According to a research report from CICC, the core contradiction in the market this year has been the extreme divergence between owning AI and not owning AI. Buying AI means buying win rate, but it requires enduring high crowding and high volatility; buying consumption means buying odds, but fundamentals have been slow to deliver. As of September 26, the top five industries under its win-rate and odds framework were energy, insurance, transportation, pharmaceuticals, and raw materials, with semiconductors ranking sixth. Looking ahead, the main events worth watching in the fourth quarter are: first, technological progress, especially Anthropic's potential listing plan and its third-quarter earnings; second, oil price changes and the Federal Reserve's October and November interest rate meetings; third, fiscal policy, especially the Central Economic Work Conference at the end of the year; fourth, geopolitical factors such as the U.S. midterm elections. CICC's main views are as follows: the core contradiction in the market this year has been the extreme divergence between owning AI and not owning AI. Buying AI is about buying win rate, but it comes with high crowding and high volatility; buying consumption is about buying odds, but fundamentals have yet to materialize. The turbulence since late June shows that relying on win rate alone does not work, as drawdowns in crowded trades can devour the returns from a trend; consumption and the real estate chain have been cheap for two to three years but are still drifting lower, showing that relying on odds alone also does not work. Allocation must answer both the questions of win rate and odds.
The interim reports provide the latest sample for testing these issues. In the second quarter, A-share earnings grew 29.4% year on year, the highest single-quarter growth since the third quarter of 2021, but about 40% of industries still saw year-on-year profit declines, with high growth concentrated in a small number of industries. Similarly, Hong Kong stock interim reports showed the same divergence. In the first half, Hong Kong stock earnings grew 18.2% year on year, with semiconductors, raw materials, and insurance leading profit growth, while consumer industries generally declined. Since the third quarter, the pattern of strong technology and weak consumption has continued; after the Federal Reserve's September rate hike was implemented, market expectations for another hike in October are still building, and external disturbances remain recurrent. At the end of September, this article answers three questions: how to build an industry comparison framework that balances win rate and odds? What clues on win rate and odds did the interim reports provide? And how should allocation be handled in the fourth quarter?
How can win rate and odds both be taken into account? In a market where both volatility and crowding are extreme, simply clustering around the main line or betting on high-low switching has its flaws. To address this pain point, we score and rank industries based on a combination of odds and win rate to find a better allocation strategy. We use MSCI China industry indices to build a win-rate and odds framework: first, win rate describes the certainty of fundamentals, that is, the probability of industry upside over a period ahead; second, odds describe the asymmetry between upside and downside space, that is, the room for mean reversion in valuations. Win rate consists of two parts: micro liquidity and earnings expectations. First, micro liquidity captures the strength of an industry's own trading, including momentum, volatility, and price-volume elasticity. Second, the earnings expectation factor compares dynamic EPS and ROE year-on-year changes across industries, but considering that sell-side consensus forecasts are often lagging, we further supplement the original framework by adjusting earnings expectations with contract liabilities plus advance receipts. Changes in the year-on-year growth rate of contract liabilities and advance receipts, or the second derivative, can reflect the direction of demand changes to some extent. This data is mainly taken from A-share financial reports. Although there may be errors for some industries dominated by Hong Kong stocks, for most industries the data can better reflect the full picture.
Odds are based on a composite of each industry's dynamic PE, PB, and PS rolling five-year percentiles. The cheaper the valuation, the higher the odds score, based on the mean-reversion characteristic of valuations; being cheap itself is an important source of space. The reason for combining win rate and odds is that either dimension alone has clear limitations. Looking only at win rate makes it easy to suffer major drawdowns in crowded trades. If industries are screened only by win rate, our backtest shows that under a positive-weight strategy, where higher scores receive larger positions, the annualized return from February 2021 to August 2026 was 8.1% with a maximum drawdown of -40.4%, clearly worse than the result after combining win rate and odds. Win rate is essentially confirmation of an existing trend. If the trend continues, positions may continue to earn excess returns; however, once the narrative or macro paradigm shifts, the varieties with the most extreme prior gains tend to fall the most when disturbances appear, the so-called winner's curse. This year's semiconductors are an example: consensus expectations and capital crowding pushed the sector rapidly higher in the first half, but under multiple disturbances it has since pulled back sharply from its high since late June.
