Back Testing with Futures

[rickek 10]

Hello,

 

I was working with Back Testing in Tradestation and a question came up. I can't seem to find the answer any where. So help would be greatly appreciated.

 

When you back test lets just say the 30 Tres bonds (ZBM9 or USM09 depending the software), that chart is for June. You want to back test your strategy back 6 months. Does that back test just this chart/contract or does it keep it the lead contract?

 

Further explanation - you look at the February results of your back test. Is that the information for the June contract only or is there some kind of rollover? Is there a way to do rollover for furtures with backtesting in tradestation?

 

If there is someone out there who understands my question and can explain it, it would be great!!!

 

Also anyone who uses tradestation know why 30 US Treasury bonds is ZB(month, one number year) in other trading platforms but in the TS it is US(month, two number year). Are these the same charts/contracts...they look to be when compared with other platforms.

[zdo 20]

rickek,

 

It may not be the best schema for a creating continuous contract but you can use @US, etc. for data that spans multiple contracts.

 

hth

[Richard Dale 10]

You need to create a continuous contract that is backadjusted to remove the gap associated with rolling from one contract to another.

 

We wrote an article on this quite a long time ago but its still applicable:

Futures Continuous Contracts Explained

[Ab763 10]

I have traded 30 yr bonds for about 3 years in a proprietary fund and found that when you backtest more than a few days back with Tradestation it curve fits. The 30 year bonds are heavily algo traded by many funds and banks, thus the movement and condition is often related to the Dow, S&P and the Nasdaq. This includes the mini contracts. The more successful 30 year bond strategies and trading systems available for Tradestation (easylanguage) use tick bars to eliminate the irrelevant factor of time and focus on market movements based on volume. 144 tick bars (plus or minus one fibonacci number) will produce clearly defined market movements and conditions for the 30 yr bonds. To optimize a system to tick bars you only need enough bars back to capture 3 or 4 examples of trending, oscillating and choppy market conditions. You can find this in just a few days of data.

 

Tradestation is not the best backtesting tool but it is great at forward testing a system. Using the simulator you can auto trade a system and watch it’s performance in real time. This allows you to see what kind of market conditions cause the system to break down.

 

One of the best tools for evaluating market conditions with tick bars is the Lasalle 3Line Trender.

 

For more information on using Tick Bars read "Building Winning Trading Systems With TradeStation" by George Pruitt and John Hill.

[sevensa 10]

I have traded 30 yr bonds for about 3 years in a proprietary fund and found that when you backtest more than a few days back with Tradestation it curve fits. The 30 year bonds are heavily algo traded by many funds and banks, thus the movement and condition is often related to the Dow, S&P and the Nasdaq. This includes the mini contracts. The more successful 30 year bond strategies and trading systems available for Tradestation (easylanguage) use tick bars to eliminate the irrelevant factor of time and focus on market movements based on volume. 144 tick bars (plus or minus one fibonacci number) will produce clearly defined market movements and conditions for the 30 yr bonds. To optimize a system to tick bars you only need enough bars back to capture 3 or 4 examples of trending, oscillating and choppy market conditions. You can find this in just a few days of data.

 

Tradestation is not the best backtesting tool but it is great at forward testing a system. Using the simulator you can auto trade a system and watch it’s performance in real time. This allows you to see what kind of market conditions cause the system to break down.

 

One of the best tools for evaluating market conditions with tick bars is the Lasalle 3Line Trender.

 

For more information on using Tick Bars read "Building Winning Trading Systems With TradeStation" by George Pruitt and John Hill.

 

TradeStation is just a tool. It cannot decide by itself to curve fit. This is how you use it for backtesting which will determine if this is curve fitted or not.

 

Considering that your first post include a link to an indicator for sale, your post appear to be borderline SPAM.

[Frank 10]

rickek,

 

your question is a good one. based on my experience, here is my 2 cents --- take the @US contract and run your strategy for the past year--- then go back and run the same strategy using the 4 individual contracts that correspond to the same time period of the continuous contract strategy results (find the rollover dates and sum the individual results into a total) --- align the dates so its comparable and see how, when and why there are differences.

 

I found that doing this was hugely informative. in reality, you are going to be trading the individual contracts, not the continuous contract, so you might as well figure out the nuances now anyway.

 

fwiw, when I did this for the S&P and Russell -- I found large differences in the tested results (on Tradestation). there are just some assumptions that have to be made in back-testing that you will learn if you go through this process on a 'bottom-up' basis. Personally, I would use the continuous contract to 'ballpark results' of a new idea I wanted to test -- but I would never trust the continuous contract when it comes time to really and truly analyzing the strategy results.

Edited April 21, 2009 by Frank

[Richard Dale 10]

Something to be aware of with futures data is that if you are basing your roll decisions on volume/open interest, is that many exchanges do not provide volume and/or OI on a timely basis. Sometimes it is provided in estimate form then clarified a day later. Other times it is only provide a day late.

