I backtested the turtles system 2 from 1995 - Nov 2010 and I got the below result, "Turtles backtesting 14 Markets.pdf". The simulation only trades 14 markets, mostly chosen from the original list of the turtles (EURO, JPY, CORN, CAD, DAX, CRUDE, COPPER, NATURAL GAS, SUGAR11, GOLD, LIVE CATTLE, SMI, SOYBEANS, COFFEE) There is a small international flair like the DAX and the SMI.
The result is not bad (I backtested on excel not on TB). The blue line is closed equity. It turns $100,000 into $3.8m within 15 years. The pink line is the VIX index. I added a small twist, basically setting my stops at 2.25ATR if the 90 day MA of the VIX is between 20-30 and 3ATR if the MA goes above 30 and 2ATR when under 20. This filter recognises that 2ATR on its down does not capture adequately sufficient volatility for defining stops during turbulent times so I use the VIX to widen the stop a bit further during volatile times. The filter worked well as without it closed equity comes to $2.2m. Still not bad though.
When selecting the markets I tried not to overfit. My criteria was benchmark futures, highly liquid with good sector representation (indices, FX, energy, softs, meats, ags, metals) and minimal correlation among each market. Backtesting worked well with other market combinations too (except GBP and the FTSE.I dont know what is wrong with the UK stuff!)
I believe turtles is not dead as long as our tradeable universe is not a very large number of markets. Thats probably why Rich and Bill gave their students only 21 markets to choose from. If the limit is 12 units each direction and we trade say 100 markets then the chances of the 5, 6 open markets actually ending in a good trend are minimal. On the other hand a universe of 14-15 liquid primary markets that are benchmarks in their respective sector stand a bigger chance to experience LT trends.
Comments are most welcome...
Please critique my Turtle backtest
- Attachments
-
- Turtles backtesting 14 Markets.pdf
- (54.5 KiB) Downloaded 2020 times
Dear sluggo,
I am not saying this is the perfect system. I am sure there are much better ones out there with more favorable risk/return tradeoffs. However we need to also think of the practical implications of anyone of those systems.
If for example a system that gives equal or better CAGR return prospects with a much better MAR requires trading 100 or more markets possibly simultaneously then we are not really compairing like with like. The resources, time devotion, automation and system needs to trade the latter well, with minimal slippage would be significantly higher than the turtles system. That system was designed at a time when computers were a privilage to have one. It's easy to implement with minimal effort.
The bottomline is this. A system with fantastic stats would not be of any use if it requires unrealistic dedication or resource needs from the average trader to take live. Still, i am sure there are systems out there requiring similar effort to trade as the turtles with better stats (not too much better though),,,
I am not saying this is the perfect system. I am sure there are much better ones out there with more favorable risk/return tradeoffs. However we need to also think of the practical implications of anyone of those systems.
If for example a system that gives equal or better CAGR return prospects with a much better MAR requires trading 100 or more markets possibly simultaneously then we are not really compairing like with like. The resources, time devotion, automation and system needs to trade the latter well, with minimal slippage would be significantly higher than the turtles system. That system was designed at a time when computers were a privilage to have one. It's easy to implement with minimal effort.
The bottomline is this. A system with fantastic stats would not be of any use if it requires unrealistic dedication or resource needs from the average trader to take live. Still, i am sure there are systems out there requiring similar effort to trade as the turtles with better stats (not too much better though),,,
In 2005 there was a discussion here about ways to trade a small account. I mentioned the possibility of trading a portfolio of "low dollar volatility" futures markets. These are markets deliberately chosen because of their historically low dollar-risk-per-contract. Hopefully the small account trader's risk budget can accommodate them. I offered a 13-market portfolio of low volatility (in 2005) markets:
In addition to higher gain-to-pain ratios, this system has another advantage: it doesn't move its stops every day. So you don't have to place a new exit-on-a-Stop order, every single day, for every single open position. After you enter (using a market order), you place the stop exactly once, Good Till Cancelled. When you exit the trade, you cancel the stop. A much-reduced workload.
With a bit of programming, you could modify the system and add a parameter "Max Number Of Simultaneous Positions". You could set that parameter to "13" which would ensure you'd never have more than 13 simultaneous positions. Then (here's the good part) you could increase the size of your portfolio to as many markets as you like. You could give yourself many more chances to find an acceptably low-risk entry signal, by following many more candidate markets. Yet you will never have more than 13 positions on at any given time. Your max-possible-risk does not increase when you add more markets to your portfolio. Very jolly.