Looking only at odds creates too much opportunity cost and may fall into a low-valuation trap. If screening only by odds, the annualized return since 2021 has been -4.3%, with losses for three consecutive years from 2021 to 2023. Most assets that the market prices as cheap have reasons for being cheap, and low valuations often correspond to missing fundamentals. The continued weakness in consumption and the real estate chain over the past two years is an example. Betting on odds is essentially betting on reversal, but if the catalyst that drives the reversal never appears, odds assets often continue to contribute negative returns. Especially for relative investors, the problem with looking only at odds lies in the opportunity cost during prolonged weakness.
By contrast, a screening strategy that focuses mainly on win rate while also considering odds can produce better results. In practice, we derive each industry's composite score using 80% win rate and 20% odds score, select the top five industries to build a portfolio, assign weights according to score levels, and rebalance monthly. Backtests show that this strategy has effectively beaten the benchmark index over the long term. From February 2021 to August 2026, the strategy portfolio had an annualized return of 13.2% and a maximum drawdown of -33.8%, significantly better than MSCI China's annualized -7.5% and -59.0% drawdown; in addition, the strategy weighted by score delivered a higher return than the equal-weight strategy, at about 8.8% annualized, also confirming the effectiveness of our scoring.
What clues on win rate can be found in earnings reports? Overall, A-share earnings growth in the second quarter was strong, but high growth was concentrated in a small number of industries, and structural divergence remained extreme. In the second quarter, A-share and non-financial earnings grew 29.4% and 25.2% year on year, respectively, the highest growth since the third quarter of 2021, but about 40% of industries still saw year-on-year profit declines. On a cumulative first-half basis, A-share and non-financial earnings grew 19.4% and 20.6%, respectively. The increase in non-financial earnings was basically contributed by four industries: electronics, non-ferrous metals, petroleum and petrochemicals, and basic chemicals, while the remaining non-financial industries together saw a net decrease. In addition, RMB appreciation magnified exchange losses, accounting for 5.8% of total non-financial attributable net profit, the highest since 2015, with export chains such as home appliances, machinery, power equipment, and automobiles bearing the brunt. Hong Kong stock interim reports were similar to A-shares. In the first half, overall Hong Kong stock earnings grew 18.2%, with the fastest growth also in semiconductors, raw materials, and insurance, while consumer industries such as automobiles and food and beverages generally declined by between 20% and 50%.
From the perspective of win rate, who is delivering and who is falling behind? Those delivering are broad external demand and AI hardware; those falling behind are broad consumption and internet. This is consistent with the current win-rate ranking from our framework. As of September 26, the top five industries by composite score were energy, insurance, transportation, pharmaceuticals, and raw materials. Most of these directions are broad external demand varieties, while semiconductors ranked sixth. First, the industries with leading earnings growth are mostly the high-win-rate directions screened out by the framework. The top six Shenwan first-level industries by second-quarter single-quarter profit growth were electronics, non-bank finance, non-ferrous metals, coal, basic chemicals, and petroleum and petrochemicals, precisely corresponding to semiconductors, which ranked sixth in the framework, and insurance, raw materials, and energy among the top five. In addition, transportation, where the weight is shipping and port shipping profit grew 32.6% year on year, and the innovative drug chain, where biological products grew 141% and chemical pharmaceuticals grew 14.8%, are also delivering on earnings. The reverse also holds: the lagging consumer sectors such as automobiles, food and beverages, social services, and home appliances are all low-win-rate, high-odds directions screened out by the framework.
Second, the industries with leading ROE improvement are also mostly captured by the framework. Among Shenwan second-level industries, the top ten by year-on-year ROE improvement in the second quarter on a TTM basis were energy metals, semiconductors, industrial metals, gaming, insurance, other electronics, communication equipment, batteries, minor metals, and components. Among them, except for batteries and gaming, the other eight belong to the raw materials, insurance, and AI hardware directions captured by the framework. The leading ROE decliners, apart from breeding and feed due to the downturn in the pig cycle, were concentrated in consumer industries such as retail, kitchen and bathroom appliances, baijiu, and white goods, also falling into the low-win-rate directions screened out by the framework.
