Hybrid Asset Allocation

One correction is to replace offensive assets with cash if their momentum values are negative.  See bolded section in the rules below.  Results are still not as good as the paper, mainly because we get caught in the March 2020 pandemic crash.  I think I'm still missing something…

1. Calculate the momentum of each asset in the (risky) offensive, defensive (BIL/IEF) and canary (TIP) universe, where momentum is the average total return over the past 1, 3, 6, and 12 months.
2. Select only the best defensive ‘cash’ asset (BIL or IEF) when TIP is bad, or else allocate 1/TopX of the portfolio to each of the best TopX half of the risky assets (equally weighted), while replacing each of those TopX assets by cash when ‘bad’ (ie. has non-positive momentum).
3. Hold all positions until the final trading day of the following month. Rebalance the entire portfolio monthly, regardless of whether there is a change in position.

Tristan F  Curious if you made any more progress? Attached is my attempt. I also got caught in the Covid dip. Any ideas?

Hi jam ,

Apologies for the late reply.  I have not made any progress on this and appreciate you taking a look at it.  Usually QC results line up well with Wouter Keller's published results, but not in this case.

1. QC has data issues that i've caught in the past, so i can't rule out other bugs issues with their accuracy.
2. Your drawdowns will never match, as all these papers use monthly drawdowns not daily, so intra day you can be -70%, but that's never caught in the stat padding that a lot of researchers engage in
3. Remember speaking to one author won't mention names and saying yeah well real drawdowns are like possibly double…wtf? lol. As they can't market strategies with -40% vs -15%

Has anyone tried combining Hybrid Asset Allocation and Bold Asset Allocation in a single portfolio to see how the combined strategy performs?  I am attempting to do it but am too much of a novice to get the code right.

with our chatgpt friend I ran this algo through it, ran kellers verbatim rules, and chatgpt said it's implemented exactly, after i gave it a look over manually also….so seems it's implemented correctly.

The issue is either a data issue on quantconnect which is a known glaring issue, not sure if they've done much to fix it. 

or like i mentioned before all these researchers use monthly data for their stats so on a monthly DD basis it looks good, on a daily totally breaks down. 

But given the massive discrepancy would assume it's bad data possibly for TIPs

One issue is your safe universe only includes BIL when it should just be whatever the best performer is in the defensive universe as i understand it. 

self.safe = ['BIL'] should be self.safe = ['BIL','IEF'] which is just self.selD = ['BIL', 'IEF']

IEF performed very strongly during COVID because of the sharp drop in interest rates. Not having that as an option would have been a big performance loss. 

I also don't follow what you're doing here:

# Determine number of canary securities with negative momentum

        n_canary = np.sum((mom_avg[self.selP]<0)*1)/len(self.selP)

        # % equity offensive

        pct_in = 1-n_canary

        # cash return

        pct_safe = 0


the canary universe only has 1 item and always 1 item. 

There are some other issues in the code as well regarding the calculation of the momentum as well as the weights. 

Also you can't simply use the prices of fixed income ETF's to get their total return which is what the momentum is meant to be based upon. Instead you need to incorporate the income they are paying out through out the period. This is a vital difference in the effectiveness of the crash protection the strategy uses.