Ripping up the old asset class labels

Asset managers, it seems, are no longer convinced the asset class labels that markets have used for decades make much sense. Research by Lazard Asset Management casts doubt on whether high-yield bonds, for instance, reliably move in tandem under similar circumstances.

So Lazard and others have been trying out AI algorithms to balance portfolios using machine-identified clusters instead. The idea is to look at the way securities behave, rather than their sector, geography, issuer or credit rating.

Allowing a computer to learn from scratch how instruments should be grouped, though, means that as many as a third are reclassified. That’s likely to alarm buy-siders, who rely on expertly mixing asset classes to maximise diversification.

Jai Jacob, who leads Lazard’s multi-asset investment team, says portfolio allocators make an “important assumption” when using conventional asset classes and their correlations to guide the rebalancing of holdings. Instead, quants could take a lead from other areas of data science and look at the data to define asset groups, he thinks.

“We’ve ended up having to live with a set of basic assumptions about asset categories for decades,” Jacob says. “It’s in every investment policy statement. It’s in lists of multiple choice questions. It’s all over the place.” The categories were only ever intended as simple ways to cover for a lack of data, though.

The use of clustering algorithms to splinter these historic asset blocks and form new groups of instruments that do behave alike would help asset managers reduce churn in portfolios when holdings fail to offset each other as expected. The new groups might also protect against big correlation shifts, such as a switchback in correlation between equities and bonds.

Moving away from static asset class definitions could also help in constructing more efficient portfolios.

Historically an investor who thought the US dollar would strengthen might invest in US small caps, which are less exposed to a rising US dollar, Jacob explains.

“That’s a logical view, and it’s commonly held. But it chains two assumptions together: first, that the US dollar is strengthening, and second, that US small cap is the right way to play it.”

Instead, by using data science investors could “on the fly” isolate a new asset class representing specific stocks that benefit from the US dollar strengthening. “That is much more powerful,” he says.

In June 2018, Bank of America launched a risk-parity strategy that employs much the same concept: balancing between asset groups as defined by a machine learning algorithm rather than between largely bonds and equities as traditional risk parity does.

For BofA, it’s a way to address the yield drag that comes with holding bonds by investing instead in other instruments that perform the same role – such as defensive stocks, perhaps.

Will the idea catch on? That’s unclear.

Pension funds, consultancies and index builders still organise their research around conventional asset class buckets and seem unlikely to change. Investment policy statements use these familiar terms.

But for firms like Lazard, the time may be coming when labels like high yield are erased completely from the playbook.