ETF Prophet

A while ago i looked into how inside bars impact mean reversion trading for SPY. The SPY is an index ETF mirroring SP500 and  is well known for its tendency to mean-revert.  I’m writing this post because i wanted to understand and document how this effect plays into high momentum stocks such as the Nasdaq 100 member.  In order to fully benefit from this research i recommend you read this post first.

A simple test

Let’s first look into a basic RSI(2) strategy: enter when RSI(2) is crossing under 50 and exiting above 50. No trend filter has been added. Some background about the testing methodology: all trades are close to close, no commission. I used the NDX100 member stocks as of today, didn’t adjust the historical index constitution. Trades are starting from 1998. You might notice a relatively low CAR%, that’s because i wanted to ensure that every possible trade is taken (meeting the entry criteria). Hence position size is 1% per trade (to allow for a maximum of 100 concurrent trades). Of course one wouldn’t trade 100 concurrent positions, but for this research i needed to do so.

The first column (“NO”) is taking the trade as soon as RSI(2) < 50. The second column (“1. bar”) represents performance for a system that does delay the trade by one bar in case an inside bar occurs. The third column (“2. bar”) represents a system that does check for two consecutive inside bars. Again, please read this post in order to understand the concept of nested inside bars. I know, this might be a bit confusing. Bellow you find the Amibroker AFL code in case you want to fully embrace this.

The good: basically we see an improvement on almost all metrics. The sample size is very significant > 40.000 trades.

The bad:  What if the improvement seen has a different cause? Namely the effect of simply delaying the trade for one or more trades might give you a better risk reward, because RSI(2)<50 isn’t terrible efficient. A more “realistic” test might help answering this (especially when entering trades at more extreme RSI(2) readings)..

A more comprehensive test

Let’s look at a strategy that goes long only when the following criterias are met:

• Rate of change (200)  >  0
• RSI(2)  < 25

Exit as soon as RSI(2) > 50.

Again, we see better performance while reducing risk.  The effect is most pronounced at the 1st. and 2nd. bar.

.

.

.

.

.

Conclusion

Entering a RSI(2) based trade at an inside bar decreases your performance.  I would rather delay the entry or reduce my bet size. This is another small brick in building a robust trading system.

Having done this comprehensive test with a very large sample size on SPY ETF as well as Nasdaq 100 stocks I’m rather confident in the outcome of this research.

.

.

AFL Code

#include <generall.afl>;

#include ;
SetBacktestMode( backtestRegular);
SetOption("CommissionAmount",0.00);
SetOption("MaxOpenPositions",100);
SetOption("MaxOpenLong", 100);
SetOption("InitialEquity", 100000);
SetPositionSize( 1, spsPercentOfEquity );
i_1 = HighRef(Low,-1);
i_2 = Ref(i_1,-1) AND HighRef(Low,-2);
i_3 = Ref(i_2,-1) AND HighRef(Low,-3);
Buy   = Year()>1998 AND RSI(2)<25 AND ROC(Close,200)>0 AND i_1==False AND i_2==False AND i_3==False;
Sell  = RSI(2) >50 ;
#include ; #include ; SetBacktestMode( backtestRegular);SetOption("CommissionAmount",0.00);
SetOption("MaxOpenPositions",100);SetOption("MaxOpenLong", 100);SetOption("InitialEquity", 100000);SetPositionSize( 1, spsPercentOfEquity ); 
i_1 = HighRef(Low,-1);i_2 = Ref(i_1,-1) AND HighRef(Low,-2);i_3 = Ref(i_2,-1) AND HighRef(Low,-3);

Buy   = Year()>1998 AND RSI(2)<25 AND ROC(Close,200)>0 AND i_1==False AND i_2==False AND i_3==False;  Sell  = RSI(2) >50 ;