This post is about a revised rotational trading system. I’m a big fan of rotational trading. Please read the first post to get more about the why. While I DON’T think rotational trading is better than the system types that buy into weakness and sell into strength, I do think it’s an excellent way to diversify your trading systems. A while ago I had two posts (first / second) on rotational trading, since then I continued my research and further worked on my database. So the system I’m going to present is build on a survivorship bias free database, hence it carries a lot more validity than the previous version.
Market filter
- The system is only going to be traded when the NASDAQ100 idx is above the 20 or 200 days moving average. To say it in other words, the NDX100 has to be bellow 20 or 200 days moving average in order to prevent execution.
- Yes I know, some of the key elements of rotational trading is to avoid market timing. However, from an emotional point of view I find it incredible difficult to execute long trades under these circumstances.
Scoring
- As previously, I’ve used TSI to measure trend strength. We only want to be invested in strongly trending stocks. Remember, TSI only measures strength of the trend and not direction.
- For trend direction i suggest to use a simple ROC(200). The value of 200 isn’t optimized. Doing some optimization you might get better returns, but it’s good enough for me.
- You might want to use only the absolute value of the ROC(200). When the market is rebounding after an extremely sharp sell off, it’s better to buy oversold stocks.
Ranking
- Once you get the two scores, you have to combine them in order to get one unique rank.
Position
- NASDAQ 100 stocks
- Top five stocks, rebalanced weekly (Friday)
- 20% capital
Result
- Bellow you find the results from January 2001 to September 2010. I like to exclude the crazy NASDAQ times from 1999-2000 as I think they have been extremely unusual on either side.
- No commission or slippage calculated
- Database includes delisted stocks, index constitution is revised at day of change.
- The impact of survivorship bias (link): I re-run the same system based on a database with survivorship bias. Basically i used the NDX100 index as of today. The systems shows better performance, of course: CAR: 62% with a slightly lower draw-down
Conclusion
The simple system presented delivers decent performance and a good starting point for your own research. Past results are no guarantee for future returns. I do NOT trade the system as presented here. Do your own research. In case you can’t or won’t do your own research, then follow David Varadi from CSS Analytics. Every week David posts his Top10 picks for the coming week. This year his picks have outperformed the index by about 10% and there are still three month left.
AmiBroker code
The AmiBroker code is extremely tricky this time, because AmiBroker doesn’t support the ranking approach as presented.
#include <ndx100history.afl>;
#include <Boilerplate.afl>;
#include <Generall.afl>;
SetCustomBacktestProc(“formulas\\include\\cbt_midlevel.afl”,True);
SetOption(“CommissionAmount”,0.00);
SetOption(“InitialEquity”, 100000);
SetTradeDelays(0,0,0,0);
SetBacktestMode( backtestRotational );
EnableRotationalTrading() ;
SetOption(“MaxOpenLong”,5);
SetOption(“MaxOpenPositions”,5);
SetPositionSize( 20, spsPercentOfEquity);
IndexMember = NDX100member(Name(),DateNum());
