What is multi-factor investing?

It is the practice of building a portfolio around more than one measurable characteristic that has historically been linked to differences in returns, for example value, quality, momentum, low volatility and size. Instead of relying on a single signal, you hold exposure to several, on the view that they do not all struggle at the same time.

What is the difference between blending and integrating factors?

Blending builds a separate sleeve for each factor and then puts the sleeves together, so a name only needs to be strong on one factor to be held. Integrating scores every stock on all factors first and then selects on the combined score, so a name has to be at least acceptable on several factors at once. The two produce quite different portfolios from the same definitions.

Does adding more factors always make a portfolio better?

No. Each added factor dilutes the others, raises turnover and adds another definition that can be wrong. Beyond a small number of well understood and genuinely distinct factors, most of what you add is complexity rather than diversification.

Multi-Factor Investing Explained: Blending vs Integrating

Multi-factor investing means building a portfolio around several measurable stock characteristics at once, such as valuation, profitability, price trend, volatility and size, rather than betting everything on one of them. The reason people do it is simple: single factors go through long stretches of doing nothing or worse, and those stretches do not usually line up with each other, so holding several can make the ride survivable enough that you actually stay invested in the approach.

That is the whole idea in one paragraph. Everything else is craft: which factors, how you define them, how you combine them, and how you decide the weights without quietly fitting the answer to the past.

What counts as a factor

A factor is a characteristic you can measure on every stock in a universe, in the same way, on the same date, which sorts companies into groups that have historically behaved differently. Value sorts on cheapness relative to some fundamental anchor. Quality sorts on profitability, earnings stability and balance sheet strength. Momentum sorts on recent relative price trend. Low volatility sorts on the variability of past returns. Size sorts on market capitalisation.

Two conditions matter more than the label. First, the characteristic has to be computable for the whole universe, not just for the names you happen to like. Second, it has to be computable using only information that was available on the date you claim to have computed it. If either condition fails, you do not have a factor, you have a story. Our longer treatment of the concept and the Indian evidence sits in factor investing in India.

Why combine factors at all

The honest answer is not that combining raises returns. It is that combining reduces the chance that any one definition being wrong, or out of favour for years, wrecks the whole programme.

Consider the shape of the problem. Value can lag for a long time while the market pays up for growth. Momentum can work steadily and then give back a large slice of its gains very quickly when the market turns. Quality can look dull through a strong recovery, when the weakest balance sheets bounce hardest. Low volatility can lag badly in a fast rally. Each of those is a normal feature of the factor, not a malfunction. A portfolio built on one of them has to sit through the full drought alone.

If the droughts do not coincide, holding several factors produces a smoother path. That is the diversification claim, and it is a claim about correlation between factor returns, not about any factor being reliably good. It is worth stating plainly: combining weak signals does not create a strong one. It just spreads the risk of being wrong.

Blending: a sleeve for each factor

The first construction method is blending, sometimes called the portfolio of sleeves. You build a separate portfolio for each factor, using that factor alone to select and weight names, and then you combine the sleeves at whatever weights you choose.

The mechanics are easy to describe. Run the value screen and take the top slice. Run the momentum screen and take the top slice. Do the same for quality and for low volatility. Then hold, say, a quarter of the capital in each sleeve, and let the overlaps take care of themselves. If a stock appears in two sleeves it simply gets held twice, giving it a larger overall weight.

The advantages are practical. Each sleeve is easy to explain and easy to attribute, so when performance is poor you can see which factor caused it. Adding or removing a factor is a clean operation. Reporting to a committee is straightforward because the exposures are visible by construction.

The weakness is the one people notice too late. A stock qualifies for the portfolio by being extreme on a single factor, even if it is terrible on the others. The cheapest names in the market are often cheap for structural reasons, and a pure value sleeve will hold them regardless of what the quality screen says about them. Blending therefore lets the sleeves fight each other. One sleeve can be buying exactly what another would reject.

Integrating: one composite score

The second method is integration. Here you score every stock in the universe on every factor first, standardise those scores so they are comparable, combine them into a single composite, and then select and weight on that composite alone.

The result is a portfolio of names that are decent on several dimensions at once rather than extreme on one. A moderately cheap, moderately profitable company with a steady price trend can outrank a name that is spectacularly cheap and nothing else.

