In late 2008, many traders watched as their “diversified” portfolios imploded in real-time. Despite having positions across assets that historically moved independently, everything suddenly started falling in lockstep.
People would say things like, “I’ve spread my risk perfectly,” or “oil has nothing to do with corn prices. Gold trades on totally different fundamentals than natural gas.”
But as the global financial crisis unfolded, correlations across disparate markets tended towards 1. Every position bled red simultaneously. It wasn’t about fundamentals anymore—it was about liquidation. Funds were selling anything and everything to raise cash.
Correlation of major asset classes during the GFC. Source: TradingView
The big thing that many traders missed (and still miss) is that correlation isn’t static. It’s dynamic, and it tends to spike at exactly the worst possible moment.
Today, we’re going to dive into correlation in commodity markets: what it actually means, how to measure it properly, and most importantly, how to use it to build more resilient trading portfolios.
Correlation is one of the most misunderstood concepts in portfolio construction. I’ve seen professionals get this wrong, so don’t feel bad if it’s fuzzy for you, too.
In simple terms, correlation measures the tendency of two assets to move together. The scale runs from -1 (perfect negative correlation) to +1 (perfect positive correlation), with 0 indicating no relationship.
But correlation only tells you about the direction of movement, not the magnitude. This is crucial to understand.
For example, if crude oil and gold have a correlation of 0.3, it means they tend to move in the same direction more often than not. But it tells you nothing about how large those moves are relative to each other.
Gold might rise 1% while oil jumps 5%. They’re correlated in direction, but not in magnitude.
And this is where you can go wrong – equating a positive correlation with risk similarity, which is only partially true.
Here’s what a correlation matrix of major commodity futures might look like:
Asset | Crude Oil | Gold | Natural Gas | Wheat |
---|---|---|---|---|
Crude Oil | 1.0 | 0.17 | 0.21 | 0.16 |
Gold | 0.17 | 1.0 | 0.05 | 0.13 |
Natural Gas | 0.21 | 0.05 | 1.0 | 0.07 |
Wheat | 0.16 | 0.13 | 0.07 | 1.0 |
Pretty low correlations across the board, right? That looks like great diversification. But these are long-term averages—they tell you nothing about what happens during stressed markets. And that’s exactly when you need diversification the most.
The fundamental principle is simple: combining assets that don’t move in perfect unison reduces overall portfolio volatility.
Let me illustrate with a straightforward example. Imagine you have two commodity futures with the following properties:
If you put 50% of your capital in each, your portfolio’s expected return remains 10%, but your volatility drops to about 17%, resulting in a substantial reduction in risk without sacrificing returns.
If the correlation were zero, your volatility would drop even further, to about 14%. That’s the diversification effect in action.
The math behind this is the portfolio variance formula:
Portfolio Variance = w₁²σ₁² + w₂²σ₂² + 2w₁w₂σ₁σ₂ρ₁₂
Where:
You don’t need to memorize this formula, but understanding it conceptually is valuable. That last term with the correlation coefficient is what gives us the diversification benefit.
What’s interesting is that the diversification benefit shows strong diminishing returns. Going from one asset to two uncorrelated assets gives you a massive reduction in portfolio volatility. Going from 20 to 21 assets? Barely noticeable.
This is why you see less bang for your buck in terms of diversification the more you add to your portfolio.
Many traders rely on simple 30 or 60-day rolling correlation calculations. And that’s fine, but there are other approaches too, and each has its own pros and cons.
Here we’ll explore three useful approaches.
This is what most traders use. You take the last N periods of returns for two assets, calculate the Pearson correlation coefficient, and that’s your estimate.
If you actually want to run the numbers, the Python code is trivially simple:
# In pandas
df[‘rolling_corr’] = df[‘asset1_returns’].rolling(window=60).corr(df[‘asset2_returns’])
Pros: | Cons: |
|
|
Here’s the rolling 12-month correlation of gold and oil returns:
Notice how those correlations, measured on a 12-month rolling basis, move around all over the place, often deviating quite far from the static value (0.17) in the previous correlation matrix.
This approach gives more weight to recent observations and less weight to older ones. It responds much faster to changes in correlation regimes.
