The Factor Book

No single new "fourth stream" has replaced the alpha/factor/market trichotomy as the dominant conceptual framework — it remains the practical lingua franca for funds in 2026 — but it has been substantially refined and extended. The most important refinements are (a) a formal split of "factor" into macro factors (growth, inflation, real rates, credit, liquidity, emerging markets) versus style factors (value, momentum, quality, etc.); (b) the collapse of hundreds of documented factors into a smaller set of ~13 robust "themes" (Jensen-Kelly-Pedersen cluster 153 characteristics into exactly 13 themes across 93 countries); and (c) the rise of machine-learning/AI asset-pricing models that treat alpha, factors and market as joint outputs of a single high-dimensional stochastic discount factor rather than three pre-defined buckets.

The chart's progression — an undifferentiated portfolio return in the 1960s, a split into alpha and market return in the 1980s, and a three-way decomposition into alpha, factor return, and market return by the 2000s — tracks the academic record well. Sharpe's CAPM (1962) gave investors a single market beta. Fama and French (1993) added size and value; Carhart (1997) added momentum; Fama and French's five-factor model (2015) added profitability and investment, with Hou, Xue and Zhang's q-factor model standing as a competing four-factor scheme. The same evolution anchors MSCI's July 2025 retrospective, Factor Indexing Through the Decades, which puts more than USD 6 trillion in factor-oriented mandates across active, indexed, and ETF vehicles today — roughly a 50% increase over the decade from about $4 trillion.

So the headline question — have we moved past this model? — has a two-part answer. The streams themselves are intact. What has changed is the resolution at which the industry now sees them.

Refinement 1: Factor splits into macro and style

The most widely adopted advance beyond the simple three-stream picture is the recognition that "factor" was never one thing. It contains two distinct layers operating at different levels of the portfolio.

BlackRock frames it as two types of factors: macroeconomic factors, which capture broad risks shared across asset classes, and style factors, which explain differences in return within an asset class. Its Market Advantage framework names six macro factors — economic growth, credit, emerging markets, liquidity, real rates, and inflation — and treats the familiar style factors (value, momentum, quality, low volatility, and so on) as a separate layer sitting beneath them. MSCI's multi-asset-class model uses the same logic, arranging drivers in a pyramid from macro factors at the top down to security-level characteristics at the base.

The practical effect is to turn the three-stream model into a four-layer one: market return, then macro-factor returns across assets, then style-factor returns within assets, then whatever idiosyncratic alpha remains.

Refinement 2: From a zoo to a handful of themes

The second shift came out of a crisis of confidence in the factors themselves.

By the early 2010s the literature had produced so many candidate factors that John Cochrane, in his 2011 presidential address, described a "zoo of new factors." The skepticism hardened into formal work: Harvey, Liu and Zhu (2016, Review of Financial Studies) catalogued 316 factors across 313 papers and argued that, given the sheer volume of testing, the threshold for a credible discovery should rise from the conventional t-statistic of roughly 2.0 to about 3.0. Harvey and Liu's later census counted more than 400 published factors, and Hou, Xue and Zhang's Replicating Anomalies found that a large share of documented effects simply did not hold up. The collective warning was that most of the zoo was the product of data mining rather than real premia. This did not create a new return stream; it forced a winnowing toward a defensible core — market, value, momentum, quality and profitability, low risk, size, and carry.

The most influential reconciliation is Jensen, Kelly and Pedersen (2023, Journal of Finance), Is There a Replication Crisis in Finance? Using a Bayesian hierarchical model, they reach a notably optimistic conclusion: the majority of asset-pricing factors do replicate; they cluster naturally into 13 themes, most of which earn a place in the tangency portfolio; they hold up out-of-sample in a new dataset spanning 93 countries; and — counterintuitively — the large number of observed factors strengthens rather than weakens the overall evidence. Their public dataset, which groups 153 characteristics into those 13 themes, is becoming a de facto organizing scheme. Robeco's practitioner work (Swade, Hanauer, Lohre and Blitz, Journal of Portfolio Management, 2024) lands in a similar place: a manageable set of robust factors, supplemented by proprietary signals.

Refinement 3: Machine learning redraws the map

The third refinement is the genuine research frontier, and it is the one most likely to reshape the framework over the next decade.

Where the zoo debate was about pruning factors, machine-learning asset pricing tends to embrace them. Rather than positing a handful of pre-specified factors, these models estimate the stochastic discount factor directly from hundreds of characteristics. Gu, Kelly and Xiu (2020) showed that neural networks and tree models substantially outperform linear benchmarks out-of-sample. Kelly, Pruitt and Su's IPCA approach treats characteristics as instruments for time-varying factor loadings — dissolving the clean line between "factor" and "alpha" into a statistical question about whether a characteristic's return is compensation for risk or a true anomaly. More recent work on the "virtue of complexity" goes further still, arguing that models with more parameters than observations can price the cross-section better, inverting the parsimony instinct that drove the zoo critique.

This is why the industry's vocabulary is quietly migrating from "factor investing" to the broader label of "systematic investing." When Invesco renamed its flagship survey in 2023, it was acknowledging a toolkit that now spans macro and regime signals, alternative data, and AI — not just the named factors.

What this means in practice

For all the refinement above, the working language at the major managers has not changed. BlackRock, AQR, MSCI, Invesco, and Robeco still describe returns as some combination of market return (beta), factor or style premia, and idiosyncratic alpha. If you are reading a fund's literature in 2026, you are still reading about three streams.

The more consequential practical change is that the boundary between them has blurred. A great deal of what was once sold as alpha is now available as cheap, replicable factor exposure — "smart beta." As a result, the useful distinction for due diligence is less a set of three discrete buckets than a spectrum running from cheapest to most expensive: traditional market-cap beta, then strategic or smart beta (long-only factor tilts), then alternative risk premia or "exotic beta" (long-short, multi-asset, market-neutral structures that occupy the ground once claimed by hedge-fund alpha), and finally pure idiosyncratic alpha. That spectrum maps directly onto fee tolerance and onto the single question that matters most: how cheaply can this return be replicated by someone else?

The honest summary, then, is that the three-stream model has not been overthrown — it has been given depth. Factor split into macro and style; the zoo was disciplined into themes; and machine learning began to estimate the whole structure jointly rather than bucket by bucket. The chart still works. It just describes the surface of something that has become considerably more layered underneath.