Concentration Is a Hidden Factor

Over the twenty-three years since both vehicles have existed, the cap-weighted S&P 500 has beaten the equal-weighted version of the same index by a cumulative 3.8 per cent in log total return. That is close to nothing. Along the way the same measure reached minus 27.9 per cent in April 2015 and plus 8.8 per cent in May 2026. The weighting rule has mattered enormously in both directions and almost not at all in aggregate.

Figure 1 is that history. It is worth starting there because the familiar version of this argument, that concentration has been rising and the cap-weighted index has been pulling away, describes the last three years accurately and the two decades before them backwards.

Section 01Breadth by count, concentration by weight

The trap is the difference between two kinds of breadth. Breadth by count is the number of names held. Breadth by risk is the number of genuinely independent bets those names represent. Capitalisation weighting drives a wedge between them: when the largest constituents carry a large share of the total weight, the index behaves, for risk purposes, more like those few names than like its long tail.

SPY and RSP hold the same five hundred companies. Everything separating their returns is the weighting rule, which makes the gap between them a clean read on how much the rule was worth in a given period. In 2023 it was worth 12.5 percentage points and in 2024 12.1. In 2009 it was worth minus 18.3, and in 2003 minus 11.3. Across the twenty-four calendar years in the sample the gap has a mean of roughly zero, a standard deviation of 6.8 points, and a positive sign in half of them.

Section 02Concentration as a factor

A gap of that size in either direction is a factor exposure, whichever way it points. The mechanism is the one Figure 2 sets out: the index narrows to its dominant weights, those weights sit heavily in one sector, and that sector loads on a small set of shared drivers, including long-duration cash flows and a collective expectation about capital spending. Because the largest names share those drivers, they move together, and a shock to any one driver reaches all of them at once.

January 2025 was the live demonstration, and it is the subject of this volume's first issue. On 27 January the cap-weighted vehicle fell 1.41 per cent while the equal-weighted one rose 0.07 per cent, a single-session gap of 1.48 percentage points that sits in the bottom half of one per cent of all daily observations in this sample. Nothing about the constituent list changed. The weighting rule was carrying a position in one forecast about AI capital spending, and that day the position was marked.

effective_constituents  N_w  =  1 / sum_i ( w_i^2 )      # inverse Herfindahl

    requires constituent-level index weights, which are licensed.
    N_w is NOT computed anywhere in this article.

what is computed instead:

    gap_t  =  r_cap,t  -  r_equal,t  =  sum_i ( w_i - 1/N ) * r_i,t

    the return consequence of the weighting rule, from two traded series
    holding an identical constituent list.

Section 03A portfolio problem and a stability problem

What makes this more than a private portfolio matter is that the same exposure is held in much the same form across the system, often through vehicles that can be forced to sell. In its May 2025 Financial Stability Review the European Central Bank recorded that equity markets remained vulnerable to sudden and sharp adjustments given persistently high valuations and risk concentration, and separately that liquidity mismatches in some open-ended funds, together with procyclical flow dynamics, could amplify an adverse shock. A shared exposure, richly priced, held through vehicles that must sell into weakness, is the same loop this journal meets from the strategy side in Issue 017.

Where the evidence in Figure 1 stops is worth marking clearly. It shows that the weighting rule has been worth a great deal recently, and that this has happened before with the opposite sign. It does not show that concentration is at a record, because it does not measure concentration: the top-five weight and the Herfindahl index need constituent-level weights, which S&P Dow Jones Indices licenses rather than publishes, and no proxy for them is constructed here. Nor does the recent run establish that a reversal is due. Twenty-four annual observations, half of them positive, is not a sample from which a mean-reversion claim can be made.

The label says diversified. The variance says whatever the weighting rule happened to be pointing at.

The useful distinction is between constituent breadth, weight breadth, and risk breadth. The first counts securities, the second measures how capital is distributed, the third asks how many independent exposures survive covariance. A capitalisation-weighted index can score well on the first and poorly on the other two, and the gap in Figure 1 is the only one of the three this research can measure from public data.

This research is analysis and commentary for general information. It is not investment advice, an offer, or a solicitation, and it contains no price forecasts. Return figures are the author's own calculations from the source cited; regulatory descriptions summarise the publications named.