There have been numerous articles and features in the financial media over the last several weeks noting the narrowness of leadership in the US equity market since the start of 2023. Some articles cite the outperformance of the cap-weighted US indices vs. their equal-weighted counterparts; others cite the outsized contribution of the largest seven or eight stocks to the overall index performance.
Whichever method is used to illustrate this phenomenon, the conclusions are the same: this level of concentration in the US equity market is unprecedented, and could be a harbinger of future market distress despite the otherwise healthy index returns year-to-date. In this article, we will take a closer look at the history of concentration in the US equity market and use Axioma risk models to try and understand the drivers and potential consequences.
A capitalization-weighted portfolio like the Axioma US Top 500 is generally considered to be broadly diversified over 500 constituents that are distributed over different sectors and sub-sectors. It has become common knowledge that the Information Technology sector and its component industries dominate the US market, with the total allocation to these companies hovering between 20% and 30% over the last decade or so. Companies such as Apple, Alphabet, Amazon, Microsoft, Meta (formerly Facebook), and now NVIDIA and Tesla (not an IT stock exactly), have soared in market value to the point where these seven companies alone constitute more than 27% of the portfolio:
Figure 1: Axioma US Top 500 top 7 constituents
That may not seem so bad from a pure portfolio construction perspective, but considering this portfolio is a passive representation of almost all the publicly traded corporate value in the United States (USD 37.4 trillion as of May 31), having USD 10.1 trillion in just seven entities seems a bit skewed — they certainly don’t represent 27% of all the revenue or earnings of the US corporate sector.
Inverse Herfindahl index: A measure of concentration
To think about this level of concentration from a long-term historical perspective, we borrow a measure from the corporate finance world — the Herfindahl index of industry concentration — to calculate the effective number of companies in the portfolio over time. The Herfindahl index is simply the squared sum of the market shares of companies in an industry. The larger the number, the higher the concentration. When we take the inverse of that sum, we get an effective number of companies competing in the industry. For example, if we have five companies in an industry but one of them has a 60% market share and the other four 10% each, then the effective number of companies is 2 ½ — not very competitive. Using this measure for a portfolio illustrates how skewed the allocation to the constituents can be. An equal-weighted index has an Inverse Herfindahl equal to the number of names in the index. For the Axioma US Top 500, the effective number of companies since December 1989 looks like this:
Figure 2: Axioma US Top 500: Inverse Herfindahl index, Dec. 1989 – May 2023
When viewed from this long horizon, it is clear that there has been a major change in the US stock market since around the second quarter of 2016. We seem to be in a multi-year bubble, where the effective number of companies has dropped and has yet to revert to the long-term mean. The long-term average effective number of companies is 130, including the last seven years. 130 effective names still seems reasonably diversified, at least relative to what we have now. The effective number fell below that average in April 2018 and has not crossed back over since, continuing to trend lower, even as more and more money flows into the market. May 31, 2023 was an all-time low of just 61 effective companies in the Axioma US Top 500. The previous low outside of the post-COVID-19 period, at the zenith of the dot-com bubble in March 2000, was 90.
Returning to more recent history, the last peak in the number of effective companies was in April of 2016, and the allocation to the top seven looked like this:
Figure 3: Top 7 constituents, Apr. 29, 2016
Most of the value in the ensuing years has accrued to Apple and Microsoft alone, with both companies achieving market capitalizations and index allocations that any other company in US history has never approached.
Since the onset of the COVID-19 pandemic, we saw the effective number of companies drop to an all-time low of 67 in the summer of 2020 as the mega-cap tech companies were seen as the big winners in the lockdown economy. It then reverted to 87 in May 2021 as the market gains became a bit more widespread, then fell again to a low of 69 in December 2021 at the peak of the “reopening trade.” With the onset of the inflation-driven sell-off throughout 2022, the effective number rose once again to a post-pandemic high of 86 as a great deal of wealth in Meta, Tesla and Alphabet was lost, only to fall back to an all-time low in the present, with a new superstar of concentration and investor enthusiasm joining the pack as shares of NVIDIA more than doubled since the start of 2023 and briefly joined the trillion-dollar club at the end of May.
Equal-weight vs. cap-weight
The idea of equal-weighting the constituents of a market index or portfolio is not new. When so-called smart beta was introduced in the 2000s, this was one of the alternative weighting schemes that was marketed as a way to get better performance from a diversified equity portfolio than just holding the market’s weight. Beyond the practical considerations of higher rebalancing costs and lower liquidity, holding 500 equal-weighted stocks should provide better diversification, but not necessarily lower risk as the equal-weighting scheme gives greater weight to the more volatile smaller-capitalization companies in the index. The expectation is that an equal-weighted version of an index will outperform the cap-weighted version because of the greater weight given to smaller stocks. It is an effective way to capture the small-cap risk premium in a straightforward manner.
Looking at the performance of our Axioma USA Top 500 market portfolio vs. an equal-weighted alternative of the same 500 US stocks validates this premise in the long run:
Figure 4: Axioma USA Top 500 equal-weight vs. cap-weight, Jan 1990 – May 2023
Over the last 33+ years, the equal-weighted portfolio has outperformed the cap-weighted one by about 71 basis points a year (not including rebalancing costs), which is a significant return premium that appears to justify the slightly higher risk one must incur to achieve it.