Integration is usually the more capital-efficient way to get multi-factor exposure, because every rupee in the portfolio is working on several factors simultaneously rather than each sleeve working on one. The trade-off is transparency. When the composite disappoints, unpicking why is harder, because no single position exists for a single reason. You also inherit a new decision that blending lets you avoid: how much weight each factor gets inside the composite, which is exactly where fitting to history creeps in.

Choosing weights without fitting the past

Equal weighting across factors is the default worth beating. It embeds a specific admission: you do not know which factor will do well next, so you decline to guess. It is boring, it is robust, and it is very hard to argue against without evidence.

Optimised weights, derived by searching for the mix that performed best over a historical window, are the tempting alternative and the most common way credible-looking work goes wrong. The optimiser will happily tell you that the best mix over the last decade was heavy on whatever happened to work over the last decade. That is a description of the past, not a forecast. If you do go down this route, the minimum honest test is to fit on one period and evaluate on a period you never looked at, which is the subject of in-sample vs out-of-sample testing.

Conviction weights, where the team tilts toward the factors it can explain economically, are legitimate as long as the tilt is written down in advance and not adjusted every quarter in response to recent results. A weighting scheme that drifts toward whatever worked last year is momentum investing on your own process, applied at the worst possible frequency.

A practical build sequence

Whichever combination method you choose, the order of operations is roughly the same.

Fix the universe first. Decide the eligible list, usually by liquidity and listing history, before any factor is computed. A universe that is defined after the fact, using today’s index membership, imports survivorship bias into everything downstream.
Write the definitions down. One page per factor, stating the exact inputs, the accounting basis, the lookback window and the treatment of missing data. If two people cannot compute the same score from the page, it is not a definition.
Score point in time. Every input must be the version that was knowable on the scoring date. This is not a technicality, it is the difference between a test and a fantasy. See why point in time data matters.
Standardise before combining. Raw metrics are on different scales and cannot be added. Convert each to a percentile rank or a z-score within the universe, so that a value score and a quality score mean comparable things.
Decide the neutralisation. Sector neutral scoring stops the portfolio from becoming a single sector bet, since cheapness and profitability both cluster heavily by industry. Not neutralising is a valid choice, but it should be a choice.
Set the rebalance rule and cost assumptions before looking at results. Frequency drives turnover, and turnover drives cost. Details in transaction costs in backtests.
Then, and only then, look at the outcome. If you tune steps one through six until step seven looks good, you have built a curve fit, not a process.

What multi-factor investing does not give you

This is the section most factor material skips, so it is worth being blunt.

It does not remove drawdowns. A multi-factor portfolio is still a long equity portfolio. When the market falls hard, it falls too. Factor diversification operates on the differences between factor returns, not on market risk, which usually dominates in a crisis.

It does not guarantee that the droughts offset. The claim that factors are uncorrelated is an average over history, not a promise about any particular year. Several factors can lag together, and periods where a broad multi-factor blend underperforms a plain index for years are entirely normal.

It does not tell you whether a factor will keep working. Every factor premium documented in research faces the same open questions: was it discovered by looking hard enough at the data, has publication and money flowing in already eroded it, and does it survive real costs. These are covered in factor crowding and factor cyclicality and drawdowns.

It does not judge a business. A composite score is an ordering device across hundreds of names. It cannot tell you that the accounting is aggressive, that the promoter has pledged shares, that the order book is concentrated in one customer, or that the industry is being structurally disrupted. Factor work narrows a universe to a shortlist. It does not replace reading the filings.

It does not survive weak inputs, and it has finite capacity. A scorecard built on inconsistent, restated or partially missing data produces confident, precise, wrong rankings, and the scores look identical whether the data underneath is clean or not. Meanwhile every additional screen tends to push the portfolio toward smaller and less liquid names, so a blend that works on paper at modest size can be untradeable at scale.

Used with those limits in view, multi-factor investing is a sensible discipline: a way to express several defensible ideas about what makes a business worth owning, applied consistently to a whole universe, with the evidence written down before the results arrive.

Portfolio metrics explained: the hub for risk, return and factor measurement.
Factor investing in India: what factors are and how the evidence looks in the Indian market.
Building a factor scorecard: turning factor definitions into a repeatable, documented score.
Factor exposure analysis: measuring what a portfolio is actually exposed to, rather than what it claims.
Common backtesting mistakes: the recurring errors that make factor results look better than reality.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.