KEI based Strategy

Here the framework algo

Simone Pantaleoni,

Good idea and in line with my personal preferences. 

Thank you for sharing.

Before we start to add on top of this framework we need to fix some existing problem.

1.Every line in orders log has  Warning: fill at stale price.

Try to change resolution for stocks and bond from Daily to Hour.

And change 

self.TimeRules.AfterMarketOpen("SPY") 

to

self.TimeRules.AfterMarketOpen("SPY", 31)

This should solve the problem.

Good luck .

Interesting project, looking forward to the discussion. Well done, Simone Pantaleoni and Strongs!

Hi all guys, this is just a base from which to start. The starting idea is to also exploit the concept of in & out and the intersection of ROC by Vladimir  . The LEI could actually be exploited for a dynamic selection of ETFs, essentially identifying phases of the economic cycle and then selecting groups of ETFs that perform best in that specific phase of the cycle.

New version with the fixes from Vladimir 😊 

Simone Pantaleoni,
Completed as requested!

The second problem is that you have initialized the position in QQQ, 
which is contrary to the main logic at the beginning of backtest.

To solve this. I define

	initial_asset = self.stock if self.mom.Current.Value > 0 else self.bond 

        if self.init:
            self.SetHoldings(initial_asset, 1)

To make self.mom visible on the chart changed self.MOM to self.MOMP

Here is "SIMON LesFlex" modified by Vladimir
with changes I mention above and some cosmetic changes.

Enjoy.

The idea would be to turn the indicator like that

First of all, great work :)

I've had some time to look at this, and I want to be sure i'm interpreting it correctly...

1. We run our rebalancing logic every week, 30 mins after market open
2. If the Quandl KEI indicator is trending upward (positive MOMP), then we  liquidate Bonds and go long Stocks
3. If the Quandl KEI indicator is trending downward (negative MOMP), and the KEI under 100, then we  liquidate Stocks and go long Bonds

Questions:

1. I see we are tracking the SMA of 1 period, so we arent actually calculating a moving average. We're just plotting the value. Is this intentional, just for plotting sake? or should we actually be trying to smooth out the KEI with an SMA?
2. The KEI seems to be recalculated every Month (at most). So, do we actually need to check for rebalancing every week? Or are we using a weekly frequency just to make sure we catch the monthly change as soon as it happens?
3. How, exactly, does one interpret the KEI? i've read the description on the page, but it isn't very helpful. For example, why do we want KEI to be under 100? What does that value represent? 

Hi .ekz. ! many thanks for appreciating it. :) 

Let me try to answer your questions:
Questions:

1. I see we are tracking the SMA of 1 period, so we arent actually calculating a moving average. We're just plotting the value. Is this intentional, just for plotting sake? or should we actually be trying to smooth out the KEI with an SMA?
   1. KEI is already pretty smoothed. I just use the SMA 1day trick to quickly address the value in the code easily (but indeed, isn't really “clean”)
2. The KEI seems to be recalculated every Month (at most). So, do we actually need to check for rebalancing every week? Or are we using a weekly frequency just to make sure we catch the monthly change as soon as it happens?
   1. Exactly, just to be sure to catch every change (which occurs anyway montly)
3. How, exactly, does one interpret the KEI? i've read the description on the page, but it isn't very helpful. For example, why do we want KEI to be under 100? What does that value represent?
   1. take a look at the link below :)

Perfect, Simone Pantaleoni. Thanks for this.

Hi guys!

As suggested, we have been working behind the scene on applying the concepts shared above about the different phases of the market. Based on that, we apply ETF-selection accordingly, out of a basket of sectorial ETFs (we kept a similar switch to shift to bond in case of possible market downturn using the leading indicator) 

Some improvement on Total Return, sharpe and so on. 
Feel free to work on this release, fix it in case I've made mistakes (here I rely on Vladimir  as PeerReview :P ) and improve it (there's still something to optimize - but don't want to share all in once so… work on it and share your outcomes with us! :) )
 

Surely there is still a lot of work to be done, but this is definitely a great start. Levels for identifying the phases of the business cycle with associated ETF groups was presented. We await more ideas from you to improve it.

