Correlation Research Suggests Diversification Fails

There’s an old saying in investing that a “rising tide lifts all boats.” While that might be true, when the tide goes out, all the boats sink and fall just as fast as when they rose together. Still, you’ll hear the pounding of drums in the name of classic diversification as the portfolio savior during these downturns. Our research suggests that even a well-diversified portfolio can still completely blow up just when you need the protection the most.

Diversification

People believe classic diversification is a portfolio-saver during downturns. However, our research suggests that even a well-diversified portfolio can still blow up when you need it most during major market downturns and black swan events.
Today, we are walking through some research we have done on correlations of pairs and asset classes, and the rolling correlations we see year by year to prove this.

What are correlations?

Correlations are a way to measure the relationship and the activity in price between two different assets, asset classes, industry sectors, or, in our case, tickers.
An asset pair (e.g., bonds and stocks) are either positively correlated or negatively correlated or have no correlation.
We measure this by tracking what they do, using math to come up with a correlation coefficient (how closely do these things mimic/mirror each other or not through time).
When you break correlations down into slices of time, tickers, or asset classes, they shift, ebb, and flow, like the sea! (things that weren’t correlated become so and vice versa)
Correlation coefficients describe a range between positive 1 and negative 1 with 0 being no correlation. Positive 1 means they will go in the same direction, negative 1 in opposite directions, and 0, no correlation.

The Aim of Diversification

In most diversification, the aim is to set up correlations where some things ebb while others flow at any time.
This might not be the best aim because if some assets go up and others down, you might still sit on a net-zero gain.
A better goal is for assets to have as little ties with each other as possible. So, if one goes up, one might just stagnate, resulting in a better gain than the net-zero example.
A good portfolio contains tickers with as much uncorrelated risk amongst each other as possible.
More tickers/assets don’t necessarily equal more diversification.

Diversification Example

If I had three tickers (Twitter, Facebook, Snapchat) in my portfolio, under the traditional diversification definition, I am more diverse than if I only had one.
The problem with these three is they are all social media positions, meaning they are highly correlated. If one goes down, the others might too.
The same goes for any basket of stocks containing tickers from a similar industry.
The aim is to build a portfolio that’s as uncorrelated to the market as possible. Thus daily market movement has as little impact on our potential trading strategy as possible (unless we want it to).

Introducing Our Correlation Research Approach

We researched a period from 1999 to 2019, covering two black swan events and many other bear markets.
We tracked a number of relationships to look for common threads we could find from correlation: the correlation of individual tickers to all other tickers every year, the overall correlation of a ticker to other individual tickers, the pairs of tickers each year looking for the three best diversification relationships each year, and the most uncorrelated assets to a particular ticker.

Observation 1: A common thread in years where volatility was low and years it was high

When markets are calm/volatility is low (S&P above the 200-day moving average, trending or not), we found correlations were very loose; things were mostly uncorrelated.
E.g., in 2007, things were uncorrelated, which seems not to make sense because that was just before the crash, so if everything was rising, why are correlations so loose? This is because when things are calm, you tend to see asset classes make runaway moves.

In 2008, the market was radically different, it was a highly volatile year. Correlations snapped magnetically to become positive 1, except bonds.

The same thing happened in 2002 and 2020.
During market crashes, diversification goes out of the window, everything starts trading together. Having positions in silver, gold, bonds, emerging markets, homebuilders, retail, and utilities made no difference.

Observation 2: During this 1999-2019 period, there was a good mix of tickers, industries, and sectors that were uncorrelated to each other

We made portfolios containing up to fifteen positions and found little correlation between them and the global market.
In years where the market was crashing vs. being high flying, the makeup of portfolios between high and low volatility years are mostly the same.
In 2008, the best set of tickers was a portfolio containing FXE, FXI, GLD, SLV, TLT, USO, XBI, XHB, XLF, and XLV.
If you overlay this set over the best portfolio of 2019, there is a lot of overlap. i.e., 2019’s best set was EWW, EWZ, FXE, SLV, USO, XBI, XLB, XLF, XLV, and XRT.
Thus, in all of these markets, there is probably a core set of tickers, sectors, industries, and asset classes you should be trading, whether in a high or low volatility environment.
FXE (Euro), FXI (China), GLD (Gold), SLV (Silver), TLT (Bonds), USO (Oil), XBI, QQQ (Tech), XHB (Homebuilders), XLF (Financials), XLV (Healthcare), and XRT (Retail) – a broad index of markets you should be trading consistently over time, not something all in one market.
Yes, this is a diverse portfolio, but it’s still highly correlated to the market during a crash. The difference is that the positions in this portfolio were as uncorrelated to each other as they could be in that environment.

