Like I said, I've been in the securities industry for a while now. I am quite aware of the difference between a limit order, stops and the like. Thank you. My question is posed to those who have backtested a system, forward tested the system on paper, and then actually traded the system. What was their experience in terms of decreased returns? And, for those that have used multiple platforms, which ones currently available have shown the least amount of difference between paper trading and playing with live ammo?
Backtesting - Execution Concerns
- Thread starter AllenTrader
- Start date
Sorry if the definitions were over the top, I just wanted to make sure that I was satisfactorily explicit _for_other_people _reading 
I have had systems that produce 100% correlation between tests and fills as well as systems that have 50% or less.
The higher the frequency (shorter the timeframe) the worse in most cases. Also, there have been systems that have fluctuated between 100% and 40% over longer test periods. It works perfectly for three weeks then ceases to line up at all... for no obvious reason either.
I have not found a way other than changing the system to accommodate better fills to keep it close.
All my recent tests were on Tradestation and proprietary software using Globex traded instruments.
I have had systems that produce 100% correlation between tests and fills as well as systems that have 50% or less.
The higher the frequency (shorter the timeframe) the worse in most cases. Also, there have been systems that have fluctuated between 100% and 40% over longer test periods. It works perfectly for three weeks then ceases to line up at all... for no obvious reason either.
I have not found a way other than changing the system to accommodate better fills to keep it close.
All my recent tests were on Tradestation and proprietary software using Globex traded instruments.
Excellent reply. How long have you been using TradeStation? Do you allow TradeStation to automate your trades? In terms of frequency, what time frame are we talking about for "high" frequency...intraminute?, intrahour?, intraday?, etc.
I tried it before. Believe it or not, I could make more than 200 ticks every day for two weeks!!!! I used limit order for entry which gave many trades that won't be filled in the real world. I was testing a very short term method just like a scalper.
I have used one or another version of Tradestation for 3-4 years, currently TS 8 . I do not trade with them though.
Regarding automation:
I have worked with some programmers and a prop. firm that trades only automated systems. Beware of turning on an automated money drain that looks good in testing...
The losses can pile up very quickly.
There are lots of potential technical problems that can be costly even with contingency plans in place. Either TS is not adequate for my needs or I just have not yet been able to fully implement adequate security measures and reliability with it. I have spent a good deal of time looking into it.
Timeframe:
The lowest timeframe is tick by tick , between 5-10 min is the highest. Despite the frequency of entry signals determined by the interval, market action dictates duration of most of the trades.
With my scalping systems, wining trades usually last under two min with the exception of systems that exit or revise orders after x amount of time/ticks/contracts elapses. Losers generally take a while to develop except in the most volatile and rangebound markets.
The longer term 5-10 min retracement systems stay in position between 1-8 bars irrespective of the end result with the average probably nearer to 5 bars than 8.
I guess the whole point is that each method/ system will have different characteristics and fill likelihood as a result of it's orientation relative to the current momentum and it's overall design.
Systems that buy on pullbacks or fade the near term direction have better fills when using limit orders in my experience.
Scalping systems that take liquidity at least on one side of the trade do better than those that try to get the edge in both directions.
I hope you can save a few $ from these observations, they seem obvious but having been seduced ( many times) by attractive simulations built around limit orders, I thought they might be of use...
Regarding automation:
I have worked with some programmers and a prop. firm that trades only automated systems. Beware of turning on an automated money drain that looks good in testing...
The losses can pile up very quickly.
There are lots of potential technical problems that can be costly even with contingency plans in place. Either TS is not adequate for my needs or I just have not yet been able to fully implement adequate security measures and reliability with it. I have spent a good deal of time looking into it.
Timeframe:
The lowest timeframe is tick by tick , between 5-10 min is the highest. Despite the frequency of entry signals determined by the interval, market action dictates duration of most of the trades.
With my scalping systems, wining trades usually last under two min with the exception of systems that exit or revise orders after x amount of time/ticks/contracts elapses. Losers generally take a while to develop except in the most volatile and rangebound markets.
The longer term 5-10 min retracement systems stay in position between 1-8 bars irrespective of the end result with the average probably nearer to 5 bars than 8.
I guess the whole point is that each method/ system will have different characteristics and fill likelihood as a result of it's orientation relative to the current momentum and it's overall design.
Systems that buy on pullbacks or fade the near term direction have better fills when using limit orders in my experience.
Scalping systems that take liquidity at least on one side of the trade do better than those that try to get the edge in both directions.
I hope you can save a few $ from these observations, they seem obvious but having been seduced ( many times) by attractive simulations built around limit orders, I thought they might be of use...