idx = Foreign(“$NDX”,”Close”);
Listnum = 1; //here you enter the number of the NDX100 watchlist you want to use
list = CategoryGetSymbols( categoryWatchlist, Listnum);
Rank1_Score = 1;
Rank2_Score = 1;
Rank3_Score = 1;
Rank4_Score = 1;
Rank5_score = 1;
Rank1_bench = trendstrength();
Rank2_bench = abs(ROC(Close,200));
Rank3_bench = 1;
Rank4_bench = 1;
Rank5_bench = 1;
MArketOk = idx > MA(idx,20) OR idx > MA(idx,200);
exec = IIf(Name()==”AAPL”,1,0); //hard coded in order to improve performance, need to to this only once per bar for all stocks
for( i = 0;( ( sym = StrExtract( List, i ) ) != “”) ; i++ )
{
Maxitems = i+1;
}
if (EXEC[1]==1)
{
for( i = 0;( ( sym = StrExtract( List, i ) ) != “”) ; i++ )
{
Member = NDX100member(sym,DateNum() );
SetForeign(sym);
sym_v1 = IIf(member,trendstrength(),0);
sym_v2 = IIf(member,abs(ROC(Close,200)),0);
sym_v3 = 1;
sym_v4 = 1;
sym_v5 = 1;
RestorePriceArrays();
StaticVarSet(“IDXMember”+(i+1),member);
StaticVarSet(“RANK1″+(i+1),sym_v1);
StaticVarSet(“RANK2″+(i+1),sym_v2);
StaticVarSet(“RANK3″+(i+1),sym_v3);
StaticVarSet(“RANK4″+(i+1),sym_v4);
StaticVarSet(“RANK5″+(i+1),sym_v5);
}
}
for( i = 1; i < Maxitems; i++ )
{
Idx = StaticVarGet( “IDXMember”+i);
Rank1 = StaticVarGet( “RANK1″+i );
Rank2 = StaticVarGet( “RANK2″+i );
Rank3 = StaticVarGet( “RANK3″+i );
Rank4 = StaticVarGet( “RANK4″+i );
Rank5 = StaticVarGet( “RANK5″+i );
Rank1_Score = Rank1_Score + IIf( Rank1_bench > Rank1 AND idx AND rank2!=0,1,0);
Rank2_Score = Rank2_Score + IIf( Rank2_bench > Rank2 AND idx AND rank2!=0,1,0);
Rank3_Score = Rank3_Score + IIf( Rank3_bench > Rank3 AND idx AND rank2!=0,1,0);
Rank4_Score = Rank4_Score + IIf( Rank4_bench > Rank4 AND idx AND rank2!=0,1,0);
Rank5_Score = Rank5_Score + IIf( Rank5_bench > Rank5 AND idx AND rank2!=0,1,0);
}
ComboRank = IIf(indexmember,Rank1_score + Rank2_Score ,0);
Score = IIf(marketok AND ComboRank>0 AND IndexMember AND DateNum()<=1100901,ComboRank,0);
Condition1 = (DateNum()>=1010101 AND DayOfWeek()==5) OR DateNum()>=1100901;
PositionScore= IIf(Condition1,score,scoreNoRotate);
Rank1_Score = IIf(indexmember,Rank1_Score,0);
Rank2_Score = IIf(indexmember,Rank2_Score,0);
Rank3_Score = IIf(indexmember,Rank3_Score,0);
Rank4_Score = IIf(indexmember,Rank4_Score,0);
Filter = DateNum() == EndValue(DateNum());
AddColumn(Rank1_bench, “Rank1_Value”, 1.2);
AddColumn(Rank1_score, “Rank1_Score”, 1);
AddColumn(Rank2_bench, “Rank2_Value”, 1.2);
AddColumn(Rank2_score, “Rank2_Score”, 1);
AddColumn(Rank3_bench, “Rank3_Value”, 1.2);
AddColumn(Rank3_score, “Rank3_Score”, 1);
AddColumn(Rank4_bench, “Rank4_Value”, 1.2);
AddColumn(Rank4_score, “Rank4_Score”, 1);
AddColumn(Rank5_bench, “Rank5_Value”, 1.2);
AddColumn(Rank5_score, “Rank5_Score”, 1);
AddColumn(ComboRank, “Combo_Rank”, 1);
SetSortColumns(-5,-3);
This looks like a great system! I wonder how much more upside there is from chaning the ROC(200) parameter. I personally use a system where I take the average of ROC(20) ROC(50) and ROC(100).
Hello PSiegert,
Thanks for your feedback.
Frank
The market filter is not clear. Any of the two sentences should say “20 AND 200″, instead of “20 or 200″, INHO.
Thank you for your posts
Hello Orion.
no, i trade the system when either criteria is meet. Either IDX>MA20 or IDX>MA200.
That’s the line of code i use: MArketOk = idx > MA(idx,20) OR idx > MA(idx,200);
Hello Frank, First of I wish you a good year 2011:)
According the market filter, I am still not sure about 2 special situations.