Beyond backward-looking earnings and ROE, some forward-looking orders and inventories also indicate that divergence is continuing, which can be mutually verified with high-frequency data: orders show that the divergence of strong technology and weak consumption is still continuing. Using advance receipts plus contract liabilities as a proxy for orders, electronics order growth accelerated quarter by quarter to 21.4% year on year, with communication equipment and semiconductors growing 31.5% and 12.4%, respectively; consumer sector orders grew only 0.1%, essentially flat, while food and beverages and commercial retail fell 4.6% and 4.0%, respectively. Further, industries where prosperity is still accelerating are concentrated in a small number of directions. Among 124 Shenwan second-level industries, only 15 simultaneously met four conditions: order growth above 10%, improvement versus the average of the past four quarters, single-quarter revenue growth above 10%, and positive single-quarter attributable net profit. Besides AI hardware, including other electronics, components, communication equipment, and semiconductors, resource products such as minor metals, non-metallic materials, new metal materials, coal mining, and rubber, as well as manufacturing and export chains such as automation equipment, commercial vehicles, shipping ports, and papermaking, were also included, broadly consistent with the high-win-rate directions screened out by the framework, such as energy, raw materials, and transportation.
Combining orders with inventories better shows the current state. In technology, semiconductor and communication equipment inventories grew 40.4% and 40.6% year on year, respectively, with orders and revenue rising together, indicating active restocking under strong demand. In the consumer sector, food and beverage orders fell 4.6% while inventories rose 9.5%, indicating passive inventory accumulation due to insufficient demand; commercial retail orders and inventories fell 4.0% and 3.3%, respectively, indicating active destocking. Since September, high-frequency data show that structural divergence in prosperity is still continuing. The computing power chain remains strong. South Korea's total exports in the first 20 days of September grew 78.3% year on year to $71.4 billion, the highest on record for the same period; semiconductor exports grew 259% year on year, further rising from 199% in August. Memory prices also rose, with the DXI index up 931% year on year in September, slightly down from 1025% in August but still at a historical high, extending the memory price increase cycle. Traditional domestic demand has not yet shown a turning point. The latest August retail sales grew only 0.4% year on year, with automobile retail down 18.5% year on year; according to the China Automobile Dealers Association, nationwide passenger vehicle retail sales in the first 20 days of September were 878,000 units, down 22% year on year, with end demand still weak.
What clues on odds can be found in earnings reports? The highest-odds directions are almost entirely consumption. In the latest results of the win-rate and odds framework, industries such as food and beverages, durable consumer goods, automobiles, consumer services, and e-commerce retail have valuation percentiles at five-year lows, with leading odds scores. The reason consumption is cheap is weak demand, and it is still weakening. At the interim report level, most A-share consumer industry profits declined, and the sector's ROE fell for four consecutive quarters from 9.4% in 2Q25 to 7.1%. Consumption is a margin-driven industry, and the ROE decline came almost entirely from margin compression, with the sector's gross margin falling from 26.4% to 25.6% over the same period. Major Hong Kong consumer sectors also weakened across the board, with first-half discretionary consumption and staple consumption profits falling 28.2% and 22.6%, respectively.
By category: staple consumption clearly weakened. Food and beverage profits fell 18.9% year on year, including a 21.4% decline in baijiu; food and beverage gross margin fell from 50.9% to 48.9%, and ROE fell from 19.0% to 15.8%, while baijiu ROE fell from 24.6% to 19.3%. The divergence between inventories and revenue also shows that the constraint is on the demand side: food and beverage inventories rose 9.5% while revenue fell 4.8%, and baijiu inventories rose 12.4% while revenue fell 17.0%, indicating passive inventory accumulation rather than active expansion. Large-ticket consumption also weakened. After the trade-in policy tapered off, automobile profits in the second quarter fell 19.3% year on year, with passenger vehicle revenue already turning negative at -2.7%, profits down 64.4%, and ROE falling from 7.5% to 3.9%; home appliance profits fell 4.8%, with white goods ROE falling from 19.8% to 15.7%; home furnishing profits fell 50.0%, with ROE falling from 8.2% to 4.3%. The same was true for Hong Kong consumer industries. In the first half, food and beverages, automobiles, consumer services, and discretionary consumer distribution and retail, mainly e-commerce retail, saw profits fall 29.5%, 42.0%, 31.5%, and 49.3%, respectively.
