GTAA - The way to the optimal model (part 2)

Dear Quins,

following my little introduction to GTAA (Global Tactical Asset Allocation) (Link 1) there were some questions about how to create your own model. So I thought for a while about how to best explain this process in a generally understandable yet exciting way. Here is the result. Have fun!

1st step: GAA - Global Asset Allocation

2nd step: GTAA 1 - GAA with Dual Momentum

3rd step: GTAA 2 - GTAA with Asset Differentiation

4th step: GTAA 3 - GTAA with factor

5th step: GTAA 4 - Optimized GTAA

6th bonus: GTAA Max

7. conclusion

8. links

Step 1: GAA - Global Asset Allocation

i. Asset selection

The first question when building a Global Asset Allocation portfolio is: Which asset classes are there and which can be invested in via ETFs? To begin with, it is sufficient to simply look at the categories of an ETF search on a good ETF portal (link 1). This shows the usual suspects: Equities, Bonds, Commodities, Money Market, Real Estate, Crypto.

Here we look for a cheap ETF for each asset class, covering the entire class if possible. Crypto is left out (sorry, dear crypto fans, but the compensation follows at the end, we promise!). That is:

Stocks: all-world ETF;

Bonds: Aggregate Bond ETF;

Commodities: All Commodity Index ETF;

Real Estate: Global Reit ETF;

Money Market: Call Money

ii. Backtest

How did this world portfolio compare with B&H All-World-ETF over the last 20 years? For the backtest comparison we use the powerful tool "Portfoliovisualizer" (link 2).

Unfortunately, there are a few limitations: Tradable all-world ETFs have only been around for about 10 years, and an all-commodity ETF has also only been available in the tool since 2008. Accordingly, you have to find similar or like-minded assets for backtesting. In our case these are:

SPY (S&P500 for All-World-ETF)

LSGBX (Intl. Bonds for Aggregate Bond ETF)

^GOLD (Gold index for commodities);

VGSIX (REITs ETF for real estate).

If one weights the assets in GAA analogous to the ARERO world fund with 50-35-15-10 and compares the portfolio with both the S&P500 and the classic 50-30-20 world portfolio, then from 10000$ in the period 2001-2022 the following became (link 3 - please enter yourself for verification!)

Results (portfolio, final amount, annual return, maximum drawdown)

S&P500: $44,046 6.97% -50.80%

World portfolio: $39,215 6.41% -55.93%

GAA:
$47,394 7.33% -34.62%

Compared to the S&P500, the classic world portfolio is slightly worse in every way: less return with more risk. The GAA portfolio is slightly better in every way.

To think further: In the savings plan, the order reverses (feel free to try it out!). A different weighting of the assets sometimes also has a noticeable impact on performance.

Step 2: GTAA 1 - GAA with Dual Momentum

i. Strategy selection

After the asset allocation is in place, the question is which easy-to-follow, low-maintenance strategy could be used to reduce risk and increase opportunity. Two classic candidates: 200-day strategy and dual momentum.

The 200-day strategy has the following maximally simple rule: hold an asset if it is trading above the 200-day average at the turn of the month, otherwise hold cash.

Dual Momentum has the following two rules. 1: On the last day of the month, add up the respective 1,3,6,12 month returns for each asset and rank them in order. Rule 2: If the strongest asset is trading above the momentum of Cash, invest the entire capital in this asset. Otherwise hold cash.

ii. Backtest

Let's apply the two timing strategies, GD200 and Dual Momentum, to GAA and compare them with each other and with B&H S&P500 (link 4).

Results (Portfolio, Final Amount, Annual Return, Maximum Drawdown)

S&P500 $44,046 6.97% -50.80%

GAA $47,143 7.30% -33.60%

GD200 $44,293 6.92% -12.70%

GTAA1
$86,959 10.33% -29.06%

The GD200 model has about the same return as B&H S&P500, but with dramatically lowered risk (return sequence important for safe withdrawal rate in retirement!). The GAA Dual Momentum model has medium risk with significantly increased performance.

To think further, other averages (e.g. 100d) or momentum periods (e.g. 1.6 months) change performance, sometimes significantly. There are a number of other simple and complex models on a monthly basis (Gebert's stock market indicator).

Step 3: GTAA 2 - GTAA with asset differentiation

i. Expansion of the assets

Can returns be enhanced if Dual Momentum selects the strongest asset not only among different asset classes, but also within them among different markets? We assume the following classic differentiation of asset classes:

Equity market: US, EU, EM (SPY VEURX VEIEX)

Bonds: US Treasuries, World Bonds (VBMFX ESICX)

Real assets: Gold, Real Estate (VGSIX ^GOLD)

ii. Backtest

Let's now apply Dual Momentum to the new differentiated GAA.

Results (Portfolio, Final Amount, Annual Return, Maximum Drawdown)

S&P500 $44,046 6.97% -50.80%

GAA $47,143 7.30% -33.60%

GTAA1 $86,959 10.33% -29.06%

GTAA2 $102,498 11.16% -30.37%

Splitting asset classes into regions further increases returns with almost the same level of risk.