1) The idx is under the MA20 but above MA200 will you open the trade or not?
2) The idx is above MA20 but under MA200, will you open the trade or not?
Early thanks for your smart answers to this silly questions:)
Hello Ilya,
1/2.) yes, it has to be either above MA20 or above MA200.
Regards,
Frank
I’m not clear on the Trend rating, does the trend score of 70 as an example, represent a 200 D ROC of 70%?
Hello Aristotle,
the table is just an example, but i use ROC expressed in percentage. So ROC200 is the percentage difference between today and 200 days ago.
Frank
hi Frank,
it is a bit of a pain to try this on different watchlists by constantly changing the hardcoded name of the
watchlist and the first symbol in the watchlist in the script. I have used in my programs the following mechanism (which you may already be aware of) to remove this hardcoding.. (I am replacing your code with the following lines)
//to get the current watch list number
EnableScript(“jscript”);
script = GetScriptObject();
Listnum = script.curWatchListNum();
//to get the first symbol in the watch list..
exec = IIf(Name() == StrExtract(List,0),1,0); //hard coded in order to improve performance, need to to this //only once per bar for all stocks
hope it helps. To give credit where it is due, most of this is derived from amibroker user group.
rgds
bgpl
Hello Bgpl,
cool! … i will try this. THANKS!!!!
BGPL – take a look at my comment concerning status(“stocknum”) as it may eliminate the need for jscript… let me know your thoughts.
Thx,
Carl
For some reason, the javascript code got removed from the earlier comment.. mostly due to the formatting..
EnableScript(“jscript”);
//Here remove the word space”
script = GetScriptObject();
oh boy.. sorry about that.. its getting real messed up. Hope this works..
put this function:
function getAAWatchListNum() {
var ab = new ActiveXObject(“Broker.Application”);
var aa = ab.Analysis;
return aa.Filter(0, “watchlist”);
}
within the separators “”
Do you have and example you could post of this using a watchlist without the NDxmember filter?
Using NDX100 as of today will return the following results (January 01 until September 10)
RAR: 90.08%
CAR: 67.97
Sharpe: 0.7
Winner: 57.60%
Avg%: 8.49
MaxDD: 20.53
Hello Frank,
Are the “include” files required to run this system? If so, can you please email them to me?
Thanks for sharing.
Burger
Hello Burger,
the only include file you realy need is the NDX100 one, because it eliminates the survivorship bias. Read this post: http://engineering-returns.com/2010/09/02/the-impact-of-survivorship-free-backtesting/
section: availability. Please understand, it’s been to much work to give it away for free.
Regards,
Frank
Frank,
When I sum up 2009 monthly returns from your chart in excel, I get 89.6 not 120
Thoughts?
Dave
Hi Dave,
that’s because you can not sum the month2month changes in order to get yearly% change.
Frank
oh duh, brain must not have been functioning last night . . .
Frank,
the only include file you realy need is the NDX100 one, because it eliminates the survivorship bias.
Yes. Can you throw a bone on how to get the start date and end date in afl? I would think each symbol will need to be hard coded, correct?
Regards,
John
Hi John,
exactly, i got a BIG AFL where beginning and end date is hardcode. Some of the stocks re-enter NDX100 multiple times.
http://engineering-returns.com/2010/09/02/the-impact-of-survivorship-free-backtesting/
Regards,
After much grief I finally got the code to work. It was the copy/paste process from html to AFL that was causing issues such as quotes not pasting correct. Anyway, works great and thanks Frank for a fine program. Here is a possible suggestion for ranking called Most Anchored Momentum. It is outlined here:
http://www.actwin.com/kalostrader/MAM-Rev3.PDF&pli=1
Burger, I took a look at the MAM paper last night. Do you think if you wanted to try that for say 6 month ROC, that you’d use something like 3DEMA/126SMA?
Aristotle, I would suggest testing it as he laid it out as an alternative to 6 month ROC.
Fantastic post Frank! I am also a big fan of rotational strategies and like you I think that rotation must be combined with timing. This really is A Grade research. Thanks for sharing!
