CITIC SEC highlights that "structural rapid rotation" has become the norm for A-shares since 2022, with extreme rotation speeds typically persisting for one to two months. The common thread behind this high-speed rotation is an insufficient breadth of earnings upgrades, leaving a scarcity of sustainable market-leading themes.

Currently, as trade frictions enter a high-incidence period and the impact of exchange gains and losses amplifies, overseas expansion—one of the most critical earnings drivers for A-shares—faces valuation constraints. This curtails the market's breadth and fosters an environment prone to rapid sector rotation. Breaking this deadlock requires fresh catalysts. Recent progress in AI has reinforced the existing trend of surging computing power demand but falls short of altering the long-term commercialization narrative. Should new developments, such as advancements in RSI (Recurrent State Inference) or anti-distillation technologies, emerge, they could unlock further valuation upside.

During phases of rapid rotation, most winning style strategies belong to the "low-valuation" camp. However, only the PB-ROE strategy demonstrates a clear advantage in such conditions, with its returns strengthening as rotation speed intensifies. Conversely, momentum strategies suffer the most damage in these environments. Under high rotation, market opportunities may increasingly manifest as valuation recoveries for stocks supported by solid earnings. In terms of allocation, our mid-year strategy proposed a "AI + Energy & Chemicals" barbell structure, with AI peaking in Q2 and the energy and chemical sectors likely to heat up progressively in the coming period.

Patterns of Rapid Rotation Since 2022

We have constructed a "mainline retention rate" indicator to gauge the speed of market rotation. Its median level saw a one-time structural shift upward in 2022 (monthly average median rising from 0.71 to 0.77), marking the onset of A-shares' "structural rapid rotation" norm. In August of this year, the indicator's monthly average reached 0.853, placing it at the 97.5th percentile historically (with only early 2016, early 2022, and September 2023 recording higher readings since 2010). As of August 27, the reading stood at 0.875, indicating the market remains in an extremely high-speed rotation state.

However, such rapid rotation phases typically do not last long. Since 2012, we have identified 38 distinct high-rotation periods, with a median duration of just one month and an average of 1.9 months. 65% of these phases conclude within the first month, and only 22% persist beyond two months. Yet, this round's extreme readings, coupled with the inertia of the 20-day smoothing, caution against relying solely on historical patterns to predict an end to high-speed rotation in September. Confirmation requires signals from earnings upgrade data.

Root Cause: Insufficient Breadth of Earnings Improvements

A narrowing breadth of earnings improvements shares a stable contemporaneous relationship with accelerated rotation. Since 2015, the proportion of A-share companies with year-on-year ROE improvements has been highly correlated with the mainline retention rate. When earnings improvements concentrate in fewer sectors, these sectors exhibit higher volatility (rapid surges followed by swift corrections). Once such sectors enter a phase of valuation digestion, the market lapses into a state of missing themes and rapid rotation. Therefore, determining whether high-speed rotation will conclude hinges on assessing the breadth of earnings improvements.

We find that the proportion of large-cap companies with upward earnings consensus revisions serves as an effective leading indicator for the breadth of earnings improvements, offering predictive power over the mainline retention rate with a lead time of around one month. A two-month smoothed measure of this revision ratio provides even stronger predictive insight. Currently, the proportion of large-cap companies with upward revisions was just 33.8% in July, rebounding to 45.5% in August. If this ratio maintains a level above 46% in September, it would provide stronger confirmation of a slowdown in rotation speed for October.

Trade Frictions and FX Losses Intensify, Narrowing Market Breadth

Recent Sino-US economic and technological frictions have visibly escalated, with policy tools trending toward broadened application from entity lists to product access, critical supply chains, and general tariffs. Since late July, the US has tightened access for foreign-made robots, networked inverters, and grid equipment, while China has strengthened export controls on dual-use items related to US drones. Meanwhile, Reuters reports suggest the US is mulling restrictions on Chinese data center components like optical modules and considering an additional 7.5% tariff on Chinese goods citing "overcapacity."