- Canadian Dollar (IMM CME)
Class III Fluid Milk (CME)
CAC-40 stock index (MATIF Paris)
Five Year Treasury Notes (CBOT)
Gold (COMEX NYC)
High Grade Copper (Nymex)
Live Cattle (CME)
Gas Oil (IPE London)
London Robusta Coffee No.406 5 tonne contract (LIFFE)
Mexican Peso (IMM CME)
Oats (CBOT)
Canola - Rapeseed (Winnipeg)
Sugar #11 (NYBOT)
In addition to higher gain-to-pain ratios, this system has another advantage: it doesn't move its stops every day. So you don't have to place a new exit-on-a-Stop order, every single day, for every single open position. After you enter (using a market order), you place the stop exactly once, Good Till Cancelled. When you exit the trade, you cancel the stop. A much-reduced workload.
With a bit of programming, you could modify the system and add a parameter "Max Number Of Simultaneous Positions". You could set that parameter to "13" which would ensure you'd never have more than 13 simultaneous positions. Then (here's the good part) you could increase the size of your portfolio to as many markets as you like. You could give yourself many more chances to find an acceptably low-risk entry signal, by following many more candidate markets. Yet you will never have more than 13 positions on at any given time. Your max-possible-risk does not increase when you add more markets to your portfolio. Very jolly.
- Attachments
-
- Fig 2. Backtest results of TMA + LV_13
- ec_tma.png (20.8 KiB) Viewed 17509 times
-
- Fig 1. Simulation setup for TMA + LV_13
- par_tma.png (20.87 KiB) Viewed 17509 times
Very interesting, thanks Sluggo. With a 6ATR stop and 10MA profit taking, I would expect the profitable to loss making trades must be quite high (much higher than turtles) otherwise risking 5.5% per trade would wipe out the account in no time.
I am not sure I understand your entries. Is it when the 10MA cross the 75MA which in turn crosses the 225MA or is it when price crosses all three or something else.
Forgot to mention, the backtest i did incorporates 4.5bps spread cost and slippage (I know its quite high but i'd rather be surprised on the upside than the downside!). How much cost does yours' above simulate.
I am only asking because spread cost can make a huge difference to the results. For example, if I run my backtest without a spread cost I get a terminal closed equity value near $6m.
ps - You might need to model higher than 4.5bps as some markets in the universe of futures you used, although come with smaller notional, are less liquid than the ones I used above.
I am not sure I understand your entries. Is it when the 10MA cross the 75MA which in turn crosses the 225MA or is it when price crosses all three or something else.
Forgot to mention, the backtest i did incorporates 4.5bps spread cost and slippage (I know its quite high but i'd rather be surprised on the upside than the downside!). How much cost does yours' above simulate.
I am only asking because spread cost can make a huge difference to the results. For example, if I run my backtest without a spread cost I get a terminal closed equity value near $6m.
ps - You might need to model higher than 4.5bps as some markets in the universe of futures you used, although come with smaller notional, are less liquid than the ones I used above.
Scaling out?
I am new to this, and I don't know how to test it yet.
Sluggo, I read your earlier posts, the ones about scaling out as well.
Just out of curiosity: would scaling out improve th MAR of this system?
Sluggo, I read your earlier posts, the ones about scaling out as well.
Just out of curiosity: would scaling out improve th MAR of this system?
I tried my above turtles system without pyramiding. Instead i fill all 4 units at the first breakout signal (instead of every 0.5N). Still, 12 maximum units each side. The results improve massively. Closed equity value in my backtest tripples. CAGR 27%, MAR also improves to 52%.
The main reason for this is the non-movement of the stops. This leaves more room for the trade to breathe so more trades catch a trend. Also all 4 units have a better entry price. The downside is risk per trade of 8% instead of 5% but that gets compensated by the above.
The main reason for this is the non-movement of the stops. This leaves more room for the trade to breathe so more trades catch a trend. Also all 4 units have a better entry price. The downside is risk per trade of 8% instead of 5% but that gets compensated by the above.