One way that I have found to improve results further is to identify the best industry groups, trade only the top stocks in those groups and only trade in the direction of the broad market. Are you panning on having any of your strategies tracked and independently verified moving forward?
Keep up the great work!
Cheers
Derry
Also this paper points towards 6 months being a particularly effective period to measure ROC for a momentum strategy – http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1350193
So instead of 200 days 126 would be worth testing.
Cheers
Derry
Hi Derry,
Thanks for the feedback and link. Will read it, for sure.
I tested 100 days, but not 126 days. -> didn’t improve results
Regarding strategy tracking, not yet. This blog is outcome & documentation of my research. If i want to run trading as a end-to-end business, i might consider this.
All the best…
Hi Derry
WoodShedder over at IBC used the above paper to build and test a momentum system. His result’s were not very promising esp. after taking into account survivorship bias. Check out the link.
http://ibankcoin.com/woodshedderblog/2010/08/10/lets-build-a-momentum-system.
The reason I ask about having a public track record is that there is theory and there is practice and they are very different. You have some great ideas so laying out how they perform (and your ability to execute them) in real time would be the logical next step. I use timertrac.com but there are many different options that you may like to consider.
Cheers
Derry
Hi Derry,
you are so right. That’s why I never post a system I trade. Thanks for the link
Regards,
Frank
Frank, would it be much trouble to run the test with 10 stocks held instead of 5?
Also, how much does the market filter add to returns?
Hello Aristotle,
running this with 10 stocks and 10% you get
RAR: 49.65%
CAR: 37.54%
Trades: 827
DD: -46.13%
Running the system with 5 positions without a market filter would result about 10% a year. In this case i would suggest to hedge
Frank
I hate to ask a newbie question here but this system is a theory I have been thinking about for awhile and would love to play around with it. I am a tradestation coder but new to amibroker
how do I build these three initial files
#include ;
#include ;
#include ;
Hi Tony,
you only would need the one related to NDX100history.afl – which is to avoid survivorship bias.
Please read here for more information: http://engineering-returns.com/2010/09/02/the-impact-of-survivorship-free-backtesting/
Regards,
Frank
Hi Frank,
If I have 10 stocks to trade (none of them on NASDAQ100) and I want choose 3 highest score,
how should I modify your AFL code?
Thank you for sharing,
Hi,
the example didn’t need the ability to filter for a specific number (highest) stocks. However it has the foundation. you would need to add something like this.
for( i = 1; i Rank1 AND idx,1,0);
}
Buy = YOU_CRITERIA AND Rank1_Score >=(Total number of stocks) – 3;
I copied this from one of my other systems, you need to check the variables and other stuff.
Would like to ask whether you’re implementing the roc outright or using the absolute value?
Wouldn’t there be a problem if you used the absolute value as if a stock is trending downwards, you’ll get a high TSI and high absolute value of ROC which combining the two rankings, might wrongly signal a buy?
Hello Yish,
I’m using the %ROC value (not the absolute ROC value)
I agree with your concern, that’s why I would trade that only in a healthy market (close above MA20 or MA200).
Regards,
Frank
Thanks for clarifying frank. Also, if say you got long the 5 stocks when the index was trading above the dmas, what happens if the index goes below the dmas the following week? Would you exit the stocks and wait for the index to me above the dmas again to put on the trade?
Hello Yish,
that’s correct. Close the trades when the Index is bellow the DMA on Friday. Reenter next Friday if above.
Frank
Hi Frank,
Excellent ranking routine.
One question. For your EXEC, if you use EXEC = iif(status(“stocknum”)==1,1,0) instead of Name()==”AAPL” wouldn’t that give you execution control for any indice and not have to rely on the identifying the symbol of the first stock in the index? Or is there another side effect that you are trying to avoid by using AAPL?
Thaks,
Carl
Hi Carl,
BGPL made me aware of this improvement and I’m using it for my own system as well. Next time I post a revision of this system I’m going to include it. The same for listnum…
Frank
I just realized that you changed the timing filter from 30 & 200 to 20 & 200. I assume 20 tests out better than 30?
