Part 6-7 of the Feynman Portfolio Study applies the SHY momentum filter to the Feynman “Momentum” Portfolio introduced in Part 5 of the Study.
Also in this Part 6-7, I introduce the concept of allocation weighting based on (momentum) Ranking as an alternative to allocation weights based on (“Dynamic”) optimization as used in Parts 3, 6-1 (Section 6.1.2) and 6-5 (Section 6.2.3) of the Study.
This information is not available for publication elsewhere on the Internet.
Part 6-7 of the Feynman Portfolio Study is available as a downloadable Word File here with detailed supporting appendices available here (Appendix 12) and here (Appendix 13).
(Visited 119 times, 1 visits today)
Discover more from ITA Wealth Management
Subscribe to get the latest posts sent to your email.
I compared the Equal Weighted Momentum w SHY filter in the study with results from ETFReplay.com that uses the same momentum model with weightings. The ETFReplay results are significantly higher for 4 ETFs from the Feynman 18 list.
The Study results: 75.95% Total Return, 9.87% CAR, -12.6% MDD
ETFReplay results: 162.1% Total Return, 15.2% CAR, -15.0% MDD
The ETFReplay results are close to the same results I obtained using the momentum model that I’ve implemented in AmiBroker (CAR 16.0, -20%MDD) for 4 ETFs from the Feynman 18.
This leads me to question the difference you show in Fig 44 between the Equal Weight and the Momentum Weighted models. That is, do your results understate both, or is there less of a difference between the two? Any ideas or comments?
Bob Krishfield
Bob,
Good observations and comments. First, the easiest part to answer, and probably the most important/significant – my results will understate performance compared to ETFReplay outputs – and hence actual portfolio performance -because I have not included dividends in the equity performance plots – therefore we might expect 2-3% CAGR to add to my performance numbers – not insignificant. Although I use (dividend) adjusted price data when calculating momentum, returns are calculated using closing price data on the (quarterly or monthly) analysis date. It is too time consuming for me go back and to manually check on dividend payments for all ETFs on a monthly (or even quarterly) adjustment period over a six-year back-test. I just recognize/accept that the performance is understated and conservative.
You can input the Feynman portfolio into EFTReplay and choose to include the top 4 ETFs and it will rebalance monthly/quarterly to an equally weighted i.e. 25/25/25/25 portfolio – I ran this from 2007 – present and came up with numbers similar to yours 167.8% total return, 15.5% CAGR, 15% Volatility, 15.1% MDD. (Quarterly rebalancing generates slightly lower returns (~140%, 13.75% CAGR) with higher Volatility (~15.6%)). Running it from 2008-present results in 123.1% total return with 14.8% CAGR (monthly rebalancing) or 101.8% return, 12.8% CAGR (monthly rebalancing) – thus the start/finish dates and rebalance frequency need to be taken into account
The fairest comparison with the performance numbers provided in the Feynman Study (mid 2007 to mid 2013) would seem to be to take the 2008-present quarterly rebalanced numbers from the ETFReplay back-test (missing 6 months on the front end but adding 4 months on the back end) and to compare these to Figure 44. Now we are comparing a 12.8% CAGR with a 9.87% CAGR – a 2.9% difference. Making allowances for the “alignment” problem, I believe this explains the differences you note.
Unfortunately ETFReplay cannot be used to back-test (only forward-test) a portfolio in which the asset weights are changed on a monthly/quarterly basis but I believe that the Feynman Results do show that there is a benefit in doing this. If we accept this, and take the results from Section 8.3 of the Feynman Study, which shows a 15.39% CAGR on a monthly rebalanced portfolio and add 2.9% for dividends, we get a total return of ~18.3% – we can compare this with the above 14.8% generated by ETFReplay from an equally weighted, monthly rebalanced, portfolio.
Hope this addresses you questions
David
Thanks for the clarification. I adjusted my dates in my AmiBroker model to 6-2007 to 6/2013 and the equal weighted momentum model numbers were closer, but I’m using total returns data from Norgate. I’d like to automate the Weighted formula in Amibroker to backtest alternatives.
You mentioned the weighted momentum formula is proprietary. Are there any hints you could provide to help me generate a close analogy to it for personal use in Amibroker. My email is bobk@bobsden.com if you want to make a private reply.
Bob