Portfolio Diversity in Action: A Live Case Study
While portfolio diversity is widely praised, it is rarely shown in action. We present a clear example from a portfolio we rebalanced today.
Most of us already know a manager can reach genuine diversity in roughly twenty to fifty names. We run twenty-two-security concentrated books and rebalance about every ninety days. After our most recent rebalance on March 24th, a single position fell roughly 45% over three trading days under a short-seller attack. We swapped in a replacement in the same economic sector after determining the position would likely not recover within our timeline.
As far as an opportunity to observe a real-time diversity benefit, we now had an almost perfect set-up to see what would happen next, because this portfolio had a stablemate that differed by three securities.
We watched as the attack did not affect the rest of the book, which matched our expectations based on the following design parameters: average pairwise correlation across the book below 0.30, approximately 0.25 at launch; and a principal component analysis where twelve components were needed to explain 80% of variance, exceeding our ten-component minimum. In theory, that structural independence should keep idiosyncratic failure idiosyncratic. We can observe this over the following fifty-nine trading days, where the rest of the portfolio, including our replacement stock, tracked its stablemate minus the short-seller loss and also beat its benchmark for the period. Let us look at the details for a finer analysis.
The sibling test
The bulk of the roughly two-point distance is the short-seller hit we could not erase over the period. The books differ by three names, but outside the affected name the damaged book tracked its sibling step for step over the period. That showed us the short-seller damage did not spread within the portfolio, and that the diversity held.
Against the benchmark
What pleased us even more was the book still beating its benchmark by a healthy margin. Over the full holding period, with the attack included rather than excluded, the damaged book finished more than four points ahead of the equal-weight S&P 500 (RSP), gross and net of our maximum 2% annual advisory fee prorated over the period (fees range from 1% to 2% annually).
We use RSP because we build equal-weight portfolios holding at least one stock from each of the eleven economic sectors. Measuring against a cap-weighted index like SPY in this regime mostly measures how much you held of the largest names, which says nothing about construction. RSP isolates the thing under test, which is whether the book was built well, not whether it rode the same handful of mega-caps as everyone else.
Why it held
Containment was design in action, with two parameters carrying the weight. Low average pairwise correlation, plus twelve components to reach eighty percent of variance against a ten-component floor, meant the book’s risk was spread across many independent sources instead of stacked on a few. When the attack landed, it loaded onto the affected name’s own idiosyncratic risk, a risk the other twenty-one names did not share.
One distinction is worth stating plainly: this kind of containment only applies to an idiosyncratic, single-name shock. A sectoral move like an oil shock would move several names at once unless the book had specifically diversified away from that factor. Independence across names is what contains a single failure. Independence across factors is what contains a sector. This was the former.
Conclusions
We rarely get to use a live, working book as this clean an illustration of construction principles in action, and we share it in that spirit.
Please treat this solely as an illustration of how independence is built into a portfolio, and not as a forecast of future results.