Thanks,
Hi Aristotle,
yes…. but it shouldn’t be a big deal as much as I remember.
Frank
Hi!
I suppose is a silly question, but I can not test the system because of the lines:
IndexMember = NDX100member(Name(),DateNum());
and
Member = NDX100member(sym,DateNum() )
I don’t recognize such a function ‘NDX100member’. Should it be in an auxiliary file?
I don’t ask for the non-supervivorship Nasdaq bias file, I understand Frank’s reasons, and I just want to test the system with survivorship bias, but.. I should be doing something wrong..
Thanks any answer..
Hello Gonzaga,
just write IndexMember = TRUE …
because it’s my function for checking if the stock is part of the index at any given day.
Frank
Thanks Frank, but I have still a trouble.
The other line,
Member = NDX100member(sym,DateNum() )
still crash in Ami. If I set
Member = true; // instead of Member=NDX100member(sym,DateNum() )
the code crash in the line
StaticVarSet( “IDXMember” + (i+1), member)
because ‘Member’ has to be a ticker, not a boolean value. ( the error is “Argument #2 has incorrect type (the function expects different argument type here) ” )
As I don’t really understand the code, (several lines are a mistery), I don’t know what to do..
Thanx in advance..
Hello Frank,
Just to let you know there is a simpler way in AFL to create a “Dynamic list of Stocks”.
1. By way of improting from a CSV file, automatically create another ticker with an extention for each stock with in out information in a price field. eg if the stock is MSFT then new ticker is MSFT-Survivor and is it is in the index it is coded as 1 for the close field and if it is out of the index it is coded as “0″ for the Close field. Of course you need to play with the excel get the CSV file like adding the extension to the tiker name and in/out info using excel functions but for THIS task it is much easier to play with excel than AFL.
2. Using the AFL Foreign function, determine the membership of the index when coding in AFL.
I have attached another code which I used for another project. as an example for the item 2.
_SECTION_BEGIN( “Use_a_specific_InOut_List”) ;
// Use a Specific InOut List; List In = 1, List Out = 2 for close of the list
Use_a_specific_InOut_List = ParamList(“Use_a_Specific_InOut_List”, “No,Yes”,defaultvar = 0);
if ( Use_a_specific_InOut_List == “Yes” )
{
List_Name =ParamStr(“Name of the In Out Stock List”, “-Star”);
// Stock List Selection
// Extract Foreign Data
sym = Name() + List_Name ;
SS = 0;
ListIn = 0;
ListOut = 0;
SS = Foreign( sym , “Close” ) ;
ListIn = IIf(SS == 1, 1, 0);
ListOut = IIf(SS == 2, 1, 0);
List = Flip(ListIn ,ListOut );
Plot( List , “List_Name”, colorBlue, styleOwnScale|styleArea|styleNoLabel,-0.1,50);
}
_SECTION_END();
Regards
Hello Asanka,
thanks for your code and contribution here. Why would you want to do that? Would you use that to replace my AFL function that checks for index membership?
Regards,
Frank
Hello Frank,
Yes so you don’t have to hard code your AFL function – with few excel functions CSV file can be created.
But in no way I am suggesting you should do that because you have already done it. !!
Regards
Asanka
No one else asked about this, so I must be missing something obvious, but what is this file:
SetCustomBacktestProc(“formulas\\include\\cbt_midlevel.afl”,True);
the cbt_midlevel.afl ?
and where would I find it?
ges
Hello Ges,
you don’t need the line to reproduce the results. I use CBT in this particular case to calculate spcific KPI’s.
However, sent you an email containing the CBT AFL code.
Frank
Hello Frank,
I want to ask for the cbt_midlevel.afl too. May I can get it as exercise-example.
Regards from Stuttgart
Helmut
Hello Frank,
very interesting study. Why is not possible to send a Amibroker script for testing without your add ons with
survivorship Nasdaq bias file. So we can test ourselfs and discuss with further ideas. I know it is
a lof of work for this survivorship issues. All other issues can everybody add on, but if I have your basic idea in a code then everybody will play and add ideas.
