Tactical Asset Allocation

Last week, the @WifeyAlpha account appeared on my timeline: for once, Twitter algo did the right thing. Not only did Twitter put the account in front of me, it chose one of the best Tweet I read in a long time:

Part 2: 15 tactical basic asset allocation models with expected max drawdowns of < -20% with Sharpe ratios > 0.75. Live trading may not represent historical.

What type of regime are we in and how will these models react? How will these tactical models help buy and hold models?

— Wifey (@WifeyAlpha) March 12, 2022

15 tactical asset allocation models? Is it Christmas yet?!? This is the stuff I love.

Why Tactical Asset Allocation Models

You are probably familiar with the 60/40 portfolio: 60% Tesla, 40% Bitcoin….ehm, 60% stocks, 40% bonds. Allocations to this portfolio are static and portfolio holdings are rebalanced with a set frequency (monthly, semi-annually and so on) to the target allocation.

In a Tactical AA model, allocations are not static but change according to a defined set of rules. Let’s take Wifey’s first model as an example.

Ivy Portfolio Timing The Ivy Portfolio is designed to mimic the investment strategies of Harvard and Yale endowments. The timing version uses a simple moving average. 20% equal weight in VTI VEU VNQ AGG DBC Ann Ret: 7.5%, Risk 6.5%, Sharpe ratio: 0.8, Max Drawdown: -13.1%. This strategy trades once per month. Rules below. A) Split the portfolio into 5 equal slices. B) On the last trading day of the month, calculate the 10-month moving average for each of the assets above. C) If the price on the last trading day of the month > 10-month moving average, allocate to that investment. If the price < 10 months moving average, allocate that portion of the portfolio to cash. D) Hold until the last trading day of the next month.

This portfolio can include six asset classes: US stocks, International stocks, real estate, bonds, commodities and cash. Instead of having a fixed allocation to each asset, like 16.6% of the portfolio, the allocation to each ETF is rule-based. We have five ‘risk on’ assets (US stocks, international stocks, real estate, bonds, commodities), while cash acts as a proxy of ‘risk off’ [we can debate if bonds should be considered as risk-on or risk-off, I am trying to simplify things here, bear with me]. The general idea is to invest in an asset when its trend is positive, as measured by its price against a moving average, and switch to cash when the trend is negative.

Compared to a static model, this strategy aims to lower the Maximum Drawdown amount by ‘timing the market’, sitting in cash when prices go down, instead of taking advantage of the inverse price correlation between the assets included in the portfolio. It is a simple way to buy-low / sell-high preying on the fact that asset prices trend, their moves up and down are not completely random if you consider a long enough period of observation.

Got problems?

Unfortunately, this type of strategy has its own issues, compared to a static one:

• False Signals: a false signal is when the model sells (buys) but the trend immediately resumes upward (downward). As per above graph, you can see that each sell signal is triggered after a market top, sometimes way after. In this events, the model sell high only to buy higher, leaving some gains on the table.
• Non-trending market: sometimes, markets do not have a clear trend but move sideways. During these periods, the model do the reverse of what it is supposed to do, it sells low and buys high. This is the worst possible situation, the reason why the model can still have high drawdowns. Static models are killed by big crashes while tactical models die by a thousands cuts, small losses that continue to pile on.
• Trading costs: tactical models trade more often than static ones. In the era of zero-cost brokers, this might not be relevant anymore but you still pay the bid-ask spread and possibly a divergence between price and NAV because…you are following the trend, selling when everyone else is selling, beeeeee:

Got solutions?

Depending on where you sit in the S&M spectrum, you might sustain better a thousand cuts than the kick in the nuts, 40% market crash, type of pain. If this helps you to follow the investment strategy, it is already a plus.

Instead of considering tactical models as an alternative to static ones, we should consider both strategies as complementary. This is where the real juice is. As I said before, their drawdowns happen in different forms and, more importantly, in different moments. In short, the strategies are uncorrelated when it matters the most, on the downside.

Composer

Should you blindly trust me?

I like PortfolioCharts a lot but unfortunately, their website can only help us to test Static Asset Allocation strategies. Luckily for us, last week Composer opened to the public. I wrote a quick post about it almost a year ago, it was a very futuristic (and brief!) way to look at custom indexing for retail investors.