When this relative return history is analyzed via the Axioma US Equity Factor Risk Model – Medium Horizon, we can see just how much of this return premium is due to the pure Size factor:
Figure 5: Cumulative returns and active size factor contribution, Jan. 1990 – May 2023
Figure 6: Cumulative returns and active size factor contribution, Jan. 1990 – May 2023
Figure 7: Returns analysis
The equal-weight portfolio maintains a significant and consistent negative active exposure to the Size factor, meaning that its weighted average market cap is significantly smaller than that of the cap-weighted portfolio. The pure Size factor has an average annual return of around -5% since 1990, which we can interpret as meaning large caps tend to underperform small caps. Over the full period, the negative Size exposure is the only factor with a meaningful positive contribution to active return, as most other factors and the specific return contribution are negative.
Figure 8: Return decomposition
In fact, while the T-statistic for the total active return does not meet the threshold of statistical significance at the 95% confidence level (T-stat would need to be 2 or greater to reject the null hypothesis that the true active return is 0), the T-stat for the Size contribution is over 4, which infers that it is really the Size factor that has driven the outperformance of the equal-weighted portfolio through time.
Figure 9: Axioma USA Top 500 equal-weight vs. cap-weight, Jan. 1990 – May 2023
Year-to-date through May, the equal-weight portfolio trails the cap-weighted one by 838 basis points. This exceeds any annual active underperformance in the last 33 years except for 1998, right smack in the middle of what is considered the biggest bubble in US equity market history. From 1994 to 1999 the equal-weight portfolio underperformed as the winners kept on winning in the go-go ‘90s. It goes without saying that if the current trend towards greater and greater concentration continues, the 1998 record of -10.15% underperformance will be broken, possibly shattered. While the Axioma Top 500 is up 10% year to date, without the ‘big 7,’ it is up just 1.7%.
Canary in a coal mine? Ultra-short term trading model
Our US Equity Factor Risk Model: Trading Horizon suite is a pair of models tuned to investment horizons of one month or less by putting much greater weight on the most recent trading history (especially the last 20 days) to forecast future volatility. The suite includes a fundamental factor model and a statistical model with very little a priori structure. Statistical models have greater flexibility in recognizing temporary trends in market data owing to this lack of structure. They are free to pick up on any factors that may be present in the data history, while fundamental models are highly specified and may treat such spikes or dips in general volatility as idiosyncratic and thus diversifiable. In the last six weeks, the statistical trading model has experienced two distinct jumps in US market risk forecasts, while the fundamental model trends lower.
Figure 10: Axioma USA Top 500 predicted risk YTD 2023
These sorts of spreads in model forecasts occur from time to time, and may be an indication that there is a transitory systematic factor in the market that the fundamental model is incapable of recognizing. While attempting to determine what the statistical factors are can be a fruitless exercise, what we can do is look at the difference in risk contributions from the constituents of the portfolio to determine which stocks may have exposure to the statistical factors adding the most risk.
Figure 11: Axioma USA Top 500 risk forecast
As of June 9, 2023, the difference in forecast risk from the statistical model is 757 basis points. The stat model also forecasts about half the level of specific risk, meaning that the additional risk is mostly in the statistical factors. At the asset level, we see the following:
Figure 12: Asset level differences in risk contribution
Every stock in the top 13 (12 if we consolidate Alphabet) besides Apple has higher forecast risk in the statistical model, and universally their correlations to the portfolio are also forecast to be quite a bit higher. It is probably not a coincidence that the ‘big 7’ are all at the top, and the next few names are all tech-related as well. The next largest difference in contribution is with the first non-tech stock, JP Morgan, which is the 13th largest constituent, but 14th on the list. Names like AMD, Broadcom, Netflix and Salesforce leapfrog it from further down the allocation ladder.
This is possibly indirect evidence of a ‘concentration’ factor being picked up by the stat model and adding considerable risk to the portfolio forecast, mostly through these 12 companies, and the ‘big 7’ in particular. We have seen similar divergence in the same group of names in the Axioma US Equity Factor Risk Models – Short Horizon as well, although not in the same magnitude. The medium-horizon models have not picked this up yet.
It seems incongruous that in an era such as ours, with global pandemics, low growth, inflation, war and political unrest in the background, the US equity market would resemble the late 1990’s, only with the same trends magnified. With hindsight, it is clear that the stock market exuberance of the 1990s was an asset bubble, but it was one in which the macroeconomic and geopolitical conditions were generally supportive. The current environment more closely resembles the late 1960s-early 1970s, in which the US stock market was sideways and volatile. The very recent euphoria around Artificial General Intelligence (AGI) and Large Language Models (LLMs) and the hardware and software needed to develop and maintain them would appear to be the operative narrative for the current bout of extreme concentration.
Perhaps AGI will dominate our economy five or ten years hence in a way that justifies the market’s allocation of capital to these seven companies. Perhaps they and their AI prospects are the only bright spots in an otherwise stagnant and directionless US economy. It could be these things, or it could be that allocating to high-growth, expensive tech is a reflex investors have developed because of a perceived margin of safety in these liquid tech behemoths, conditioned as we have all become to the expectation that when the bid fails, central banks will provide unlimited liquidity to stabilize risky asset markets. If (or when) they do, one might expect this cycle to then repeat itself and intensify, as it has done with increasing frequency since the turn of the century.
- Leon Serfaty, CFA, is Senior Product Specialist at Qontigo.