Great stuff Guys thanks for sharing. I have a question and a comment. I have a Quandl free account have not used it much. I plugged in my key and started looking at the algo. My question is how does the self.history statement get the Quandl data to sync up with the start of the algo? i see the request for the 1400 bars, so does the Quandl request use the start date of the algo as it start retrieval date? I see the data frame adjusting to the different start dates I tried. Any clarification greatly appreciated.

My comment is i have been following a youtube channel Hedgeeye. The gentleman who runs the channel is an ex-Hedge fund manager who also uses a macro view of investing. He uses inflation and gdp rates of change for his strategy much like this algo uses the KEI data as leading indicators.. He uses ETFs like this algo but also uses different styles based on his macro view of 4 Quadrants. Depending on the Quads as he calls them he picks the ETFs and the investment style (Momentum, V:alue, High Beta, Low Beta etc.)

A full explanation of his approach is on Hedgeeye.com, but here is a link to his strategy

https://app.hedgeye.com/insights/77156-chart-of-the-day-what-works-in-which-quad?type=macro

Frank

Sorry, the grid got messed up when i pasted it

          
Quad 1  Growth accelerating  Inflation slowing
Quad 2  Growth accelerating Inflation accelerating
Quad 3  Growth slowing  Inflation accelerating
Quad 4  Growth slowing Inflation slowing

Hi Frank,
History method takes the StartDate as the anchor retrieves data before it for the requested period. Refer to the following source code.
 

keihist = self.History([self.kei], 1400)

returns the last 1400 bars of history but we have the option to pass start and end dates to the History method.
 

keihist = self.History([self.kei],self.StartDate-timedelta(100),self.StartDate-timedelta(10))

returns the historical data for custom 90 day period.
Refer to the attached backtest, which demonstrates both approaches just comment out any one.
Best,
Varad Kabade

Varad,

Thank you for the explanation now i understand 

Frank

Hi there,

interesting stuff!!

Here are my 2 cents:

• KEI data was Europe, is now Total to work together with US stocks.
• The number of regimes is increased to 4 for rising and 4 for declining.
• A regime plot is included: increasing = regime 1,2,3,4, and decreasing = regime 3.8, 2.8, 1.8, 0.8 so that the regime curve follows the momentum curve intuitively.
• A benchmark bool is added using QQQ and TLT only.
• The attached backtest does not include my personal Quandl code, so to reproduce you have to register yourself.

And the results using QQQ and TLT only:

Results using an amended version of Sector ETF's:

Results using an overfitted version of Sector ETF's:

Suggestions for the way forward:

• Maybe the weighting of ETF could be done from a preselection based on the regime, plus a dynamic based on their actual performance.
• Also, further ETF such as SPY growth / SPY value could be included.
• I found a note that this KEI has a delay of 2 months, so it needs to be confirmed if it can be used for trading. Maybe an addition would be possible using High-Yield Spread: 

Have fun,

Frank

Expanding on Frank's work, I've added to the number of assets allocated depending on the regime.

I've attempted to replicate an asset allocation chart that I have which tells you where you should put your money depending on the trajectory of the economy. 

For example, in a scenario of better than expected growth forecasted, it suggests allocating to 37% US equities, 18% non-US equities, 25% US-fixed income, 3% non-US fixed income, 13% alternative investments, and keeping 4% cash.

This doesn't translate perfectly to this strategy but I've done a crude implementation that follows more ETF's weighted based on this chart. For the US-equities component I've maintained the same sectors suggested in the previous version, only with different weighting.

The optimal implementation of this version of the strategy in my opinion would be to use universe selection to choose the best companies in each index, rather than following a whole sector. I think this would increase performance dramatically. Unfortunately, I'm not advanced enough to do this sort of dynamic universe selection.

Thanks to Frank Giardina and Tom Penrose for the good work and for moving the thread forward!

You're going on the exact direction I was meant to drive this discussion, when we've started it :) 
I've been a bit busy lately on other projects, but next step will be exactly as Tom outlined, about stock-selection to pick up best 4-5 stocks for each sector.
Challenge there is to make the code decently quick, so I think a quite strict pre-selection of stocks will be needed

I think we got the wrong Frank, I made some comments and certainly want to help but Frank Schikarski  did more meaningful work with his post.