Factor Exposure: The Risk Your Sector Breakdown Hides

Factors are the shared characteristics that explain why groups of stocks move together — regardless of sector. A portfolio spread across five sectors can still be one concentrated factor bet: if every holding is a large, expensive, fast-growing company, the portfolio rises and falls with the growth factor, and the sector labels are decoration.

Every loading in this article is a real regression output, not an illustration. Each one comes from an ordinary least squares fit of a fund's or stock's daily excess returns on the Fama-French daily market, size and value factors, over the trading days from August 10, 2021 to May 29, 2026 — the window where the price series and the published factor series overlap, which is 1,206 observations for each of the funds below and one fewer for the individual stocks. Factor data is from the Ken French Data Library, whose factor definitions were checked September 14, 2026; prices are the daily closes PortLens caches for its own calculations. These historical outputs use the stated window. A new regression may produce different loadings. The methodology explains the calculation.

What are investment factors?

Decades of academic research (Fama and French's work being the foundation) identified persistent return drivers:

The first three — market, size, value — form the Fama-French three-factor model, the standard workhorse for portfolio analysis. Factor exposures typically explain more of a diversified portfolio's behavior than its sector weights do.

What do real factor loadings look like?

Numbers make the abstraction concrete. Below, each fund is regressed on its own, and the ± figure is the coefficient's standard error — how precisely the window pins the loading down.

The fund labels and the loadings agree, which is the first useful thing to establish: the model recovers what the fund names claim. A growth fund loads negatively on value; a value fund and a dividend fund load positively on it, at +0.39 and +0.42 — nearly the same exposure bought under two different labels. VB has the largest size loading in this sample, at +0.64. Most other funds have negative size loadings; VXUS is a small positive exception at +0.09. A loading describes behavior relative to the chosen factors, not a fund's complete holdings composition.

R² is the second thing worth reading. It says how much of the holding's daily movement the three factors explain at all — 99.5% for a broad index fund, 66.8% for an international one. The unexplained variation in VXUS can include currency effects, local-market conditions and other influences. R² alone cannot identify those causes or assign a share to each.

Why do sector labels hide factor concentration?

Because sectors classify what a company sells, while factors classify how its stock behaves. Microsoft (technology), Amazon (consumer discretionary), and Alphabet (communication services) sit in three different sectors and one factor profile: mega-cap growth. A portfolio holding all three plus a growth ETF is quadruple-exposed to the same driver — the same stacking problem that shows up in ETF overlap, one level deeper.

That trio is measurable as a portfolio. Equally weighted, Microsoft, Amazon and Alphabet regress to a market loading of +1.17 ±0.02, a size loading of −0.31 ±0.04 and a value loading of −0.50 ±0.03. Three sectors, three tickers, and a growth tilt larger than the dedicated growth fund's. Add VUG and the tilt does not dilute — a 50/50 mix of VUG and VGT still computes to −0.36 on value, because both funds sit on the same side of it.

The contrast is what a real spread looks like. A 50/50 mix of VUG and VTV — a growth fund and a value fund — lands at +0.02 ±0.00 on value. Their estimated value loadings nearly offset in this window. That does not remove market risk or establish how many companies the funds share; the dated VTV vs VUG holdings comparison measures that separately.

The danger is regime risk. Factors rotate: growth dominated 2017–2021, value snapped back hard in 2022. A single-factor portfolio doesn't just underperform when its factor rotates out — it does so all at once, across every holding, precisely because the holdings were never really different bets.

How is factor exposure actually measured?

The standard method is a time-series regression: regress your portfolio's daily excess returns (returns minus the risk-free rate) on the daily returns of the factor portfolios. The fitted coefficients — the loadings — tell you how much of your movement each factor explains:

• A market loading of 1.2 → amplified equity exposure.
• A positive size loading → behaves like small caps; negative → mega-cap tilted.
• A negative value loading → growth-tilted; positive → value-tilted.
• The regression's R² is the share of return variation explained over the fitted window. Alpha is the intercept: the average excess return left after accounting for the fitted factor exposures. Residuals are the individual daily differences between observed and fitted returns.

A meaningful regression needs enough overlapping history — around 60 trading days at minimum; more is better. This is how PortLens computes it, as part of Pro: an ordinary least squares regression of your portfolio against the daily Fama-French market, size, and value factors from the Ken French Data Library, with the loadings, their standard errors, and R² reported. The full recipe and its fallbacks are documented in our methodology.

Two practical details decide whether the output means anything.