Hello Thomas,
you just need to change a line or so in the AFL statement and remove the check.
” .. AND IndexMember ” remove the check
That’s pretty easy.
Frank
Hello Frank,
actually it is not so pretty easy because if you cancel member then your staticvarset is missing this.
Could send me the the adjusted code without any survivorship issues.
Hi Thomas,
got your point:
just replace:
…. IndexMember = NDX100member(Name(),DateNum());
with
… IndexMember =TRUE;
Frank
You did this suggestion above to someone else, but if i do this you get LN 43 Col 24 Error 30, Syntax error
with member = NDX100member(sym,DateNum() ); And by the way I exchange trendstrength() for
TSI in the script as
Ratio = abs(Close-Ref(Close,-10))/ATR(10);
TSI = MA(MA(Ratio,10),100);
this is not working but I did your suggestion:
SetOption(“CommissionAmount”,0.00);
SetOption(“InitialEquity”, 100000);
SetTradeDelays(0,0,0,0);
SetBacktestMode( backtestRotational );
EnableRotationalTrading() ;
SetOption(“MaxOpenLong”,5);
SetOption(“MaxOpenPositions”,5);
SetPositionSize( 20, spsPercentOfEquity);
Ratio = abs(Close-Ref(Close,-10))/ATR(10);
Trendstrength = MA(MA(Ratio,10),100);
IndexMember = True;//NDX100member(Name(),DateNum());
idx = Foreign(“$NDX”,”Close”);
Listnum = 0; //here you enter the number of the NDX100 watchlist you want to use
list = CategoryGetSymbols( categoryWatchlist, Listnum);
Rank1_Score = 1;
Rank2_Score = 1;
Rank3_Score = 1;
Rank4_Score = 1;
Rank5_score = 1;
Rank1_bench = Trendstrength;//trendstrength();
Rank2_bench = abs(ROC(Close,200));
Rank3_bench = 1;
Rank4_bench = 1;
Rank5_bench = 1;
MArketOk = idx > MA(idx,20) OR idx > MA(idx,200);
//exec = IIf(Status(“stocknum”)==1,1,0)
exec = IIf(Name()==”AAPL”,1,0); //hard coded in order to improve performance, need to to this only once per bar for all stocks
for( i = 0;( ( sym = StrExtract( List, i ) ) != “”) ; i++ )
{
Maxitems = i+1;
}
if (EXEC[1]==1)
{
for( i = 0;( ( sym = StrExtract( List, i ) ) != “”) ; i++ )
{
Member = NDX100member(sym,DateNum() );
SetForeign(sym);
sym_v1 = IIf(member,Trendstrength,0);//IIf(member,trendstrength(),0);
sym_v2 = IIf(member,abs(ROC(Close,200)),0);
sym_v3 = 1;
sym_v4 = 1;
sym_v5 = 1;
RestorePriceArrays();
StaticVarSet(“IDXMember”+(i+1),member);
StaticVarSet(“RANK1?+(i+1),sym_v1);
StaticVarSet(“RANK2?+(i+1),sym_v2);
StaticVarSet(“RANK3?+(i+1),sym_v3);
StaticVarSet(“RANK4?+(i+1),sym_v4);
StaticVarSet(“RANK5?+(i+1),sym_v5);
}
}
for( i = 1; i Rank1 AND idx AND rank2!=0,1,0);
Rank2_Score = Rank2_Score + IIf( Rank2_bench > Rank2 AND idx AND rank2!=0,1,0);
Rank3_Score = Rank3_Score + IIf( Rank3_bench > Rank3 AND idx AND rank2!=0,1,0);
Rank4_Score = Rank4_Score + IIf( Rank4_bench > Rank4 AND idx AND rank2!=0,1,0);
Rank5_Score = Rank5_Score + IIf( Rank5_bench > Rank5 AND idx AND rank2!=0,1,0);
}
ComboRank = IIf(indexmember,Rank1_score + Rank2_Score ,0);
Score = IIf(marketok AND ComboRank>0 AND IndexMember AND DateNum()=1010101 AND DayOfWeek()==5) OR DateNum()>=1100901;
PositionScore= IIf(Condition1,score,scoreNoRotate);
Rank1_Score = IIf(indexmember,Rank1_Score,0);
Rank2_Score = IIf(indexmember,Rank2_Score,0);
Rank3_Score = IIf(indexmember,Rank3_Score,0);
Rank4_Score = IIf(indexmember,Rank4_Score,0);
Filter = DateNum() == EndValue(DateNum());
AddColumn(Rank1_bench, “Rank1_Value”, 1.2);
