Factor Investing: Value, Momentum, Quality, Size and Low Volatility

A portfolio returns 11.0 percent in a year against a 2.0 percent risk free rate. Excess return, 9.0 percent. Regressed against the standard factor series, 8.2 points of that turn out to be exposures anyone could have bought in an index product, and 0.8 points are attributable to the manager. The manager, in most cases, did not know which of the two they were producing.

That decomposition is the reason to care about factors at all. A factor is a sorting rule with a return series attached: rank a universe by a number you can compute, hold the top slice, short the bottom slice, rebalance on a schedule, and record what the portfolio earned. Everything else is an argument about which numbers to rank on and whether the resulting premium is payment for risk or a mistake investors keep repeating. How portfolio risk is measured is the prerequisite, since the arithmetic here is built on beta and Sharpe ratios.

Run the regression before anything else

Rp - Rf = alpha + b(MKT-RF) + s(SMB) + h(HML) + m(UMD)

Take the portfolio above. The regression over the prior sixty months returned loadings of b = 1.05, s = 0.30, h = 0.45 and m = -0.10. Over the year in question the factors returned MKT-RF = 7.0 percent, SMB = 2.0 percent, HML = 1.5 percent and UMD = 4.0 percent. The explained portion is

1.05 x 7.0 + 0.30 x 2.0 + 0.45 x 1.5 + (-0.10) x 4.0 = 7.35 + 0.60 + 0.675 - 0.40 = 8.225 percent

leaving alpha of 9.0 - 8.225 = 0.775 percent for the year.

Most portfolios that feel like skill decompose approximately this way. If h comes back at 0.6 and s at 0.4, you own a small value fund assembled by hand, and the honest comparison is against a cheap product with the same tilt, on cost and on drawdown. Feed the residual series into the Sharpe ratio calculator to see whether the tilt was paid for in risk adjusted terms or merely in volatility.

What a factor is, in construction terms

The canonical construction is a double sort. Split the universe by size at the median. Split again into three buckets by the characteristic. That gives six portfolios. Value is then

HML = 0.5 x (small value + big value) - 0.5 x (small growth + big growth)

and size uses the same six portfolios along the other axis:

SMB = (1/3) x (small value + small neutral + small growth) - (1/3) x (big value + big neutral + big growth)

Eugene Fama and Kenneth French built the model that put those two alongside the market factor. Mark Carhart added momentum, written UMD for up minus down, constructed the same way but sorting on the return from twelve months ago to one month ago.

That skipped month is a construction detail. It has real consequences. Include the most recent month and the measured premium shrinks noticeably. Recent losers bounce. Anyone comparing momentum results across studies without checking the skip is comparing two different strategies.

The five that survive most scrutiny

Factor	Sorting variable	The usual economic story
Value	Book to price, earnings to price, cash flow to price	Payment for distress risk, or overreaction to bad news
Size	Market capitalization	Illiquidity and higher failure risk among small firms
Momentum	Return from month 12 to month 2	Slow diffusion of information, and herding
Quality	Gross profitability, accruals, leverage, earnings stability	Investors underpay for durable profitability
Low volatility	Trailing volatility or beta	Leverage constraints push investors into high beta names

Value and size are the Fama French pair. Momentum is the Carhart addition. Those three series are what nearly every published attribution runs against. Quality and low volatility arrived later, have shorter clean histories, and are defined in more ways by more providers, which makes comparing results across studies harder than it looks.

The growth versus value guide covers the same ground from the stock picking side, and the momentum trading guide covers what momentum looks like in a single name.

Risk premium or persistent mistake

Return data alone cannot separate the two explanations. That is worth stating plainly, because a great deal of marketing material pretends otherwise.

The risk story says a factor premium compensates for holding something genuinely dangerous: cheap stocks are cheap because their businesses are fragile, small companies fail more often, momentum crashes without warning, and under that reading the premium should persist, because the risk does.