Economic data also show that the weakening of consumption and the real estate chain is continuing. In August, retail sales grew 0.4% year on year, down another 0.2 percentage points from July, while retail sales of consumer goods above designated size fell 3.7% year on year in August. The root of weak demand lies in the household sector, with employment and real estate the core constraints on household credit expansion: first, employment and income expectations are weak. The urban surveyed unemployment rate rose to 5.3% in August, and the employment sub-index of the August manufacturing PMI fell 0.3 percentage points to 48.7%, showing that hiring conditions are still contracting; second, real estate has not yet stabilized, with floor space sold for new commercial housing down 12.1% year on year in January-August, with the decline widening by another 0.5 percentage points from the first half. This also leaves the household credit impulse in a low range, lacking the impetus for spontaneous repair.
Policy signals are currently not enough to reverse the weakness in consumption and domestic demand. With household credit difficult to repair endogenously, the key to a sustained trend in consumption and domestic demand lies in whether fiscal policy can make a major push and tilt toward consumption. The policy strengthening signal from the July Politburo meeting was weaker than the two meetings in September 2024 and April 2025. Even after fiscal efforts later began to catch up, the intensity was still insufficient to drive consumption and domestic demand into a trend recovery. The issuance progress of local government special bonds can support this view. Issuance indeed accelerated after the meeting, with about 520 billion yuan and 620 billion yuan of new special bonds issued in August and September, respectively, and September hitting a new single-month high for the year; however, progress was slow. As of September, cumulative issuance progress was 80.6%, still 3 percentage points below the same period last year and the lowest for the same period since 2023. In other words, at this stage, allocating to consumption mainly earns money from valuation repair and volatility hedging; even if catalysts arrive, consumption still needs to be screened internally.
How should allocation be approached in the fourth quarter? Technology is waiting for industrial catalysts, cyclicals are waiting for rate hikes to land, domestic demand still has to wait for policy, and if none of that works, the only option is dividends. The interim reports confirmed the fundamental basis of K-shaped divergence, and high-frequency data since the third quarter show that divergence is continuing, with strong technology and weak consumption. After the volatility since late June, technology is temporarily in a hesitant position, and external disturbances such as Federal Reserve rate hikes also remain recurrent. Looking ahead to fourth-quarter allocation, sectors can be divided into four major categories according to different drivers: technology is waiting for industrial trends, cyclicals are watching for the Federal Reserve's rate hikes to materialize, domestic demand still has to wait for policy; if none of these work, then the market can only return to the dividend strategy, roughly returning to the pre-September 24, 2024 pattern.
First is technology. Downside risk is easing, while upside space awaits catalysts. This is technology's current dilemma. Three factors mean that technology's downside risk is easing. First, fundamentals are still improving. Our AI bubble pressure index continued to fall in the week ending September 25, with cash flow at -2.99 and external financing at -1.45 as the main sources of improvement, benefiting respectively from improved expectations for free cash flow and cloud business revenue, and improved credit spreads on technology debt. On the demand side, it weakened slightly due to slower growth in average daily token usage; the relative performance of the Philadelphia Semiconductor Index, which had not rebounded clearly earlier, also recovered somewhat. Second, bubble pressure at the trading level continues to ease. Third, the Federal Reserve's September rate hike has landed. Unless there is an aggressive rate hike, the most difficult phase of the rate shock is passing.
But opening up upside space requires industrial catalysts. Without new model breakthroughs or earnings delivery, valuations will find it difficult to break above previous highs. Within technology, certainty differs across segments: models are the most fundamental segment, but it is difficult to identify company-level winners. Without continuous iteration of frontier models, application rollout, new scenarios, capital expenditure, and hardware demand cannot be discussed. This is also why the market worries that slower frontier model iteration could slow overall industry progress. But model investment can judge industry trends, though it is difficult to select individual stocks. Gemini, which the market unanimously favored last year, has clearly fallen behind this year, and today's leading model companies may not still lead next year. The model segment is more suitable for participation through a basket approach.
Applications are important, and the imagination space is mainly in to C, but who can succeed and how to trade it remain uncertain. Progress such as Meta's Muse has led the market to begin expecting AI agents for individual users, and the revenue effect of individual payments is broader than enterprise payments; however, it is currently uncertain who can build it and when it can be done. Hong Kong internet currently remains more of a valuation repair logic, and further trading space also requires application-level breakthroughs similar to a DeepSeek moment.