Composer is the perfect ‘no-code’ app to build and test Tactical Asset Allocation strategies. Instead of trying to explain what Composer is, I’ll borrow the perfect teaser from the master himself, Packy-NotBoring:

{Composer is] a platform for creating and investing with rules-based trading strategies. It’s basically algo-trading without Python or Excel. They have dozens of pre-made investing templates, but what’s really cool is that you can edit those templates or build your own strategies from scratch, backtest them, and then invest. 

You can edit their “Buy the Dips Nasdaq” template to only buy the top 20 holdings if they were up over 30% last quarter and rebalance every two weeks. Or, build your own strategy to execute paired switching trades between QQQ, USO, and Bitcoin ETFs – no code required.

I use to test trading strategies in Excel, with the help of the Bloomberg add-on. To create a full test of the Ivy Portfolio it would have taken me at least half a day, while I had it tested in Composer in less than two hours…and I spent most of the time just to understand that you can select a back-test period only after you saved and closed-open again your Symphony (how a model is called in Composer). The beauty of Composer is that it has embedded some concepts like buy/sell and reports (average gain, max drawdown) that I would have to define by myself in Excel: these are the most time-consuming activities that Composer do for you if you are interested in its back-testing capabilities.

[I discovered it by watching Mr Rip videos, Google Sheets has some functionalities similar to the Bloomberg add-on…for free. Some of the strategies in Wifey thread cannot be tested in Composer, Google Sheets might be a great solution for the non-Bloomberg user]

In a past life, I use to test trading models in TradeStation and I wanted to mention it before someone commented “Composer does what TradeStation did 20 years ago”. It is true, at least as I remember it, but I think Composer interface is still a bit more intuitive while being less flexible. Composer is also free to use while I got my ‘free’ TradeStation because someone accidentally dropped a pack in front of my house with the free code in it.

The user can send, via Composer, buy/sell signals to a broker so that their trading model can be fully automatized. This feat. is available only to US-based investors so, for the purpose of this post, I will ignore it (yes, TradeStation might be able to do the same but I am not 100% sure because it could not when I was using it).

The non-code

Here is what the ‘symphony’, as they call it, looks like in Composer:

• this is only part of it because the whole code did not fit a single screen but I think you get the point, each ‘block’ is similar to the other and the only thing that changes is the ETF you invest when you are long
• you can set the Moving Average as # of days instead of months. If you assume 200 trading days in a year, 10 months is 166.6 days. I just rounded it.
• I do not think Composer accept ‘cash’ as an investment option, so I used the ETF BIL, which should be close enough (it invest in 1 to 3 months T-Bills)

If you want the full code just drop me a DM and I will send it to you.

The Back-test

I could only run the back-test for the period 08/07/08 – 18/03/22, most probably one of the ETFs I used was launched around that date. Here are the results according to Composer:

They are almost in line with what Wifey reports but not exactly there. I suspect the reason is the relatively short period of my back-test.

The Juice

What happens if I combine the Wifey strategy with a standard 60/40 portfolio (here SPY and AGG)? I called it ‘Wifey 2’ and it is a 50% Wifey + 50% 60/40, rebalance monthly.

As…expected (eheheheh) look at that Sharpe ratio. The two strategies combined have a Max Drawdown that is almost half of the 60/40, while maintaining above 7% Annualised Returns. [I might or might not notice that if you leverage that shit up to 2x you get double-digit returns for the same risk of a 60/40…probably nothing ;)]

Conclusion

One of the principles of Return Stacking/Portable Alpha is to invest your newly-freed capital into uncorrelated strategies (if you do not know what I am talking about, check my previous post about Capital Efficiency); trend models are a very good candidate and Composer offers you the ability to run one conveniently ‘in-house’ without having to rely on someone else ETF…if they will ever be available in Europe.

I also had the time to model and test only the first strategy in Wifey’s thread. If you consider all the static strategies that you can find on PortfolioCharts as well, the possible combinations are quite large. 2008 to 2022 has been a period during which risky assets have basically traded only in one direction: up to the right. It is hard to appreciate the benefits of a defensive strategy when it basically generated only false signals.

What I am reading now:

Follow me on Twitter @nprotasoni