July 2026 was a challenging period for Chinese equity investors. The CSI 300 Index declined 7.4%, marking its largest monthly retreat since January 2016. Broader market indices saw even deeper drawdowns, with the CSI 500 and CSI 1,000 indices falling 16.8% and 19.6%, respectively. Yet the headline figures only tell part of the story. Beneath the surface, the unwinding of concentrated market leadership created a difficult environment for many quant equity strategies, producing significant performance dispersion across the onshore industry. While headline commentary has focused on the sudden momentum reversal, we believe the divergence in manager performance came down to two capabilities: alpha diversification and benchmark-aware portfolio construction techniques.

Understanding July’s moves first requires a review of the quarter that preceded it, when the Chinese equity market was characterised by historic sector bifurcation. Performance was dominated by the Information Technology and Communication Services sectors, as the global AI trade propelled supply-chain constituents to returns often associated with past investment bubbles. This narrow leadership resulted in notable industry dispersion, with the return spread between the top and bottom performing industries reaching its highest level for the CSI 300 and CSI 500 indices since 2015.

Figure 1: Q2 and July 2026 CSI 300 Index return by sector (top) and industry quarterly return spread, best versus worst (bottom)

Source: Bloomberg and Internal Man Group Databases. Date range: January 2015 – July 2026. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

Problems loading this infographic? - Please click here

From a factor perspective, these second-quarter moves favoured Barra Momentum, which recorded its highest single monthly return of the past decade in June 2026. This trend was amplified by a surge in market leverage, with aggregate margin trading balances reaching CNY three trillion by late June, up by more than 60% since January 2025. A significant portion of this leverage was concentrated in technology stocks, which increased the sensitivity of the market to any change in narrative.

Figure 2: China A-share margin trading balance by sector

Source: Shanghai and Shenzhen Stock Exchange and Internal Man Group Databases. Date range: January 2014 – July 2026.

Problems loading this infographic? - Please click here

In July, the dynamic shifted, with global sentiment turning against semiconductors and AI hardware names, triggering a reversal in onshore technology leaders and a broad deleveraging event across the market. The top-performing factors and sectors of the second quarter saw severe losses in July, unwinding months of gains in just a few weeks.

The bifurcation over the two periods was most evident in the Momentum factor, which had essentially become an expression of the AI euphoria that had manifested across various sectors and perceived beneficiaries of the technology boom. June and July 2026 represented the best- (+18.0%) and worst- (-28.9%) performing months for the factor since 2015, with the scale of these returns in part reflecting the increased leverage in the market, particularly in the Technology sector, which saw the most significant increase as it evolved into a crowded momentum trade.

Figure 3: Q2 and July 2026 Barra factor return (top) and ranked monthly Barra momentum returns (bottom)

Source: MSCI and Internal Man Group Databases. Date range: January 2015 – July 2026. The factors do not represent investible vehicles and are presented for illustrative purposes only. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

Problems loading this infographic? - Please click here

The distortion of benchmark composition within the traditionally more diversified CSI 500 and CSI 1,000 indices was a notable side effect of the second-quarter rally in technology names, primarily semiconductor manufacturers exhibiting strong large-cap growth characteristics. As these constituents surged, index concentration reached historical highs, visible in both the aggregate top 10 constituent weights and effective constituent counts.

Figure 4: Sum of top 10 index constituent weights (top) and effective constituent counts (measured by 1/Herfindahl-Hirschman Index (HHI)) (bottom)

Source: CSI, Bloomberg, and Internal Man Group Databases. Date range: January 2015 – July 2026. The indices do not represent investible vehicles and are presented for illustrative purposes only.

Problems loading this infographic? - Please click here

Problems loading this infographic? - Please click here

With concentration levels spiking to historical highs in the CSI 500 and CSI 1,000 indices, the typical risk controls employed by many benchmark-aware quant equity strategies were placed under significant strain. When market sentiment shifted in July, the pullback in crowded technology constituents triggered a sharp factor reversal, resulting in a difficult month across various onshore quantitative strategies and impacting both market-neutral and index-enhanced formats.

However, that impact wasn’t uniform across the quantitative landscape. July exposed positioning built up over previous months, where some managers had taken polarised bets to maintain returns. Navigating the reversal came down to two areas: diversifying across independent alpha sources and managing risk around benchmark concentration.