 

So if you have a trigger to roll, ensure that your rolls are done in accordance with when the data is available to you.

 

That being said, unless your trading systems are extremely sensitive, you shouldn't see much overall difference in a roll that is timed +- 1 day. If you do, perhaps the system needs refinement.

[Ab763 10]

Sevensa, yes tradestation is a tool. The back testing engine is designed to curve fit to some extent. That’s just how the Tradestation algorithms work the more data you give it the more it curve fits. Building continuous contracts with time data is almost useless in Tradstation unless you are putting on positions that last several days (like trading stock or spreads, the discussion is about futures). If you are trading intraday then a more modern approach is warranted. That would be using tick bars and tools designed for them. The Lasalle link has the best screen shot example of 30- year bonds with 144 tick bars and the market condition patterns it makes (this is what rickek was talking about trading). When back testing using tick bars it is necessary to get enough data to cover a few instances of these patterns. This would usually be less than two weeks, sometimes two or three days. There are other market condition identifiers; this is the one I like. I use it on everything.

 

There are better commercially available back testing engines than Tradestation. But again forward testing is Tradstation on their simulator is excellent. The Tradestation back testing is ok if you keep it to a minimum as recommended by George Pruitt and John Hill.

 

I trade the ES, YM, NQ, and US almost every trading day. Because the US moves as slow as watching paint dry, I use an automated Polynomial RSI system on it and seldom hold overnight positions. The US seems to work well with overbought oversold type systems and indicators designed for tickbars.

 

If you want to build you own system start with a CCI or RSI indicator and add order execution signals. Backtest with optimization and find the median inputs for the best 10 or so optimization report input settings. Apply these settings to the indicator so you can see on the chart how the strategy trades in relationship to the market conditions and movement. With practice on the sim you can visually optimize the indicator for best performance in the current market conditions and apply these setting back in to the strategy.

 

There are several commercially available trading systems that use the strategy-indicator system. Most of these have high math algorithms to enhance the ability to capture oscillations.

 

Quality data for tick bars is a must. Tradestation has good data (low compression) thus works well with tick bars. In a side-by-side test with Tradestation, IQ feed had 25% less ticks than Tradestation. This is data compression and will it render back testing unreliable.

[sevensa 10]

Sevensa, yes tradestation is a tool. The back testing engine is designed to curve fit to some extent. That’s just how the Tradestation algorithms work the more data you give it the more it curve fits. Building continuous contracts with time data is almost useless in Tradstation unless you are putting on positions that last several days (like trading stock or spreads, the discussion is about futures). If you are trading intraday then a more modern approach is warranted. That would be using tick bars and tools designed for them. The Lasalle link has the best screen shot example of 30- year bonds with 144 tick bars and the market condition patterns it makes (this is what rickek was talking about trading). When back testing using tick bars it is necessary to get enough data to cover a few instances of these patterns. This would usually be less than two weeks, sometimes two or three days. There are other market condition identifiers; this is the one I like. I use it on everything.

 

There are better commercially available back testing engines than Tradestation. But again forward testing is Tradstation on their simulator is excellent. The Tradestation back testing is ok if you keep it to a minimum as recommended by George Pruitt and John Hill.

 

I trade the ES, YM, NQ, and US almost every trading day. Because the US moves as slow as watching paint dry, I use an automated Polynomial RSI system on it and seldom hold overnight positions. The US seems to work well with overbought oversold type systems and indicators designed for tickbars.

 

If you want to build you own system start with a CCI or RSI indicator and add order execution signals. Backtest with optimization and find the median inputs for the best 10 or so optimization report input settings. Apply these settings to the indicator so you can see on the chart how the strategy trades in relationship to the market conditions and movement. With practice on the sim you can visually optimize the indicator for best performance in the current market conditions and apply these setting back in to the strategy.

 

There are several commercially available trading systems that use the strategy-indicator system. Most of these have high math algorithms to enhance the ability to capture oscillations.

 

Quality data for tick bars is a must. Tradestation has good data (low compression) thus works well with tick bars. In a side-by-side test with Tradestation, IQ feed had 25% less ticks than Tradestation. This is data compression and will it render back testing unreliable.

 

Again TradeStation is the tool. Backtesting is done based on YOUR algorithms and trading strategy. Not some magical internal backtesting algorithm from TradeStation which curve fit on time based charts, but not tick based charts.

 

Just because you are unable to find a profitable system based on time charts with backtesting that also works in real time, doesn't mean it doesn't work and can't be done. All this means, is that you were not able to find something that doesn't work.

 

Thanks for all your other advise, but I am quite happy with my backtesting strategies. To be honest, I would take any advise from someone who think backtesting over three days is sufficient and that a screenprint of three moving averages is the holy grail, with a grain of salt.

[rickek 10]

Agreed -- 3 days of backtesting..isn't backtesting - it's..