Observation 3: The variance in correlations from 1999-2019

If a correlation pair, e.g., the Dow with gold, didn’t change, our map shows it as zero.
A lot of people only look at correlations over a short period of time.
Doing this variance analysis allowed us to see which pairs had the most variance.
If they had a high variance, trading them as a pair is not a good idea.
Nasdaq and S&P were zero (.009), meaning they traded in lockstep. Their tracking had little variance. Similarly, GLD and SLV had a variance of (.04).
Conversely, other things had a wide disparity. E.g., XLU (Euro), and FXE (utilities) had at some points a .41 differential. Similarly, OIH (Oil services) and EWW (Mexico ETF) had a .416 variance at some points. This would be a terrible pair to trade.

Observation 4: Each ticker’s correlation through time.

Some tickers increased in relation to the basket. They tracked more each year, becoming more correlated. Other securities had the opposite effect.
The correlation of bonds over this period is one example. In the first half, bonds are uncorrelated. In 2001, 3, 5, 7, 8, 9, 10, 11, 12, 13, bonds were relatively uncorrelated compared to the peer basket.

After that, a shift occurs, and bonds become more correlated. This could be through government intervention.
Classically, bonds are negatively correlated, but in this observation, their correlation becomes positive.
In the future, you might get a different reaction from bonds in a meltdown.
Conversely, before 2010, XLE (Energy ETF) was highly correlated, becoming less so after that.

Takeaways

We know correlations are important.
They shift from year to year, especially during high vs. low implied volatility markets.
These correlations are going through cyclical/systemic transitional shifts.
Paying attention to this can reframe abnormalities as less so.

Wrapping Up: What do we do with this info?

In black swan events, correlations that were loose become high almost instantly. It happens and will happen again. Markets continue to be highly correlated during a crash, high volatility environments. Recognizing this means you won’t be caught off guard.
Create flexible strategies that can be adjusted. While trading uncorrelated tickers is great, we have to be able to switch to correlated ones quickly. You can’t do much with long stock during a crash. Adjust by rolling contracts and moving positions out as the market changes.
You have to build in asymmetric positions that profit from an increase in volatility and the correlation of everything else. These could be long wings on positions or XPS puts or long volatility (VIX) positions.
Cash provides lots of options. In a situation where correlations go loose, be careful, and have cash on hand.

The genesis of the idea of building out correlated portfolios is the question: ‘If everything starts to become like the S&P, starts to act like the broad market during these high volatility environments, then why do we trade this diverse set of tickers during low volatility? What benefit is there because if it has no benefit during black swan events, why do it?’ Answer: Even in low volatility environments, individual tickers, industries, and sectors will go through their own mini black swan events, requiring the necessity to have a diverse, uncorrelated portfolio.

Option Trader Q&A w/ Daniel

Trader Q&A is our favorite segment of the show because we get to hear from one of our community members and help answer their questions live on the air. Today’s question comes from Daniel:

Hi Kirk. Question regarding Beta weighting: How are we seeking Alpha while also seeking Beta? If we’re trying to stay Beta weighted to an index for example, how are we trying to beat the index? Is it just from the Theta decay? Is that the main thing? I’m just not sure on the complete reasoning behind it. I hope that you can answer that question. Thank you.

Remember, if you’d like to get your question answered here on the podcast or LIVE on Facebook & Periscope, head over to OptionAlpha.com/ASK and click the big red record button in the middle of the screen and leave me a private voicemail. There’s no software to download or install and it’s incredibly easy.