[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

CASH index is poor model for Oddball & Variants



PureBytes Links

Trading Reference Links

 

It has been suggested by some to use the cash index (SPX, SPY, INX, etc.) to 
chart and trade the OddBall system and variants. 

My research shows that historical backtesting results in TS using the cash 
index can generate wildly inflated, overoptimistic performance results. 

Beyond the well known fact that trades placed at market open on the cash 
index do not reflect the futures open due to globex activity (in which case 
price can have already moved substantially during globex away from prior day 
close), there seems to be some other differences in price behavior that add 
to the over optimistic exuburance of cash index backtesting results as well. 

Recent example using actual SP futures SPM2 data: 

TNP on data from 3-14-02 till 5-17-02 using SPM2 data with Original OddBall 
at 7,3,1 = $11,925 TNP with $18,700 MIDD 

TNP on data from 3-14-02 till 5-17-02 using Cash index data with Original 
OddBall at 7,3,1 = $14,960 with $17,427.50 MIDD 

In this case the cash index results show approx. a 25% performance inflation 
for TNP. 


I have an interesting, initially promising OddBall variant that shows: 

TNP on data from 3-14-02 till 5-17-02 using Cash index data = $27,787.00 

TNP on data from 3-14-02 till 5-17-02 using SPM2 data with = -450.00 
(negative) 

If you compare longer term backtested results for the Cash index and the 
futures contract you will also see wide enough differences to conclude that 
the cash index is unrealistic for projected performances. 

 

My results on SP futures data show that by the end of SPZ1 contract, short 
term backtesting (optimization) showed clearly
that the 11:00 OddBall system would have earned more using inputs 7,7,7 and 
that traders who would have continued using these inputs would have earned 
substantially more than those who would have used the default inputs of 
7,3,1 in year 2002. 

There appears to be a case for on going research on determining length of 
look back period for determining optimal inputs and then applying those 
inputs to a predetermined walk forward period in the future. 


Johan