The Sector Rotation Playbook

Every few years, a new cohort of investors discovers sector rotation and concludes they have found the key to the market. The logic is appealing in its simplicity: different sectors of the economy perform differently at different points in the economic cycle. Financials lead recoveries. Consumer staples hold up in recessions. Energy outperforms during inflationary expansions. Technology dominates secular growth periods. If you can identify where you are in the cycle and position accordingly, you capture the upside of each sector's strongest phase without suffering through its weakest.

The idea is not wrong. The evidence that sector rotation, executed systematically, generates excess returns is real — modest, inconsistent, and far more complicated in practice than the theory suggests. The evidence that AI can meaningfully improve on systematic sector rotation is newer, thinner, and genuinely interesting.

The History of the Idea

The formal articulation of sector rotation as an investment strategy dates to Sam Stovall's work at Standard & Poor's in the 1990s^[35]. Stovall mapped the historical outperformance of eleven S&P sectors across the four phases of the economic cycle — early expansion, late expansion, contraction, and recovery — and documented that certain sectors reliably led or lagged at each phase. The framework became a standard reference in equity strategy.

The introduction of GICS (Global Industry Classification Standard) in 1999, jointly developed by MSCI and S&P, gave sector rotation a standardised vocabulary and a tradeable universe^[37]. The subsequent proliferation of sector ETFs made tactical sector rotation accessible to any investor with a brokerage account.

Academic examination of the evidence was, characteristically, more skeptical than the practitioner consensus. Work by Moskowitz and Grinblatt in 1999 identified that much of individual stock momentum could be attributed to industry momentum, suggesting that sector trends were more persistent than individual stock trends^[36]. Subsequent research — including the broader value-and-momentum framework documented by Asness, Moskowitz, and Pedersen — found that simple momentum-based sector rotation generated excess returns above a market-cap-weighted index in most periods studied^[38].

The qualification embedded in "most periods" is important. Sector momentum has profound vulnerability to sharp reversals. The March 2020 COVID crash reversed years of energy and financial sector outperformance in weeks. The 2022 technology correction inflicted losses on momentum investors who had been concentrated in the sector's decade-long outperformance.

Where the Market Cycle Theory Breaks Down

The appeal of cycle-based sector rotation obscures a fundamental problem: in real time, it is very difficult to know where you are in the economic cycle, and by the time it is obvious, the market has already priced the positioning.

Economic cycle phase identification requires, in practice, a judgment about the current state of the economy based on lagging indicators (GDP, unemployment, earnings) that are published with a delay of weeks to months and subsequently revised. The market, which is forward-looking and aggregates the views of millions of participants, is making its own cycle judgment continuously.

Research by Friesen and Sapp (2007) documented that the average mutual fund investor's timing of sector fund purchases and sales destroyed approximately 1.5% of annual return relative to buy-and-hold, because investors rotated into sectors after their strong performance rather than before it^[39]. The cycle-based sector rotation that looks compelling in retrospect is, in real time, executed after the cycle signal has already moved the market.

The Thematic Alternative

The more defensible version of sector rotation is not "I think we're in early recovery, therefore buy financials." It is "I believe this structural economic trend is real and is developing faster than consensus estimates suggest, and these companies are specifically exposed to it in ways the market hasn't fully repriced."

The distinction is between cyclical timing and structural identification. Structural shifts — the multi-decade buildout of AI infrastructure, the energy transition, the demographic pressure on healthcare systems, the reshoring of semiconductor manufacturing — unfold over years or decades, not quarters. ASML's effective monopoly on EUV lithography, for example, is the kind of structural positioning that compounds over a full cycle rather than reversing on a single macro print^[40].

Where AI Has a Genuine Potential Advantage

A large language model can read an entire regulatory docket, a year of earnings call transcripts across an industry, the relevant academic literature on a technology, and the patent filings of the key players in an afternoon. Whether it can integrate this information into a thesis that is both correct and non-consensus — rather than summarising the consensus view that is already embedded in its training — is the genuinely uncertain question. The honest answer is that it can, sometimes, in specific domains where the relevant information is dense, textual, and not yet fully synthesised by the analyst community.

The practical implication is a constraint on which sectors are worth researching. Technology and consumer discretionary themes are extensively covered and efficiently priced. Industrials, materials, energy infrastructure, and international markets — sectors with less analyst coverage and more complex, technical information flows — are where AI synthesis is more likely to surface something that isn't already priced. The sectors that feel boring and technical are where the potential information advantage is largest.