To think further, different time periods have different average performance. The GAA model shows its full strength especially in difficult market periods, e.g. 2001-2009.

Step 4: GTAA 3 - GTAA with factor

i. Optimization of the assets

Can perhaps the classic factor premiums (Growth, Small Cap) be used for further differentiation into uncorrelated asset classes so that Dual Momentum can show its strengths even better?

A lot of research is needed here. 1. not all factors are testable in the tool until 2001. 2. not all testable assets are tradable as ETF in Germany or at your own broker. 3. the offer changes constantly. So you have to compare, compromise, keep searching.

I found the following tradable factors that have little correlation:

US-Large-Growth: Nasdaq100 (QQQ)

EU-Small: EU Small Caps (PRIDX)

EM-Small: EM Small Caps (DEMSX)

US-Bonds: 30y Treasuries (VUSTX)

World Bonds: Global Bonds (LSGBX)

Real Assets: Gold (^GOLD), Real Estate (VGSIX)

ii. Backtest:

Results (portfolio, final amount, annual return, maximum drawdown).

S&P500 $44,046 6.97% -50.80%

GAA $47,143 7.30% -33.60%

GTAA1 $86,959 10.33% -29.06%

GTAA2 $102,498 11.16% -30.37%

GTAA3
$117,392 11.85% -29.16%

Performance continues to increase without an increase in risk. The model already comes close to a 12-fold increase in capital 2001-2022, but there is more to come!

To think further: the markets can be divided into further factors, e.g. value, momentum or combinations of these, e.g. EU Small Cap Value. Or into regions: Japan, Frontier Markets. There are ETFs for almost(!) everything.

Step 5: GTAA 4 - Optimized GTAA

i. Optimization of the model

Is it possible to further increase the performance and at the same time reduce the risk? Here again a lot of research work is necessary to find 1. comprehensible adjusting screws and 2. the optimal application. I have found the following optimizations so far.

To increase the return: 1. Focus the model on the best assets. Those assets that lower the overall return get kicked out. 2. not only focus on one top momentum asset, but also on two or three. The strongest combination stays.

To reduce risk: 1. Apply GD200 strategy. 2. spreading the risk on top 2 or 3. the lowest drawdown remains.

ii. Backtest

GTAA4/1: GTAA3 with GD200

GTAA4/2: GTAA3 without REITs with GD200 and Top2

Results (Portfolio, Final Amount, Annual Return, Maximum Drawdown)

S&P500: $44,046 6.97% -50.80%

GAA: $47,143 7.30% -33.60%

GTAA1 $86,959 10.33% -29.06%

GTAA2 $102,498 11.16% -30.37%

GTAA3 $117,392 11.85% -29.16%

GTAA4/1 $218,607 15.05% -18.96%

GTAA4/2 $260,531
15.97% -14.39%

The optimized factor GTAA shows a significant increase in return with a significant decrease in risk. Comparatively, the optimized GTAA model was able to increase initial capital 26-fold, compared to B&H S&P500's 4.4-fold increase (see chart).

To think further, what is the optimal GTAA model - with the highest return at the lowest risk? Can the values also be applied to the euro area? How can we estimate the probability whether the performance can be expected in the future?

6. bonus: GTAA Max

i. Top asset selection

Now the bonus for all return fans! How far can GTAA be used without increasing the risk too much? The following conditions apply: 1. everything that is tradable as ETFs in Germany is allowed. 2. all averages and momentum periods are allowed. 2. no borrowing.

The following ETFs are eligible:

ETF on 3x leveraged Nasdaq100.

ETF on bitcoin (yes, dear bitcoiners, nyknyc! But the above condition applies and you may also use your favorite exchange).

ii. Backtest:

Assets: TQQ PRIDX VUSTX ^BTC (time period 1/2015-3/2023).

Results (portfolio, final amount, annual return, maximum drawdown).

GTAA5 Top 2: $358,174 54.30% -32.36%

GTAA5 Top 3: $128,297 36.25% -19.79%

GTAA5 Top 4: $79,330 28.54% -14.60%

7. conclusion

Which is the optimal GTAA model, everyone must decide for himself: some want rather yield, the other rather security, the third a good combination. GTAA offers something for everyone! But finding the individually suitable GTAA model to implement with ETFs is not an easy task. GTAA is a process! Maybe now one or the other of you has got the desire to go on a GTAA exploration tour by yourself or together? Feel free to share your insights and thoughts!

PS.: Yes, I am implementing the GTAA strategy with my own money, because the logic behind it and the backtest results convince me. I am slowly building it using two models, one defensive and one offensive. Within the technical capabilities on GQ, I'm trying to map them both as best I can. Just stop by if you're interested (following and unfollowing always works).

8. links

Link 1: https://getqu.in/HTC7DCSKVZ8H/gj9PsZf4V2/

Link 2: ETF Search: https://de.extraetf.com/etf-search

Link 3: https://www.portfoliovisualizer.com/

Link 4: https://www.portfoliovisualizer.com/backtest-portfolio

Link 5: https://www.portfoliovisualizer.com/test-market-timing-model?timingModel=6