I have put a tremendous amount of effort into the development of an ETF 2X model which will produce a more consistent equity curve than those produced by my existing models. In choosing the ETF’s to include, I considered only those which are highly liquid. The cross correlations were determined and I adjusted the weights of the selected ETF’s to produce a high Calmar Ratio (high compound annual growth rate and low maximum drawdown) as well as a low Ulcer Index. ETF’s which did not improve the Calmar Ratio were discarded. Canadian ETF’s for international equity markets have pathetically low trading volumes so I had to use US ETF’s.
As I believe in using 2:1 leverage in conjunction with my timing algorithms, the model is based on borrowing an equal dollar amount of existing capital for each ETF. For example, if the model calls for a $5,000 purchase of GLD then $10,000 is purchased and interest is calculated based on a 6% interest rate. Dividends are not included in the calculation of the returns.
Below is the equity curve for my Combined ETF 2X model based on month-end returns:
The ETF’s that have been used in this new model are:
DBC – Powershares DB Commodity Index
EEM – iShares MSCI Emerging Markets Index
EFA - iShares MSCI Europe, Australasia, Far East Index
EWZ - iShares MSCI Brazil Index
GLD – SPDR Gold Trust
IYR – iShares Dow Jones Real Estate
IWM and RWM - iShares Russell 2000 Index and Proshares Short Russell 2000 Index
QQQQ and PSQ – Powershares QQQ and Proshares Short QQQ
TLT - iShares Barclays 20+ Year Treasury Bond Index
DBC contributes nothing to the above equity curve for the first 29 months and IYR contributes nothing for the first 16 months since these two ETF’s are relatively new.
I believe in always pointing out the caveats with my models, so here goes. It is possible that the correlations between the ETF’s will change going forward and affect the results. Of course, the correlations are never static and thus weren’t static throughout the backtest. I may have optimized the model by determining what worked the best in the past and therefore future results may be different. However, from the testing I have done, the model is robust. Finally, past performance is no guarantee of future performance.
For what it is worth, the model currently holds GLD, RWM, PSQ and TLT all of which are currently profitable.
The maximum drawdown based on month-end returns is 11.5%.
Now, the question is “What do I do going forward?”. This model will have perhaps 75 to 85 trades per year and requires more work each day to update than anything I have presented before. I lead a hectic life at times and therefore I am not certain that I will have the required time every evening to update my files in order to determine if trades are warranted the next day. What I am considering is e-mailing the trades to my Collective2 subscribers for a period of time to determine if I can keep up with the required effort plus generate a real time performance history of the model.
Feel free to e-mail questions or provide feedback.
FJP