Although some measures are not yet formally implemented, and some may serve as bargaining chips ahead of leader meetings, the rising density, sector coverage, and policy hierarchy of friction events have raised tail risks for high-value overseas expansion supply chains. These effects may manifest in market risk appetite and valuations before actual export changes occur. Sino-EU trade negotiations are also entering a critical validation window. September to October will test whether technical consultations can translate into tangible outcomes on market access and trade balance. Any setbacks could expose European-heavy companies to rising market access and compliance costs, earnings forecast volatility, and valuation discounts.

Additionally, the mid-year reporting season has again validated the negative impact of exchange losses on corporate profits amid sustained RMB appreciation. For non-financial firms that have published interim reports, overall financial expenses rose 67.4% year-on-year, with the increase accounting for 9.8% of the net profit base from the same period last year. For companies with overseas revenue exceeding 30%, the financial expense ratio rose by 1.41 to 2.52 percentage points year-on-year (sample averages across different overseas revenue brackets). While we believe trade frictions will not alter the long-term trajectory of Chinese companies' internationalization and industrial competitiveness, overseas expansion—one of A-shares' most vital earnings drivers—indeed faces valuation constraints in the near term, narrowing market breadth and fostering rapid sector rotation.

Breaking the Deadlock: AI Advances Reinforce, but Not Transform, the Narrative

NVIDIA's latest earnings report, particularly its significantly better-than-expected forward guidance, positively impacted numerous supply chain companies. Yet, the stock performance of these related companies shows signs of fatigue. Rapid growth on North American TaaS platforms (especially AWS Bedrock), combined with Anthropic's potential massive IPO fundraising (and the accompanying expectations for computing power investments and stronger model releases), has largely dispelled concerns about decelerating commercialization of computing infrastructure. However, these marginal positive developments are more an intensification of the "computing power shortage" narrative, without fundamentally altering the long-term commercialization story for end-user applications. In essence, these industry improvements may support earnings forecast upgrades but cannot transform hardware stocks' cyclical valuation framework.

The market has formed a consensus: frontier model developers consume substantial computing costs to advance models, open-source models narrow the capability gap via distillation and post-training to cut costs, capture high-end market share through pricing advantages, and application companies leverage low-cost models with proprietary data for vertical markets. Recent announcements challenge this consensus. Zhipu AI stated that all traffic for GLM-5.3-Flash is carried by a cluster of over 100,000 domestic chips. Tencent's Hy4 demonstrates strong capabilities in most office tasks at a fraction of frontier models' pricing. These developments directly challenge the business models of frontier model developers, thereby affecting the economics of massive future (rather than current) computing investments.

To alter the existing narrative, marginal improvements in model capabilities or TaaS platform growth will have limited impact. What's needed is either a fundamental technological breakthrough like RSI, establishing a solid link between computing advantages and model capability gaps, or progress in anti-distillation, preventing capabilities from being replicated at low cost and thereby reviving the story of competing for computing power to build moats.

Winning Strategies in Rapid Rotation Markets

Since no change in the rapid rotation landscape has yet emerged, we must seek optimal coping strategies. Examining 15 style strategies, low-PE and low-PB strategies show monthly average return differentials of 219 and 196 basis points between high and low rotation periods, respectively. PB-ROE (residual method), HML value factor, PB-ROE (distressed reversal), and low-priced stock strategies show differentials of 147, 136, 125, and 124 basis points, with both PB-ROE-based strategies achieving statistically significant positive differentials. Most strategies performing well during high-speed rotation share "low-valuation" characteristics.