Short-term certainty lies in hardware, especially bottleneck segments. Before the outlook for models and applications opens up, investments with higher certainty are supply-constrained assets such as U.S. computing power, Chinese chips, optical modules, and glass substrates. In terms of timing, we believe this round's hardware profit cycle may already be in the middle-to-late stage, and the period around 2Q27 is a key window for fundamental verification. Taking optical modules as an example, with financial reporting roughly corresponding to Shenwan third-level communication network equipment and components, demand is still strengthening but has not reached historical extremes. Second-quarter order growth rose to 45.9%, while order growth at the peaks of the two prior solar and battery cycles was close to or above 100%, and both set new historical highs for their respective periods. Fixed asset turnover on a TTM basis bottomed in 1Q25 and has risen for five consecutive quarters. Referring to the 7 to 10 quarter upward cycles in the wind power, solar, and battery cycles, the peak window for this round is roughly 4Q26 to 3Q27.
Cash flow has still not fallen to the key threshold. The ratio of free cash flow to revenue for optical modules on a TTM basis was 2.0% in the second quarter. In reviewing four rounds of emerging technology manufacturing cycles, we found that the main upward wave of excess returns all ended when this ratio turned negative and deteriorated rapidly, at around -10%. Extrapolating linearly at the pace of about 2.5 percentage points of decline per quarter over the past six months, the threshold would be reached roughly in 2Q27 to 3Q27.
Beyond positioning, whether the market can go further depends on catalysts. The high in AI hardware in June this year is closer to the first top of the M-shaped top in the 2021 to 2022 new energy vehicle and solar cycles. At that time, after the first top appeared, the sector pulled back about 30% to 40%, broadly similar to this round. If fundamentals continue to deliver, there may still be repair opportunities, but a clear break above the previous high requires sufficiently strong industrial catalysts, such as AI demand continuing to exceed expectations and leaders' operating cash flow being sufficient to cover rapidly growing capital expenditure, similar to Anthropic's progress in coding in the first quarter of this year.
Second is cyclicals. The logic is relatively simple: watch external demand and the Federal Reserve. On external demand, the September Markit PMI data clearly beat expectations, validating U.S. economic strength and implying that external demand may still have persistence. On monetary policy, current market rate hike expectations remain recurrent. Referring to CME FedWatch, the market expects about a 64% probability that the Federal Reserve will raise rates by another 25bp at its October meeting, with expectations for one to two more cumulative hikes by December roughly balanced. But we believe this is not a long rate-hiking cycle, and the broad direction is still a gradual peak.
Third is domestic demand, where the key still depends on policy. Household credit is difficult to repair endogenously, and without policy support, consumption and domestic demand will struggle to enter a trend move. If policy lands and the household credit cycle restarts, the more efficient allocation choice is Hang Seng Tech. Hang Seng Tech itself can be viewed as a larger version of consumption. During a phase of improvement in the household credit cycle, Hang Seng Tech may perform better than the consumer sector and with greater elasticity, as shown after September 24, 2024.
Finally, if technology does not receive catalysts, cyclicals still face disturbances, and domestic fiscal policy remains absent, then the market has only the dividend direction left, roughly the state of the economy and market before September 24, 2024. The results of our win-rate and odds framework also match the above analysis. As of September 26, the top five industries by composite score were energy, insurance, transportation, pharmaceuticals, and raw materials. Energy and raw materials correspond to the continued delivery of upstream price increases, pharmaceuticals correspond to earnings improvement from innovative drugs going overseas, transportation's win-rate improvement mainly comes from shipping, corresponding to the simultaneous delivery of orders and earnings in shipping ports, and insurance corresponds to high interim profit growth while also combining dividend attributes with not-low odds. Semiconductors ranked sixth by composite score, with win rate still among the leaders but odds already clearly low. The market is still expected to continue with fundamental support, while a break above the previous high requires new industrial catalysts. Broad consumer sectors show typical odds characteristics, consistent with the weak demand shown by both interim reports and high-frequency data and the absence of fiscal catalysts.
Looking ahead, the main events worth watching in the fourth quarter are: first, technological progress, especially Anthropic's potential listing plan and its third-quarter earnings; second, oil price changes and the Federal Reserve's October and November interest rate meetings; third, fiscal policy, especially the Central Economic Work Conference at the end of the year; fourth, geopolitical factors such as the U.S. midterm elections.
Comments