Eight factor indices, three real axes

If you read the names alone, NSE's strategy index family looks like a well-spanned set of exposures. Low volatility, quality, value, alpha, momentum, and several two-factor blends, each with its own published methodology.

We pulled the constituents and looked at what the indices actually hold rather than what they are called.

Three clusters, not eight exposures

Across the family, roughly 48% of constituent slots are redundant once overlap is accounted for. Cluster the indices by holdings rather than by label and they collapse into about three groups.

The names do not predict the grouping especially well. Indices with different factor labels sit in the same cluster, and indices with adjacent labels sometimes do not.

Size is the axis nobody labels

The strongest single explanation for which cluster an index lands in is not its stated factor. It is the market-cap band its screening rules effectively select from.

None of these indices is called a size index. Size is nonetheless doing a large share of the differentiating work, because the eligibility and liquidity screens that sit in front of every factor rule are themselves size screens.

This has a practical consequence. Combining two of these indices to diversify factor exposure often does not diversify very much, because the shared size tilt is a larger common component than the factor difference being targeted.

The momentum caveat

Momentum's excess return over the family, where it exists, concentrates in a specific environment: calm markets, rising, with small caps leading.

That is a narrow enough condition that it invites an obvious follow-up, which is to build an overlay that holds momentum only in that state. We tried that. It failed out of sample, for the third time.

Knowing when an exposure works and being able to trade that knowledge are not the same thing, and the gap between them is where most of the alpha people believe they have goes.

A note on the data

The constituent endpoint for these indices returns HTTP 200 with an HTML body for names that do not exist, rather than a 404. Anything ingesting it needs to validate the payload shape rather than the status code, or it will silently record empty membership for a mistyped index and quietly change every downstream result.