-
presentmomentum
- Contributor
- Posts: 2
- Joined: Mon Apr 04, 2011 10:25 am
Backtesting on Excel
You said, "I backtested on excel. . . ."
I work mostly with Excel myself. If you're open to sharing your Excel backtesting work, I would love to have a copy of the spreadsheet itself.
Thanks!
I work mostly with Excel myself. If you're open to sharing your Excel backtesting work, I would love to have a copy of the spreadsheet itself.
Thanks!
It is hard! The xls file is now approx 180mb. I make a change and it takes 30 seconds to fully update.
The beauty of working backtests with spreadsheets is that it doesn't get more transparent than this. You know exactly your backtest drawbacks and merits because you have a verifiable and visible walk through. This is unlike working with code where there can be invisible elements which can have a massive impact during real trading.
This is not to say spreadshhets are the way to go. Spreadsheets dont handle intraday trading properly unless for each day you use intraday data but we can imagine how much larger the file would be!
There is another reason i use spreadsheets. I don't know TB code!
The beauty of working backtests with spreadsheets is that it doesn't get more transparent than this. You know exactly your backtest drawbacks and merits because you have a verifiable and visible walk through. This is unlike working with code where there can be invisible elements which can have a massive impact during real trading.
This is not to say spreadshhets are the way to go. Spreadsheets dont handle intraday trading properly unless for each day you use intraday data but we can imagine how much larger the file would be!
There is another reason i use spreadsheets. I don't know TB code!
-
presentmomentum
- Contributor
- Posts: 2
- Joined: Mon Apr 04, 2011 10:25 am
AceofAce Excel Spreadsheet
Are you open to sharing the spreadsheet? Love to give it a go myself.
I recently did a back test on AFJ Gardner's modified turtles parametes in viewtopic.php?t=7301&highlight=turtles
I slightly changed the parameters as follows:
Portfolio – The 14 futures noted in the beginning of this thread plus S&P500
LMA: 300
SMA: 50
Go long on entry breakout and SMA>LMA, short on entry breakout and SMA<LMA
Trade Direction: Trade All
Risk Per Unit:3%, Max no of units: 4 added every 0.5ATR
Entry Breakout Days: 90
Exit Breakout Days: 45
ATR Days: 20
ATR Stop: 6
No ATR Profit Target
Max Sector Risk: 6 units
Max Total Risk: 20 Units both sides
I got the attached results for the period 23/5/1995 - 19/11/2010 which are very encouraging: On closed equity CAGR% 40%, Max DD 34.6%, MAR ratio 1.15. No of trades 755.
I wonder how the backtest would look like on TB.
I slightly changed the parameters as follows:
Portfolio – The 14 futures noted in the beginning of this thread plus S&P500
LMA: 300
SMA: 50
Go long on entry breakout and SMA>LMA, short on entry breakout and SMA<LMA
Trade Direction: Trade All
Risk Per Unit:3%, Max no of units: 4 added every 0.5ATR
Entry Breakout Days: 90
Exit Breakout Days: 45
ATR Days: 20
ATR Stop: 6
No ATR Profit Target
Max Sector Risk: 6 units
Max Total Risk: 20 Units both sides
I got the attached results for the period 23/5/1995 - 19/11/2010 which are very encouraging: On closed equity CAGR% 40%, Max DD 34.6%, MAR ratio 1.15. No of trades 755.
I wonder how the backtest would look like on TB.
- Attachments
-
- TurtleBacktestingTurtles9045(15INSTR).pdf
- (56.96 KiB) Downloaded 1112 times
AFJG's backtest results, using Trading Blox to simulate the system he published in ActiveTrader magazine, are here on this site: (link)AceofAce wrote:I wonder how the backtest would look like on TB.
However, his system and the one you describe are different in several respects.
- AFJG's portfolio is quite a bit larger and more diversified than yours.
- AFJG does not pyramid ("scale in") but you do.
- AFJG triggers partial exits ("scale out") when profit targets are hit but you do not.
- AFJG limits total heat and sector heat according to current risk (distance to stop) of each position, but you monitor unit-count instead of current risk.
The last line of his post is
AFJ Garner wrote:Buy the fully programmable version of TB – anything else and you are severely restricting your future growth. Enjoy!
Another possibility would be to hire one of the Trading Blox System Development Partners (link) to write Blox code for your system, and then run the backtests that you specify.