The behavioural story says investors systematically misprice these characteristics, extrapolating recent growth too far or paying up for lottery like volatility, and under that reading the premium should shrink once the mispricing is documented and capital arrives to exploit it.

Both stories predict positive average returns in the historical sample. They disagree about the future. The data available cannot settle which one is right.

The long stretches of nothing

Every one of these premiums has disappeared for periods long enough to end careers, and this is the part that gets underweighted in fund literature.

Value underperformed growth for most of the decade after the 2008 financial crisis. That is long enough that a large share of the investors who allocated at the start had given up before any reversal arrived, which means their realized experience of the value premium was entirely negative regardless of what the long run series says. Momentum carries a different shape of pain: positive on average with occasional crashes that hand back several years of gains in a few months, typically when a market rebounds violently off a low and the short leg, stuffed with beaten down names, rips higher. Size has been the weakest of the classic three in live data since publication.

The framing that survives scrutiny is that a premium is payment for holding something uncomfortable, and the discomfort is the mechanism. A premium that never hurt anybody would have been arbitraged away. That framing does not establish that any of these premiums will exist going forward, and an allocation to them should be treated as a probabilistic bet on a decade scale horizon.

Crowding, capacity and the cost of turnover

A published anomaly attracts money. Money changes the thing it is chasing. Once enough capital sorts on the same variable, valuations on the long leg rise against their own history, the trades become correlated so the drawdowns arrive together, and rebalancing costs rise because everyone reconstitutes on similar dates.

Turnover is where paper premium dies. Value portfolios turn over slowly, often 20 to 30 percent a year. Momentum portfolios turn over several times a year by construction. The point is to hold what just rose. At institutional scale that turnover is priced with the same tools as any other execution, which is where market microstructure stops being academic. A factor with a 3 percent gross premium and 200 percent annual turnover can deliver a negative net premium in mid cap names, and the backtest that ignored costs will never mention it.

What a long only factor ETF actually gives you

A long only factor ETF holds market beta plus a diluted tilt: if a fund sorts the S&P 500 into its cheapest third and equal weights the result, its realized h loading might be 0.3 against the 1.0 of the academic long short series. You are buying roughly a third of the factor. Plus all of the market.

Price it that way. The decision gets easier. A 0.25 percent fee for an h of 0.3 costs 0.25 / 0.3 = 0.83 percent per unit of factor exposure, against a plain index fund that charges a fraction of that for the beta alone. Check what the sort does to sector weights too, because a naive value screen frequently resolves into a financials and energy position, which the sector ETF guide covers.

Combining factors without doubling the same bet

Value and momentum have historically been negatively correlated, which is the strongest practical argument for holding both: the combination has usually produced a higher Sharpe ratio than either alone. Quality and low volatility overlap heavily. Both tilt defensive. Holding both is closer to one position than two.

Measure the overlap. Put the candidate return series into the correlation matrix and read the pairwise numbers across several sub periods, because factor correlations move around and tend to rise in stressed markets, which is the point the diversification guide makes about everything else in a portfolio.

Two implementation choices then matter more than the choice of factors. Combining signals, by scoring every stock on both and holding the best composite, avoids buying and selling the same stock in two separate sleeves and usually carries lower turnover than combining portfolios. Rebalance frequency trades signal freshness against cost. There is no universally right answer.

Where this fails

It fails as a timing tool. Attempts to overweight whichever factor looks cheap against its own history have a poor live record, and the valuation spread between value and growth stayed extreme for years before it meant anything.

It fails when the horizon is wrong. A premium that shows up over decades is no reason to hold a concentrated tilt with money needed in three years. Size the tilt against the drawdown you can sit through, using the same discipline as any other position sizing decision, and remember that the estimation error around a factor premium is wide enough that the honest input to a sizing formula is a range.

Next, macro regimes and asset allocation covers which economic conditions have historically favoured which of these tilts, and why that relationship is weaker than it is usually presented.