Returns must be aligned by calendar date, not by row. US, London and crypto calendars differ, and matching the 200th observation of one series to the 200th of another compares different days. That biases every covariance toward zero, which makes a concentrated portfolio look diversified.

Recent dates simply drop. The published factor series lags the market by roughly two months, so the most recent weeks of your own price history have nothing to regress against. That is why the window here ends in May 2026 while the price data runs to August.

How do I read a loading I've just computed?

Four questions, in order:

1. Is the loading distinguishable from zero? Compare it to its standard error. VTV's +0.39 ±0.01 is a real tilt. An estimate of +0.03 with a standard error of 0.04 is not clearly distinguishable from zero; the positive sign alone is weak evidence of a tilt.
2. Is it large relative to the market loading? Every long equity portfolio loads near +1 on the market. That is not a finding. The size and value loadings are where portfolios actually differ from one another.
3. What does R² leave out? A single stock is mostly not a factor story. Coca-Cola's daily returns over this window regress to a market loading of +0.38 ±0.03 with an R² of 14.7% — the three factors explain about a seventh of what it does. NVIDIA's R² is 58.6%, Apple's 57.0%. Diversified portfolios sit far higher: VOO at 99.5%. The more concentrated the portfolio, the more of its risk lives outside the model.
4. Would a different window change the answer? It will. These loadings describe August 2021 to May 2026, a period with one particular rate cycle and one particular growth regime in it. A loading is a measurement, not a constant.

Can factor concentration be diversified away by adding holdings?

Not by adding more — only by adding different. This is the point where factor analysis and a holdings count part company most sharply.

Adding a fourth mega-cap growth stock to Microsoft, Amazon and Alphabet raises the position count by a third and moves the value loading further negative. Adding VTV, whose loading has the opposite sign, moves it toward zero. Both are "adding a holding". Only one is adding a bet.

VGT and XLK illustrate the distinction. Their historical market loadings both round to +1.25, while their value loadings are −0.37 and −0.35. The dated VGT vs XLK holdings comparison answers a separate question: how much company exposure they share. Similar regression outputs do not prove identical holdings, and neither measure guarantees how the funds will behave next.

What should I do about factor concentration?

1. Start with what you own. A free portfolio scan shows company exposure, overlap and portfolio beta. Factor loadings require a separate regression; PortLens provides Fama-French analysis with Pro. The free scan does not calculate those loadings.
2. Compare factor and holdings exposure. Opposite value loadings can offset a measured tilt, while both funds still share market risk. Read the regression alongside the dated VTV vs VUG comparison.
3. Watch drift. A winning factor grows its own weight; yesterday's balanced portfolio becomes today's momentum bet without a single trade.
4. Accept deliberate tilts, kill accidental ones. A conscious growth tilt is a strategy; an accidental 90% growth loading discovered after a drawdown is a mistake.

None of that says which tilt to hold. A negative value loading is not a flaw and a positive one is not a virtue — they are descriptions of what your portfolio is exposed to, and what to do about them depends on what you were trying to build, which the regression knows nothing about.

What are the limits of factor models?

The model is a lens, and every lens has edges worth knowing.

• It is descriptive, not predictive. A loading measured over one window says what a portfolio did, not what it will do. Nothing here supports a return forecast, and the factor premia that appear in academic work are long-run historical averages with wide dispersion around them.
• The factors are a choice. Three factors is a convention, not a fact about markets. Adding momentum or quality changes every other loading, because factors are correlated with each other.
• Concentrated portfolios sit largely outside it. At an R² of 14.7%, a factor model is not describing Coca-Cola. Single-stock risk is the residual, and the residual is where a −60% company-specific move lives.
• Loadings move. Both because the portfolio drifts and because the relationship itself changes with the regime. Re-measure rather than remember.

Key takeaways

• Factors describe how holdings behave; sectors describe what companies sell. Only one of them is a risk model.
• Real loadings recover real labels: growth funds compute to −0.35 on value, value and dividend funds to +0.39 and +0.42, and only a small-cap fund carries a meaningful size loading at +0.64.
• Three stocks in three sectors reached a −0.50 value loading — a bigger growth bet than the dedicated growth fund it resembles.
• Diversifying factors means opposite signs, not more line items: 50/50 growth and value computes to +0.02 on value; 50/50 growth and technology computes to −0.36.
• Read R² alongside the loadings. It tells you how much of the portfolio the model is entitled to speak about.

This article is for information and education only and is not investment advice. See our methodology and disclosures.