Stock & bond investors miss out on 70% of the diversification available.
One of the attractions of trading futures is that the universe of instruments is large and wide. Do you want to trade gold? There is an instrument for that. Oil? Sure. S&P 500? Korean government bonds? The Swiss franc? Oats? Yes, yes, yes, and yes!
The diversity of futures allows us to explore the relationships between asset classes, because they are all represented in the futures universe. We can easily calculate their correlations and examine them.
Correlations measure the similarity of asset returns. Two assets that tend to move together will have a high correlation. Assets that move completely independently of one another will have zero correlation. And assets that move opposite to each other will have a negative correlation.
We track the trading history of 286 instruments in our universe of futures contracts. There are 286 x 286 = 81,796 elements in the correlation matrix. Staring at a table with over 80,000 entries is unlikely to give us insights. We need to distill the correlation matrix into something more informative.
Clustering algorithms give us a distillation of the correlation matrix. The algorithm will split the universe of futures into a given number of clusters and assign contracts to the clusters so that within each cluster, the contracts are more similar to each other than to contracts outside of their cluster. The number of clusters is arbitrary: we can run the analysis with 2, 3, or even 100 clusters. But if we opt for two clusters, we’ll end up with two big undifferentiated blobs. If we choose 100 clusters, they will have two or three contracts each, which is not helpful either. A useful number of clusters is somewhere between half a dozen and a dozen. For the purpose of this article, I chose 10 clusters because I found the resulting groupings informative.
I used the procedure outlined in Rob Carver’s blog post from January 2022. Robert Carver is a former hedge fund manager and author of excellent books about investing and trend following. He concludes: “This just goes to show that the traditional grouping of asset classes may not make as much sense as you think.” Well, let’s see.
Cluster 1: Safe haven assets
Bonds, safe haven currencies (Swiss franc, Japanese yen), utility and real estate stocks
Cluster 2: Equities
US & European equities
Cluster 3: Global trade beneficiaries
Asian equities, metals (both base and precious metals), agriculture (soybeans, corn, cotton), Brazilian real, Mexican peso
Cluster 4: Wild & crazy
Crypto, coffee, sugar, European natural gas, palm oil, Japanese government bonds
Cluster 5: Oil
Crude oil & distillates, oil & gas equities
Cluster 6: Grains and edible oils
Agriculture (wheat, rice, oats, milk, cheese, soybean oil, canola)
Cluster 7: Breakfast foods
Agriculture (cocoa, orange juice, cattle)
Cluster 8: Currencies
All currencies except safe haven currencies (CHF and JPY), Brazilian real, and Mexican peso
Cluster 9: Volatility
Volatility indexes in the US (VIX) and Europe (VSTX)
Cluster 10: Natural gas
US natural gas
Several observations jump out:
- Out of ten clusters, most equities are only in two of them. The majority of US and European stocks are actually in the same cluster. It’s hard to find much diversification in stocks: they tend to move together. The exceptions are oil & gas stocks which move with oil prices, and real estate and utilities which move with interest rates.
- The distinction between developed & emerging equities is hard to see in the data. The real split is US and Europe versus Asia. Non-Asian emerging markets (including Latin America) cluster with US and European stocks.
- Bonds are all in a single cluster, with the exception of Japanese government bonds. Japan is in a unique fiscal situation that makes its bonds a potentially explosive asset (not in a good way).
- Stocks and bonds together are 3 clusters out of 10. The standard allocation to stocks and bonds misses out on 70% of the diversification available.
- Agriculture has the most diversity: it has two clusters of its own plus presence in two more clusters.
- Volatility is its own asset class.
- US natural gas behaves like nothing else.
What is the takeaway from this analysis?
For investors, the standard stocks & bond universe represents only a small portion of the available exposures. Benefits of diversification cannot be fully explored in just those two asset classes. When they both decline, as they did in 2022, the rest of the investable universe should come to the rescue. Especially in times of high inflation, as during the 1970s and 1980s, commodity investing protected investors who had the courage to include it in their portfolios.
Futures fund managers should think beyond the standard split of futures universe into stocks, bonds, currencies, and commodities. Agriculture should be a sector of its own, and it should have the largest risk allocation, given the diversity of the sector. Volatility is an important asset class too and deserves its sleeve. And finally, US natural gas is so unique that it should receive the highest risk allocation of any futures contract.
And a final technical note. Clustering is a statistical technique that relies on historical correlations. When correlations change, the clusters change. There is a fair amount of stability but assets do migrate between clusters from time to time. We looked at historical clusters too and the observations above remained valid in recent history.