Methodology — Factor Lens

Transparency is a feature. This page explains what the factor-lens panel shows and why — in plain terms. The precise formulas and numeric conventions are documented in our internal methodology; the summary here is deliberate, not an omission.

Some of your portfolio's day-to-day movement comes from broad forces — the equity market, interest rates, credit spreads, the dollar, commodities, and the classic equity styles (size, value, momentum, quality, low volatility). The rest is specific to the companies you actually picked. This panel splits your book's variance into exactly those two parts and shows the company-specific (residual) share of variance as its headline.

The factor set

Each factor is a transparent spread of liquid ETFs, named verbatim in the panel (for example, market = SPY; size = IWM minus SPY; credit = HYG minus IEF). These are ETF proxies — not the academic factor series — and every construction is shown.

Why the factors are orthogonalized

The proxies overlap by construction: the size spread contains SPY, the credit spread contains IEF. Adding up raw per-factor contributions would double-count that shared movement. So for the variance split, the factors are put through a published hierarchy — market first, then rates, credit, dollar, commodities, then the equity styles — and each factor is replaced by the part of it the earlier factors do not already explain. After that, the shares add up exactly: factor shares plus the company-specific share equal 100%, to rounding.

The order matters, and we say so. Variance shared between two factors is credited to the one earlier in the hierarchy. That choice affects how the explained part is divided among factors — it does not affect the company-specific share itself, which is the same under any order.

The raw factor loadings are also shown, in groups (macro / equity styles), with a collinearity measure (VIF) that quantifies how much the raw factors overlap — which is exactly why they are displayed in groups rather than as ten independent numbers.

Rolling loadings, without look-ahead

The rolling-beta view asks how the book's factor loadings drifted over time. Computing it against factors orthogonalized over the whole window would quietly let each past point see the future. Instead, the orthogonalization at each date uses only data up to that date — the way it would have looked at the time.

When parts of the panel are withheld

Every view has a minimum-data floor (scaled to the number of factors; per-name columns need roughly a year of aligned history). Below a floor the panel says so and why — a withheld number beats a noisy one. Whenever the full lens is withheld, the headline still shows the company-specific share from the live three-factor fit, labeled as such.

Every number on this panel is a model output estimated from one window and one proxy set — a description of that window's co-movement, not a prediction of returns.