ETF Rotation Strategy

Jul 20, 2017

Why do you rank them before? I mean, you do the returns for each ETF (20d, 3m, 6m, 1yr), but you don't need to rank them yet. First, do the score, and then rank them.
I would delete this:
day20_rank=day20_ret.rank(ascending=False)
day3mo_rank=day3mo_ret.rank(ascending=False)
day6mo_rank=day6mo_ret.rank(ascending=False)
day1yr_rank=day1yr_ret.rank(ascending=False)
vol_rank=volatility.rank(ascending=True)

Jul 20, 2017

Thank you,

So just to clarify. I want to rank by the average of their returns, not the average of their ranks.
i.e. I was doing
average_ranks=(1+3+7+4+3)/5
but you say
average_returns = (0.01 +0.09 +0.15 +0.10+0.19)/5
then:
rank = average_returns rank

That makes sense, here is the backtest.

Jul 24, 2017

I still don't know why the alpha is so low compared with other similar strategies like this. Any one? https://www.quantopian.com/posts/global-market-rotation-strategy-buggy-implementation

Jul 25, 2017

I think it has to do with the performance of emerging ETFs, I used sector ETFs. I took out everything but the 3 month returns and the volatility. It performed a lot better. There's a leverage problem though, so it's not perfectly accurate right now.

Jul 25, 2017

Here is the performance with the sector etfs instead of emerging market. Interesting!

Jul 25, 2017

I had to replace EDV with TLT. The partial fills from the stocks ordering were messing up the leverage. I'm sure there is a better way, to fix the problem and still trade EDV.

Jul 26, 2017

When you introduce the volatility of the ETF, I think (might be wrong) that you are rewarding those with more vol. because you are adding it.

-ETF A has a 4% return, and 2% volatility = 0.7*4%+0.3*2%=2.8+0.6=3.4 Score
-ETF B has a 4% return , and 3% volatility = 0.7*4%+0.3*3%=2.8+0.9=3.7 Score

Jul 26, 2017

I did this based on your approach, but still not as profitable as yours

Jul 26, 2017

Also picking SHY instead of TLT may be a more accurate option

Blue Seahawk

Start date of the EB code showing 800% is being changed to 2009. Tried this, same:

def initialize(context):  
    set_symbol_lookup_date('2004-01-31')

@Blue So are you saying it is not trading until 2009?
@Ro FG You are exiting every time it is below the 200 day SPY, and putting it in EDV. It doesn't work as well as a safety because after the recession you get those signals, and it still goes up also.

I normalized the returns by dividing them by their volatility . I am not sure if this is correct. I also combined all the etfs from both strategies. I am trying to lower the drawdown, because the returns seem way to high. I wouldn't want to just invest in emerging etfs. What comes up must come down!

Relatively a flat strategy the last 3 years or so. Did really well up until 2010.

Blue Seahawk

@EB I was saying the start date field changed to 2009 automatically. Clicking on that calendar after the test, there was a message to the effect of first available date for securities being traded, surely you've seen that a few times eh. If there's a log message about it as well, I missed it.

Switching to the second-most recent backtest above with apparent returns of 197, its profit per dollar was just 60 due to margin. A common problem. This version has apparent returns of 233 and profit per dollar is 232, so almost no margin. So not only is 233 higher than 197 but also 232 is higher than 60. Using profit per dollar activated|invested is the only way I know for comparing two algorithms apples-to-apples and seeing more clearly what they are really doing by neutralizing, or adjusting for, their use of capital, i.e. amount of risk. Maybe people don't go that route because of the reduced speed.

There were evidently so many partial fills (the cause of all that margin along with the inherent unpredictability that goes with it hand-in-hand, unintended margin with order_target_percent when there are partial fills) that I took an unusually extreme approach in reactivating close() by running an entire day before balance and again the morning of that monthly balance, as this, the first try. @Ro FG, you might want to use similar and see what happens in your version.

Aug 1, 2017

interesting point. Dividing by the volatility lowers the drawdown, but the ratio returns/drawdown is worse.