signal.
In reality, that does not happen using limit orders. Especially in the ER and YM, limit orders will partial fill or miss filling altogther quite often. This week alone, I had several ER trades prestaged to enter on limit orders fill one - four contracts only when I was asking for five - ten.
The results of these partial fills would be limited profits, while losing trades ALWAYS fill completely as they go against us. That is one of many differences between back-testing and reality.
*
Another difference is money management. Testing assumes all trade signals are taken, and/or all trades left to run toward full profit objectives.
In reality, traders usually take some but not all trades. I myself commonly stop trading well before the day is thru if I'm profitable enough already. Also, I sometimes skip the morning session if charts are muddy and just trade in afternoon.
Back-testing would assume I traded thru all of the session and took every signal. Reality is, on trend /swing days when I'm up +8pts ~ +12pts ES / ER by 2:00pm est, many times I walk away. There may be more trade signals in the final two hours, but I'm done.
**
Those are just two of many reasons why testing vs trading live will differ. When testing any method, I expect to be 50% efficient at best. If the method in question remains robust enough at only 50% efficient operation, that's good. If its edge isn't big enough, I'd pass and look to refine or rebuild parameters from there.
In reality, that does not happen using limit orders. Especially in the ER and YM, limit orders will partial fill or miss filling altogther quite often. This week alone, I had several ER trades prestaged to enter on limit orders fill one - four contracts only when I was asking for five - ten.
The results of these partial fills would be limited profits, while losing trades ALWAYS fill completely as they go against us. That is one of many differences between back-testing and reality.
*
Another difference is money management. Testing assumes all trade signals are taken, and/or all trades left to run toward full profit objectives.
In reality, traders usually take some but not all trades. I myself commonly stop trading well before the day is thru if I'm profitable enough already. Also, I sometimes skip the morning session if charts are muddy and just trade in afternoon.
Back-testing would assume I traded thru all of the session and took every signal. Reality is, on trend /swing days when I'm up +8pts ~ +12pts ES / ER by 2:00pm est, many times I walk away. There may be more trade signals in the final two hours, but I'm done.
**
Those are just two of many reasons why testing vs trading live will differ. When testing any method, I expect to be 50% efficient at best. If the method in question remains robust enough at only 50% efficient operation, that's good. If its edge isn't big enough, I'd pass and look to refine or rebuild parameters from there.
