rikderParticipant03/17/2015 at 5:04 pmPost count: 22
I came across a couple blog posts related to Momentum rotation strategies and obtaining clean backtest results that I wanted to get your comments on. The links are here: part 1: http://dtr-trading.blogspot.com/2015/03/momentum-rotation-strategies-and-data_15.html part 2: http://dtr-trading.blogspot.tw/2015/03/momentum-rotation-strategies-and-data_16.html
Rather than summarize, I’ll use the author’s own words, since he put is so well. Dave @DTRtrading says: “Now to the crux…the data used by rotation systems has an impact on individual fund ranking scores. Most rotation strategies that are published on the internet (commercial systems, blogs, etc) use split and dividend adjusted price data to generate the ranking scores. The entire adjusted price data series changes with every dividend that is issued, which changes all past prices in the time series. When a dividend is issued for one of these funds today, all of the past adjusted prices change, which will change the past rank scores for that fund. When the past rank scores change, a fund that was in position 1, can move to position 2, which then changes the backtest results.”
My understanding is that on a go forward basis you are using non-adjusted data therefore your backtests need to use non-adjusted data (although your results should reflect adjusted data to account for actual returns). If you backtest with adjusted data you will inadvertently rank funds differently in relation to one another on a look back basis than you actually would have real time. This occurs because each fund adjusts its data at different intervals throughout the year.
Dos this dynamic affect the funds you use for your rotation systems? If so how do you account for it? It seems pretty important so I assume it must be something you have factored into your testing.
Thank you very much. I appreciate all your great work!
VangelisModerator03/19/2015 at 10:26 pmPost count: 165
Thanks for the post; I read the bloggers work when it came out and I was immediate bothered by the errors at the time.
The blog author appears to be operating under the impression that price appreciation is the sole guide to relative value, momentum, etc. However, when comparing an asset with no dividends/interests etc with another with substantial dividends, the analysis is totally distorted by not properly including the dividend impact when choosing which asset to own for the next allocation. I have recreated Yahoo’s dividend adjusted data methodology, and have even constructed alternative methodologies. While Yahoo does occasionally make mistakes, which is why often we check several sources, the underlying approach is a solid methodology to create a total return stream for an asset. The total return stream is an excellent way to compare different asset relative value to the investor. The concept of fund adjusting for dividend data at different times is wrong — it is all done tied to the ex-dividend data, which is when the economic value is acquired by the investor.
Sorry to be harsh, however, the blogger made mistakes in the framework of his analysis.
I hope this helps.
VangelisModerator03/20/2015 at 5:49 amPost count: 165
I will agree with Scott that when comparing assets we need to take into account dividends. Its’ like comparing two cash deposit accounts, on paying 0.5% and the other 1% interest. They are not equal.
There is another subtle point however. Let’s say today is Monday. We download Yahoo daily data (not historical) and rank. Then we trade. If a dividend was present, tomorrow or in the future, when we download adjusted data, the prices will change and the rankings will change. But that is not due to Yahoo’s methodology but rather on using both adjusted (historical) and unadjusted data (daily). This problem is present in daily ranking systems where yahoo may delay the adjustment of data. It is less of an issue on monthly or weekly strategies as long as one uses yahoo’s “historical” feed.
VangelisModerator03/30/2015 at 7:37 pmPost count: 165
Kudos to DTR Trading blog for admitting the error in the earlier conclusion regarding dividend adjusted data.
We are all learning the hard way as we dig into this stuff. Good to know they figured it out.
You must be logged in to reply to this topic.