AddColumn(Rank1_score, “Rank1_Score”, 1);
AddColumn(Rank2_bench, “Rank2_Value”, 1.2);
AddColumn(Rank2_score, “Rank2_Score”, 1);
AddColumn(Rank3_bench, “Rank3_Value”, 1.2);
AddColumn(Rank3_score, “Rank3_Score”, 1);
AddColumn(Rank4_bench, “Rank4_Value”, 1.2);
AddColumn(Rank4_score, “Rank4_Score”, 1);
AddColumn(Rank5_bench, “Rank5_Value”, 1.2);
AddColumn(Rank5_score, “Rank5_Score”, 1);
AddColumn(ComboRank, “Combo_Rank”, 1);
SetSortColumns(-5,-3);
_SECTION_END();
Frank,
This is really useful. I’d been putting off tangling with static arrays and this code is a great jumping off point.
To get it to run I had to upgrade AmiBroker, because static arrays were only introduced around version 5.24.
Pre 5.24 versions might explain some of the problems people were having with this code up thread.
Please consider posting this code to the yahoo user list, as there is very little there re: static arrays.
Keep the posts coming and thanks for sharing all your hard work,
ftm
You mention TSI as a software.
Would be very helpful to have your system coded in their language as well.
Anybody able here to do that?
Hello Flipper,
TSI is an indicator that I developed, see the menue entry above.
Frank
Hi Frank, where do I find following includes ?
//#include ;
//#include ;
//#include ;
Hello Ton,
those aren’t available.
the most important one is the NDX100… which is responsible for eliminating survivorship bias, see here
http://engineering-returns.com/2010/09/02/the-impact-of-survivorship-free-backtesting/
Frank
Regards,
Frank
Hi Frank,
could you update the performance till now. How was the 2010 return and DD in %.
Thanks.
HI Thomas,
have planned to do an “updated” post soon…
Frank
Hey Frank. If an NDX100 stock has an great TSi, and ROC but is total illiquide, what is your approach in such situation? Do you have any volume filters for the rotational trading?
Early thanks!
Ilya
Hello Ilya,
not sure I understand your question. Based on my experience I haven’t seen any illiquide stock being part of the NDX100 / SP500. I theory that can be the case, just think of Volkswagen a few years ago. But this is an EXTREMELY rare scenario. That’s actually the reason why I trade SP500 / NDX100 stocks only.
Frank
Hi frank, great system to stimulate ideas. To confirm are the orders placed as market on close or market on open ?
Hi,
Close2Close.
However, I did a test on Open2Close … hardly any difference in performance.
Frank
How do you realized Open2Close? I can set these paramaters in Amibroker Backtest settings (and in the code as well) but checking the results reveals that Amibroker is using for selling the same price that is defined for buying. E.g. Buy at Open, Sell at Close is carried out as Buy and Sell as Open.
Hello David,
in my opinion there is bug in AmiBroker. The backtester takes the AFL specific commands, e.g. Buyprice = Close or Sellprice=Close….
However, this isn’t true for rotational trading. In this case the backtester WILL always use the settings from the settings-page. Hence AFL comands are ignored.
Frank
Hello Frank,
I agree. But in my Amibroker environment work the parameters set in the settings-page neither. Close will carried at buy, independent from the sell parameters. Can you consider this behaviour as well?
