[Research #5]: Beating Bridgewater? How to Build Your Own All-Weather ETF Portfolio
Build A Defensive Portfolio Utilizing Asset Correlations
Recently Bridgewater launched its All Weather strategy ETF (ALLW) . Apparently, Bridgewater’s strategy has earned legendary status for a reason. Designed to handle everything the economy throws at it — booms, busts, inflation spikes, deflation scares — it’s the poster child of risk-balanced investing. But here’s the real question: can regular investors, using simple ETFs, build something just as good, or maybe even better?
I think the answer is yes.
In this post, we’ll:
Break down the core idea behind All Weather investing
Recreate it using ETFs
Apply some smart tweaks to make it more resilient
Test it against Bridgewater’s own ETF during drawdowns
🌦 What’s the All-Weather Idea, Anyway?
The All Weather portfolio was the brainchild of Ray Dalio and his team at Bridgewater. The idea is simple on paper: combine asset classes that perform well across four key environments — rising growth, falling growth, rising inflation, and falling inflation.
In theory, when stocks stumble, bonds pick up the slack; when inflation heats up, gold and commodities come to the rescue. Bridgewater’s original All Weather portfolio has delivered steady, moderate returns (around 7–9% a year historically) with much lower drawdowns compared to a stock-heavy portfolio.
Rising Growth → Stocks
Falling Growth → Bonds
Rising Inflation → Commodities, Gold
Falling Inflation → Long-term Treasuries
The classic mix looks like this:
30% U.S. stocks → growth engine
40% long-term Treasuries → deflation hedge
15% intermediate Treasuries → stability
7.5% gold → inflation hedge
7.5% commodities → broader inflation play
A Popular Version Built With ETFs
Here’s a popular DIY version using ETFs as a proxy for Ray Dalio’s portfolio:
30% U.S. stocks → VTI or SPY
40% long-term Treasuries → TLT
15% intermediate Treasuries → IEF
7.5% gold → IAU or GLD
7.5% broad commodities → PDBC or DBC
For a typical brokerage account, this is straightforward to set up and cheap to maintain.
Here’s Where We Can Improve
Now, let’s be honest: Bridgewater built this strategy for massive institutional money. Retail investors can afford to be a bit more nimble.
Remember the goal is to design a portfolio that has relatively low drawdowns, which means its volatility should not be large. Once you following this train of thought, two key factors could help improve the design of the all-weather portfolio:
Select ETFs with relatively low volatilities
Pick a group of ETFs that have low correlations to each other.
In the following, we will run some analysis to attack both, using historical prices of the ETF universe. Before we do that, we should be fair and separate the ETF price history into two parts:
Training period part: price data prior to 2020. We also require ETFs to have data with inception day at least since 2007, so that we can find out how the ETFs perform during the financial crisis.
Out-of-sample period: price data since 2020.
We use 2020 as the separation point since the market during 2020 serves as a good out-of-sample test for all-weather strategies.
Step 1: Screen for Low-Vol and Resilient ETFs
The data I have in hand have around a little over 1000 ETFs, hence the final result could change as the amount of data changes, however the methodology would be still valid. Once we require the inception dates of ETFs to be at least 2007, then the number of ETFs goes down. Following is the Annualized Return and Volatility for these candidate ETFs.
Obviously some of them have quite large volatilities, and I will filter out those that have high vols. Some of them also have large maximum drawdowns, which I will put in as filter too:
Volatility (between 2007-2019): < 20%
Maximum Drawdown (between 2007-2019): no more than -45%
I also require the realized return per unit of volatility to be at least 0.4 (similar to Sharpe Ratio except without the risk-free rate here).
With these filtering, The number of ETFs goes down significantly:
Step 2: Find A Cluster of ETFs That Have Low Correlations
The correlation of the candidate ETFs from Step 1 looks like the following
Warm colors correspond to high correlations, and cold colors (while or blue) correspond to low correlations. We would like to select 5 ETFs from the grid above and hopefully their pair-wise correlations are all low, i.e., they are all cold to each other.
Correlation Clustering Method:
There are many methods to find clusters from a group of objects, one of them is called “Hierarchical clustering”. Without going into details, I will post the code below
dist_matrix = correlation_matrix
from sklearn.cluster import AgglomerativeClustering
from scipy.cluster.hierarchy import linkage, fcluster
# Hierarchical clustering
linkage_matrix = linkage(dist_matrix, method='complete')
n_clusters = 3
clusters = fcluster(linkage_matrix, n_clusters, criterion='maxclust')
My original idea was to find three cluster from the correlation matrix:
high-correlation cluster
medium-correlation cluster
low-correlation cluster
The results I have ended up delivering the following three clusters:
Cluster 1:
Cluster 2:
Cluster 3:
However the results aren’t very satisfying as my goal is to find 5 ETFs which are “cold” to each other, and any of the three clusters above doesn’t directly deliver what I want.
As a result, I decided to use brute-force and use combinatorics to choose 5 ETFs out of the candidates, and then calculate the average correlation for each choice, then pick the one group that has the lowest average correlation. This adds some computation time but works well, and the algorithm picked the following five ETFs:
RTH: VanEck Retail ETF
IDU: iShares U.S. Utilities ETF
TLT: iShares 20+ Year Treasury Bond ETF
GLD: SPDR Gold Trust ETF
DBC: Invesco DB Commodity Tracking ETF
Interestingly, among the five, TLT, GLD, and DBC overlaps with the popular strategy version. Now let’s look at their pair-wise correlations:
Yes they are quite “cold” to each other. Their return and volatility characteristics (during Training period) are below:
Their volatilities are near or below 17.5%
DBC actually had negative returns over the period
Performance of All-Weather Portfolios
We will use the 5 ETFs selected above and build a All-Weather portfolio. In stead of hand-picking the portfolio weight for each ETF, I decided to use optimization to figure out the weights which minimizes portfolio volatility. To be a bit more realistic, I added the following constraints during optimization:
I put a upper limit of 7.5% on both GLD and DBC as it makes sense not to put too much Gold or Commodities in your portfolio
The optimization delivers the following portfolio weights, and I name this portfolio “Correlation-Based”:
TLT: 39.7%
RTH: 27.42%
IDU: 17.89%
GLD: 7.5%
DBC: 7.5%
To make it a bit more fun, I will compare the portfolio above with the popular version, which I will name “Ray Dalio Proxy”.
To examine the performances, I compare their portfolio metrics of two periods: 1. Training period, and 2. Out-of-sample period.
A few observations:
“Correlation-Based” portfolio delivers better returns during both periods
“Correlation-Based” portfolio also has slightly higher volatility than the popular version, and it has slightly drawdown during the training period -14.82% vs -13.61%
Another measure I often look at is the market participation rates. As can be seen from the chart below, “Correlation-Based” version has slightly higher participation rates in both up and down markets.
Real-Time Stress Test: 2025 Drawdown
ALLW was launched in March 2025, hence it would be interesting to compare it with the two portfolios above during the drawdown period in early 2025. (Note since ALLW only starts in March 2025, some of the performances below for ALLW doesn’t cover days prior to that).
ALLW had bigger drawdowns than the other two ETF portfolios, though all of them had much less drawdowns compared to the market (SPY as benchmark)
ALLW also had better recover later on
Final Thoughts
You don’t need to settle for whatever the market throws your way. With a simple all-weather ETF portfolio, you can build something resilient, cost-efficient, and surprisingly powerful.