It ranks second in performance on Allocatesmartly both for 10 and 20 years.
All of the criteria are not stated on the website, so there is some room for guesswork. Here is my take:
The strategy uses the following Etfs: the S&P 500 (SPY), the Nasdaq 100 (QQQ), US real estate (VNQ), US mortgage real estate (REM), intermediate-term US Treasuries (IEF), long-term US Treasuries (TLT), US TIPS (TIP), Europe stocks (VGK), Japan stocks (EWJ), international small cap stocks (SCZ), emerging market stocks (EEM), international real estate (RWX (GLD).
Select the top six assets based on momentum, using the average of each asset’s three, six, and twelve-month returns.
Using this in a ranking system: “ROC(63)+ROC(126)+ROC(252)”
Choose three of the six, which are the least correlated.
The rule number three is what I have problems with. What can the rule (code) be to pick the three least correlated of the 6 ETFs with the highest momentum (rule2)?
I did some tests using “Correl(5,52,GetSeries(“SPY, QQQ, VNQ, REM, IEF, TLT, TIP, VGK, EWJ, SCZ, EEM, RWX, GLD, DBC, BWX”)”. But I couldn’t get it to work the way I wanted it to.
According to the original approach “It starts with its list of” top half “assets and asks: What’s the most robust portfolio of 3 assets one could make from this list? It defines” robust "as the 3 assets with the lowest average correlation to the other remaining assets in the portfolio (akin to Varadi’s Minimum Correlation algorithm). All other assets are discarded. " (https://breakingdownfinance.com/finance-topics/modern-portfolio-theory/minimum-correlation-portfolio/)
Is it possible to put the Viradis Correlation Algorithm to the test in this strategy in P123?
OK, so something’s a little confusing. If you’re doing the top half of the assets and you have fifteen assets, you want the top 7, not the top 6. Then you discard the other 8. And then you invest in the three with the lowest correlations to the 7 that are left.
This method calculates weights for max return, use inverse of weights for min return. The objective of the minimum correlation algorithm is to choose the portfolio weights such that the assets are weighted proportionally to their average correlation with the other assets in the portfolio. That way, we obtain a portfolio where assets that have lower correlation to all the other assets in the portfolio get a higher weight.
In the above there is a spreadsheet link for three ETFs which calculates the optimum weights for a combo of the three.
P123 could incorporate this into the Book simulation module, or an expanded version for more ETFs or strategies.
Inputs are all available from P123:
1.Correlation matrix from Simulated Book.
2. Standard Deviations for each ETF or strategy from Risk Measurements .
I just tested this and it actually does produce the highest returns if the weights are calculated accordingly.
I understand how this works but it seems to be missing one essential component: the expected return of each ETF. As I see it, the three essential variables in portfolio weighting are expected return, correlation, and volatility. If the expected returns are all equal, then this formula, which takes into account the other two variables, should work well. But if they’re even slightly unequal, does correlation make a lot of difference? I’m still trying to figure out the answer to that. If you have any insights along those lines or recommended reading, I’d love to hear.
I don't suppose anyone has made progress on developing this idea of using dual momentum and low average correlation?
The above examples are very interesting @yuvaltaylor but the outcome diverges a lot from the Allocate Smartly backtest, so I think there is some crucial feature missing. Did you find anything else in your research @geov ?
These momentum strategies tend to be timing sensitive. Most set the trading date to last trading day of the month, so does Allocate Smartly.
I have tried many of these momentum strategies you find online (on a different software), many completely fails by just moving the trading day a few days.