Once again, you can run the numbers using a simple snippet of code:
# Using pandas
span = 252 # Approximately equivalent to a 252-day simple window
ewm1 = df[‘asset1_returns’].ewm(span=span).std()
ewm2 = df[‘asset2_returns’].ewm(span=span).std()
ewmcov = df[‘asset1_returns’].ewm(span=span).cov(df[‘asset2_returns’])
df[‘ewm_corr’] = ewmcov / (ewm1 * ewm2)
Pros: | Cons: |
|
|
Here’s how the exponentially weighted correlation of gold and oil compares to the rolling window approach:
Instead of looking at all market conditions, you can specifically examine correlations during periods of market stress.
# Simple example: look at correlation when VIX is above 30
stress_periods = df[df[‘vix’] > 30]
stress_corr = stress_periods[‘asset1_returns’].corr(stress_periods[‘asset2_returns’])
Pros: | Cons: |
|
|
Here’s what that might look like:
The million-dollar question: Does past correlation tell you anything useful about future correlation?
The answer is yes, but less than you probably hope.
Correlation does show persistence. That is, the correlation between assets today is somewhat predictive of their correlation tomorrow. But it’s incredibly noisy, and regime changes can happen overnight with no warning.
Here’s the 30-day correlation of gold and old plotted against the next 30-day correlation, with overlapping data removed:
You can see that there’s weak evidence of persistence: the correlation today is indeed related to the correlation in 30 days time, but it’s a very noisy effect with a ton of variance. Look at how spread out those points are.
In practical terms, this means that yes you can use correlation estimated from past data to assist with risk management in the future – but don’t expect it to work out consistently. That’s OK – this level of variance is a feature of most predictive relationships in asset returns.
Theory is great, but let’s get practical. Here are some approaches you can implement today to better manage correlation risk in your commodity trading:
This approach allocates capital so that each position contributes an equal amount of risk to the portfolio, accounting for correlations between positions.
The math gets hairy quickly, but the concept is simple:
You don’t need to implement the full optimization algorithm. A simplified heuristic version works well enough:
This approach naturally reduces exposure to assets that become more correlated to the rest of your portfolio.
If you find ERC too complex, here’s an even simpler heuristic:
For example, you might use a rule like:
This creates an automatic “correlation brake” that reduces overall exposure when diversification benefits start to disappear.
If you’re already using volatility targeting (and you should be), you can incorporate correlation as a secondary adjustment:
This approach gives you the best of both worlds. You’re primarily sizing based on volatility (which is more stable and predictable than correlation), but you’re still adjusting for correlation effects.
The important takeaway is that, given the level of variance in our correlation predictions (see the scatter plot above), the method you choose is less important than actually doing it at all.
Here are a few practical tips:
Daily recalculation of correlations is overkill for the frequencies at which most traders will be trading. Weekly is more than sufficient, with one major exception: during market stress, you should update your views more frequently.
Monitoring the rate of change of correlation can be more valuable than the absolute level. When correlations are increasing rapidly across your portfolio, that’s can be a warning sign of a regime change.
Keep an eye out for these red flags that correlations might be about to break down:
When you see these warning signs, it’s often wise to reduce overall exposure first and ask questions later.
You don’t need fancy software to manage correlation effectively:
By far the biggest mistakes people make with this are overconfidence in correlation estimates (they’re noisy!) and treating correlation as a precise measurement rather than a rough, dynamic guide. Precision is an unrealistic objective.
When it comes to using correlation for risk management, the right approach sits somewhere between “ignore it completely” and “optimize everything to the fourth decimal place.”
Be careful not to fall into one of these camps:
The middle path recognizes correlation as an important but imperfect tool:
Remember that the goal of risk management isn’t to maximize returns—it’s first and foremost to ensure survival, and secondarily to ensure that no single component dominates your returns.
If you want to improve your correlation-aware risk management, here are some concrete actions to take:
Start simple, then add complexity only as needed. Even a basic awareness of correlation dynamics puts you ahead of many commodity traders who still believe diversification is simply about trading different markets.
The commodity markets will always throw correlation surprises at you. It’s not about predicting these perfectly, but about having systems in place to detect and respond to them as best you can. No system will get you out perfectly unscathed, but doing something is better than doing nothing.
Kris Longmore is the founder of Robot Wealth, where he trades his own book and teaches traders to think like quants without drowning in jargon. With a background in proprietary trading, data science, engineering and earth science, he blends analytical skill with real-world trading pragmatism. When he’s not researching edges, tinkering with his systems, or helping traders build their skills, you’ll find him on the mats, in the garden, or at the beach.