Alpha diversification

Over recent years, the onshore Chinese quant equity industry has witnessed a secular decay in certain traditional technical factors. As high-frequency price and volume signals became widely adopted and crowded in large- and mid-cap spaces, the alpha available from these more generic strategies steadily deteriorated. To illustrate this industry-wide trend, Figure 5 shows the yearly returns of a generic technical factor long/short spread,¹ split by market capitalisation, to highlight where the decay has been most pronounced.

Figure 5: Generic technical factors long/short spread

Source: Man Group Databases. Date range: January 2015 – July 2026. The factors do not represent investible vehicles and are presented for illustrative purposes only. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

In response to this decay, quantitative frameworks had to evolve. To maintain historical return profiles without introducing new signal types, many strategies migrated down the market capitalisation spectrum to capture residual technical alpha in less efficient, smaller cap spaces.

However, leaning into smaller caps introduces notable structural vulnerabilities. As illustrated by the short Barra Size factor return (Figure 6), a persistent small-cap bias has historically carried substantial drawdown risk, as seen during early 2024 and through the first half of 2026. Because smaller companies have restricted trading volume and free float, simultaneous exits by systematic strategies create meaningful execution friction when sentiment turns.

Figure 6: Short Barra Size factors return

Source: MSCI and Internal Man Group Databases. Date range: January 2015 – July 2026. The factors do not represent investible vehicles and are presented for illustrative purposes only. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

This dynamic created a polarised factor environment in 2026, where generic technical strategies delivered volatile, regime-dependent returns based on their underlying factor tilts.

While broader small and mid-cap indices, like the CSI 500 and CSI 1,000, suffered sharp headline declines in July due to the collapse of their heavily-weighted technology constituents, underlying factor spreads told a different story. As shown in Figure 7, the long/short payoff across generic technical combinations² was bifurcated. Momentum-based technicals performed strongly in the second quarter before unwinding in July. Conversely, combinations with small-cap, low-volatility and lower-liquidity tilts saw the inverse pattern: facing a prolonged drag across the first half of the year before rebounding on a relative long/short basis in July as market participants rotated away from crowded momentum names.

For single-pillar strategies, this polarisation created a difficult trade-off. Holding either profile required absorbing a significant drawdown in one of the two periods. For strategies that attempted to rotate exposures to chase trailing performance, it often meant taking the loss in both.

Figure 7: The polarised payoff of generic technical factors

Source: Man Group Databases. Date range: January 2015 – July 2026. The factors do not represent investible vehicles and are presented for illustrative purposes only. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

Rather than relying on a single factor style or attempting to time market regimes, we believe that a quantitative framework might be better served by diversifying across orthogonal alpha pillars.

Figure 8 illustrates this dynamic across proprietary models, split into alternative, fundamental, and technical sleeves. The primary strength of this framework is diversification across market regimes. In the second quarter, while fundamental signals faced headwinds during the narrow technology rally, alternative data models captured the upside. In July, as market dynamics rotated, fundamental signals rebounded. Furthermore, proprietary model design, such as focusing on short-term liquidity dislocations and mean-reversion, may help to eliminate the unintended factor bets that affect generic models, as historically demonstrated by the outperformance and stability of proprietary technicals across both periods.

A balanced allocation across these independent pillars may provide a more consistent return profile, seeking to ensure that performance is not beholden to any single alpha source.

Figure 8: Long/short spread return by proprietary model³ over Q2 2026 and July 2026

Source: Man Group Databases. Date range: April – July 2026. The factors do not represent investible vehicles and are presented for illustrative purposes only. Past performance is not indicative of future results.

Problems loading this infographic? - Please click here

In our view, alongside a balanced allocation to innovative alpha styles, diversification within those buckets is equally as important. Within the alternative data sleeve, for instance, one compelling area of research involves looking beyond domestic datasets to map global industry networks.

By systematically tracking over 120,000 active cross-border relationships, quantitative models may connect China A-share companies directly to international technology hubs across the US, Europe and Asia. Using heat diffusion matrices and similarity metrics across global supply chains, these systems trace how innovation cycles and demand shocks propagate from multinational tech leaders down to domestic component suppliers.

Figure 9: China A-shares’ global connections growth

Source: Man Group Databases. Date range: January 2020 – July 2026.

Problems loading this infographic? - Please click here

Global industry networks are just one example of how new sources of information may be incorporated to seek to improve alpha signals beyond traditional domestic datasets. The value lies not in any one signal, but in combining differentiated sources of alpha. As the differing outcomes across the second quarter of 2026 and July demonstrated, we believe diversification is essential to support a more consistent return profile across market environments.

