Discussion in 'Journals' started by globalarbtrader, Feb 11, 2015.
Portfolio construction; long only portfolios; geared towards cross asset ETFs
I use your python system to run a backtest using all instruments you have uploaded in github, but for some reason the first time a portfolio is being formed is approx. around year 2000 in spite of the fact many individual instruments having much longer history. How can I make the portfolio start in, say, 1980?
Can you make sure you are using the latest version and post the enterity of exactly what code you ran; including any config changes.
I think I found out why I get that - I use EUR as base currency and understandably EURUSD exchange rate data starts in 1999. If I use the default base currency of USD, the system forms a portfolio starting in ~ 1970/1980.
Let me know if you still would like to know the exact code I ran.
Makes sense. Glad you worked it out.
Brief monthly update: up around 4%
That will be all as I am on holiday. I will do a full update for 2 months in September.
In a previous blog post, you mention IB was pretty much the only game in town for retail API/Futures. I want to see if you still think this is the case?
I haven't had the time to look into other options, as there isn't any immediate benefit to doing so.
There would be big switching costs to setup another broker. But purely for redundancy it's worth doing. I'm very happy to hear about other options.
Hey there.. I was looking at your code and I have a question about what's in this yaml file. At the very bottom, you define what looks like some sort of custom groups (rule_groups, style_groups etc) which could be presumably used for grouping subsystem p&l's OR constraining forecast or instrument weights along these dimensions. Do you have any example of how you use these groupings?
Second question, when I optimize all of the instruments you provide (38 of them I believe) using the with-cost optimization method you provide here, I see that I get instrument weights that are less robust than the no-cost optimization. I have not delved into your code but I'm just curious, do you expect this to be case?
Hey GAT, when I look at the performance of Winton Futures Program (at the bottom of the page http://ctaperformance.com/wntn), it looks like they have milder drawdowns than those of your system at the same vol level. By using decreasing capital multiplier in drawdowns, your system can reduce drawdowns somewhat, but are there any other drawdown reduction techniques that you have come across?
Separate names with a comma.