David
Hi David,
i tested Open2Open trades. Didn’t find any issue. Changed settings manually to buyprice = open and sellprice = open
Furthermore I’ve got this:
SetTradeDelays(1,1,0,0);
Frank
Hi Frank,
Open2Close leads in my backtesting trade report to buyprice = open and sellprice = open instead of buyprice = open and sellprice = close if I compare the price shown in the backtest report with the original share prices.
David
Hallo Frank,
your approach is very nice. I read your article in Traders too, but it seems, that I didn’t understand your ranking algorithm fully.
Is there no precedence between TSI and “Trend” (i.e. derived from ROC). What, if the stocks A to E have top five TSI, but “Trend” (however) is poor? Maybe, stocks V to Z are best in “Trend”, but their TSI is low. Do I misunderstand something, can you help?
Hello Burkhard,
Thanks for your feedback. Both attributed are equal weighted. In theory what you described can happen, in practice it’s rarely the case.
Regards,
Frank
Frank, what would be the results if you
a) used the market filter AND only used ROC (200) for ranking the stocks?
b) used the market filter AND only used TSI for ranking the stocks?
Thanks.
Hello Anton,
in the bottom of the post you find the AFL source code I used for the test. Please use that to run your own tests.
Regards,
Frank
Thanks, Frank. I do not have skills to use the code. Approximately how much does TSI add to the returns in this system?
Hi Frank,
Your rotational strategy is nice. I want to test the strategy in Indian index Nifty.
I saw the following codes in the AFL
#include ;
#include ;
#include ;
SetCustomBacktestProc(“formulas\\include\\cbt_midlevel.afl”,True);
May I know, how can I get those include files?
Also the RSI/TSI formula is giving sintax errors:
SetOption(“CommissionAmount”,0.00);
SetBacktestMode( backtestRegular);
SetOption(“InitialEquity”, 100000);
tsi = trendstrength();
LongCond = Close>MA(Close,50) OR ROC(Close,150)>0;
Setup1 = Longcond==True AND ((tsi>1.65 AND RSI(2)<18) OR (tsi<1.65 AND RSI(2)1.65 AND RSI(2)<13) OR (tsi<1.65 AND RSI(2)= 1999 ) * sigScaleIn;
Sell = (tsi>1.65 AND RSI(2) > 61) OR (tsi 39);
SetPositionSize(33, spsPercentOfEquity );
I feel that error is for the tsi = trendstrength();
Where can I found the trendstrength() function?
Waiting for your kind reply.
–
Regards,
Indrajit
Hell Indrajit,
these include files are not really relevant, one is for my default settings and the other contaings my self-developed AFL functions. The NDX100…AFL contains my survivorship bias free index constitution.
TSI can be found here: http://engineering-returns.com/tsi/
By the way, the most recent version of the strategy can be found here:
http://engineering-returns.com/2011/02/15/strategy-review-rotational-trading/
Regards,
Frank
Hi Frank,
What happens if a metric (such as TSI) gives exactly the same values for different stocks (E.g. TSI(A) = 1.4 and TSI(B) = 1.4)? In such cases, do both stocks get the same score?
Regards,
Wils
Hi Wils,
yes, you can give them the same score.
Frank
Thanks Frank. Have you considered other means of normalizing the scores across assets, such as percentile-rank or zero-mean-unit-variance? This way one can perhaps combine both continuous and discrete metrics.
Wils
Hi Frank,
Could you post the code for:
#include ;
#include ;
#include ;
I could not find it.
Thank you very much!
Hello,
these are my own functions. You don’t need it to get the system running.
#include ; -> function for survivorship bias free index constitution
#include ; -> default settings
#include ; -> prop indicators, e.g. TSI
Regards,
Frank
Hi Frank – Excuse my ignorance…
How to get these files listed on your amibroker code.
#include ;
#include ;
#include ;
SetCustomBacktestProc(“formulas\\include\\cbt_midlevel.afl”,True);
Pontric
Hello Pontric,
these files aren’t part of the source code….
Just remove it ..
Frank