The benchmark-aware hurdle

Balanced allocations across innovative alpha signals are only half the equation. In our view, for enhanced index strategies, portfolio construction is equally crucial. Unlike pure long-short setups where sector and style risks may be hedged out, long-only strategies must operate directly within the constraints of the underlying index.

A common approach is heuristic optimisation: ranking stocks by alpha and applying broad aggregate constraints, such as sector weight bands, to force benchmark alignment. However, aggregate constraints operate only at high levels. They miss the constituent-level concentration and stock-specific crowding that often drive benchmark performance during market turning points.

To test this, we simulated 5,000 randomly selected long-only portfolios,⁴ each constrained to match the CSI 500 Index across Barra-style exposures (+/-0.3 standard deviations) and active industry weights (+/-3.0%). If aggregate constraints were sufficient to manage benchmark risk, these portfolios should have tracked the index closely.

Instead, the index realised extreme tail outcomes relative to the distribution. During the second quarter, the CSI 500 Index returned +17.6%, sitting at the 99th percentile of simulated portfolios. In July, as the index fell -16.8%, it dropped to the 1st percentile (Figure 10).

Figure 10: Performance of 5,000 randomised simulated portfolios versus CSI 500 Index

Source: Man Group Databases. Date range: January 2015 – July 2026. Q2 2026 and July 2026 are shown for illustrative purposes only to demonstrate the statistical extremity of returns. Box-and-whisker plots show the distribution of simulated gross-of-fee returns, including the median, interquartile range and outliers, for 5,000 randomly selected, non-investable portfolios. The sample size was determined at discretion to provide statistical robustness. Simulated or hypothetical results have inherent limitations, do not represent actual trading and are not indicative of, or a forecast or guarantee of, future performance. Results are subject to change.

Problems loading this infographic? - Please click here

For this simulation it confirms that the extreme benchmark moves were not driven by broad sector or style factors, which all 5,000 simulated portfolios shared, but by highly concentrated, idiosyncratic stock risks trapped inside the actual benchmark. Portfolios optimised purely on aggregate rules of thumb remain blind to these tail risks.

When constructing an active benchmark-aware portfolio, rather than relying on rigid heuristic constraints to force a fit, we believe that an optimiser must factor for return, risk and trading costs simultaneously. This manages real-world implementation while aiming to maintain the portfolio tightly anchored to the underlying benchmark at a constituent level, seeking to avoid the unmanaged concentration risks that heuristic models miss.

Conclusion: next time will be different

One key lesson from the July reversal is that the next market dislocation will potentially look entirely different. Risk management in the onshore China market has at times been fragmented and reactive. In 2024, the industry learned the dangers of unconstrained style exposures and rushed to implement basic controls. Today, managers are discovering that aggregated constraints are entirely insufficient without a rigorous approach to portfolio construction relative to the benchmark.

When market dynamics shift again, crowding will likely take a different form. Patching heuristic rules after the fact may leave portfolios one step behind. Ultimately, we believe that seeking to deliver investors a consistent return profile across market cycles demands two fundamental disciplines: genuinely diversified alpha and constituent-level risk controls.

1. The generic technical factor proxy is constructed using approximately 200 factors from onshore market data provider technical factor libraries (as a proxy for inter-day signals) and order-book flow factor libraries (as a proxy for intraday signals). The factors are equally weighted to form a technical composite and tracked as a long/short spread.
2. Constructed from the Footnote 1 proxy library as of 31 March 2026. The Momentum combo equally weights the top 20% of factors by Barra Momentum exposure. The low-vol, low-liquidity and small cap combo equally weights factors in the bottom 20% of Barra residual volatility, liquidity and size exposures. Both are tracked as long/short spreads over the second quarter and July 2026.
3. Constructed from the equally weighed long/short return spreads of the proprietary alpha models within each category over the second quarter and July 2026
4. Data is based on a simulation of 5,000 long-only portfolios. Each hypothetical portfolio was randomly generated from a universe of approximately 5,000 China A-share stocks and comprises 200–500 stocks, with individual weights below 1%. Portfolios were constrained to remain within ±3% of the CSI 500’s sector allocation and ±0.3 standard deviations of its Barra style-factor exposures. The 5,000-portfolio sample size was selected at the portfolio manager’s discretion to support statistical robustness. The portfolios are non-investable and are presented for illustrative purposes only.