Plotting strategies on a map with rotation sensitivity on one axis and long-term expected returns on the other, we can categorize them into three groups: 1) "Rotation offensive" strategies—low-PE, low-PB, HML value factor, and PB-ROE (residual method)—exhibit high rotation sensitivity, but some lack outstanding long-term average returns, making them suitable for phased increases when market themes are unstable and sectors shift rapidly; 2) Quality, earnings drift, and profitability factors are insensitive to rotation states but offer relatively stable long-term returns, ideal as foundation positions independent of market state judgments; 3) Momentum and momentum-chasing strategies have significantly negative rotation sensitivity, favoring clear themes and persistent trends, but warrant reduced exposure during rapid rotation phases. Thus, in current conditions of insufficient theme sustainability, style allocation should tilt toward low-valuation and valuation-earnings matching strategies, reducing reliance on short-term trends.

PB-ROE's Edge: Strengthening with Faster Rotation

Using the mainline retention rate to segment rotation intensity into three tiers, only PB-ROE-based and momentum strategies exhibit statistically significant return differences across varying rotation speeds—with PB-ROE benefiting and momentum clearly suffering. Within the PB-ROE framework, the traditional residual method strategy outperforms the CSI All-Share Index by an average of 158 basis points per month during high-speed rotation periods. Incorporating the "distressed reversal" factor lifts this outperformance to 179 basis points monthly, with the strategy's historical performance notably surpassing the residual method. Under the PB-ROE (residual method) strategy, low-valuation and high-quality factors contribute primary excess returns; current suitable sectors include industrial metals, coal, and logistics. For PB-ROE (distressed reversal), supply-side clearing driving improved supply-demand dynamics and valuation recoveries contributes primary excess returns; current suitable sectors mainly include chemicals, batteries, and aquaculture.

Strategic Allocation: The "AI + Energy & Chemicals" Barbell

Our mid-year strategy proposed an "AI + Energy & Chemicals" barbell structure: AI as the offensive end, catering to aggressive capital seeking high growth and new paradigms, and energy & chemicals as the defensive end, satisfying allocation capital's demand for high-certainty returns. Contrary to our expectations, the two did not coexist but rather alternated in time—AI dominated in Q2, drawing capital away from energy & chemicals, which now appears to be reversing. This round of energy & chemicals strength likely reflects spreads rather than solely rising oil prices. According to shipping research firm TankerTrackers, recent transit volumes through the Strait of Hormuz have notably increased, with at least 15 STS transfer operations simultaneously occurring in the Gulf of Oman, involving approximately 25 million barrels of crude oil and some refined products, sourced from nearly all regional countries except Iran. Actual vessel transits may exceed apparent figures (nighttime convoy passages, some vessels disabling transponders for "dark sailing," satellite imagery typically captured in the morning, leading to systematic underestimates in AIS data).

However, global refining capacity impact remains significant. Diesel cracking spreads have widened notably, with a reality emerging where crude oil is not scarce but refined products are. Underground capacity (crude oil) can recover quickly once the strait is clear, but above-ground capacity (refining and petrochemicals) recovers much more slowly. Combined with China's persistently low operating rates and inventory levels, weak demand factors are already priced in. While the market's cyclical attention has been focused on non-ferrous metals, the energy & chemicals chain may currently offer better elasticity. In terms of allocation strategy, controlling expectations and avoiding "grand narratives," approaching the market with a range-bound mindset remains the core principle. Within the tech sector, rotate toward core assets (such as gas turbines, wafer manufacturing platforms, semiconductor equipment), emphasizing "volume certainty" while treating "price explosiveness" cautiously. For non-tech sectors, focus on increasing allocations to energy & chemicals, innovative drugs, and leading brokerages with overseas expansion potential.

Risk Factors

Potential risks include: intensifying Sino-US frictions in technology, trade, and finance; domestic policy intensity, implementation effectiveness, or economic recovery falling short of expectations; unexpected tightening of domestic and global macro liquidity; further escalation of conflicts in Russia-Ukraine and the Middle East; and slower-than-expected digestion of China's real estate inventory.