A-Share Market Recap | 2026-09-16 Wednesday

Read timestamp: quotes are 2026-09-16 close-of-session settlement values (Tencent qt.gtimg.cn timestamp 202609161534, independently cross-checked against Eastmoney push2delay; across 8 key names, turnover and percentage change differ by 0.00% between the two sources). Price limits: SSE/SZSE main board ±10% | ChiNext (30x) / STAR Market (68x) ±20% | BSE ±30%. Every stock in this report is labeled with its board. Limit-up / limit-down / failed-limit (炸板) counts use the exchange-official pools, validated with tc == len(pool) (ZT 89=89, DT 4=4, ZB 11=11). Whole-market benchmark: 5,551 valid samples, average +1.44%, median +0.98%, 75.1% advancing. Every "hit" in this report is judged on excess return (beating +1.44%), not on "closing green" — 75.1% of stocks closed green today, so the absolute measure carries almost no information. Dragon-Tiger List (龙虎榜): this report was generated at 15:30; the Dragon-Tiger List is only published after 18:00 and was not obtained for this issue, so §5 contains no seat-level analysis. Northbound flows: the exchanges stopped disclosing daily net buying as of 2024-08-19, so this report provides no such reading (the API silently returns 0.0; treating that as real would be fabrication).

---

0. Today's Recap in One Line

1. Sentiment jumped straight from "freezing point" to "broad-based euphoria": limit-ups 32 → 89, limit-downs 27 → 4, failed-limit rate 42.9% → 11.0%, promotion rate 13.0% → 37.5%, gap-up hold rate 17.4% → 72.9%. Every reason to "avoid chasing" that applied yesterday reversed today.

2. The strongest theme was electronics / semiconductors / optical communications; the branches ranked #2 and #3 pre-market won, the one ranked #1 lost. Semiconductors +4.82% (184 up / 1 down), semiconductor materials +6.14% (31 up / 0 down), semiconductor equipment +5.03% (25 up / 0 down); the CPO concept took +RMB 15.720 billion of main-force net inflow, first among all concepts. Meanwhile the pre-market top branch, copper-clad laminate (CCL), was +1.25% on the day, excess −0.20pp — the only one of the three branches to underperform the market.

3. The pre-market report's top arbiter returned a self-contradictory verdict, and that has to be cleared up first: the §8① primary criterion (CCL group average >0 and at least 3 names up) passed; the secondary criterion (CCL group > generic-board group) failed (+1.25% vs +3.88%, a gap of −2.63pp). But the primary criterion carried no information today — with 75.1% of the whole market up, the base pass rate for "at least 3 of 5 randomly drawn names up" is 89.8%. It measures the market, not CCL. The only meaningful reading is the secondary criterion: CCL lost, and it lost even though the "contamination source" assumed pre-market went the other way (崇达 (Suntak) and 满坤 (Mankun) were supposed to pull back after their "sentiment overheating" announcements; 崇达 actually rose +7.94%).

4. The pre-market report broke the gross-margin yardstick, and the new yardstick it swapped in failed today's test too. Full-sample measurement: Spearman(9/15 single-day main-force net inflow, 9/16 % change) = +0.055 (n=399), consecutive net-inflow days = +0.071. The top-decile 9/15 net-inflow group was +2.36% today; the bottom-decile net-outflow group was +2.49%, a gap of −0.13pp. Single-day fund flow is not a contrarian indicator — it is simply uncorrelated. Pre-market swapped a negatively correlated ranker (gross margin −0.415) for a zero-correlation ranker.

5. Almost all of the picks' excess return came from the overnight gap; not one extra cent was earned after the open. Day change +2.10% vs market +1.44% (excess +0.66pp), but open→close +1.53% vs market +1.58% (excess −0.05pp). Executed at today's open exactly as written, the list returned zero relative to the market.

6. Today's prettiest call was passing on cybersecurity: 天融信 (Topsec, 002212) −10.04% limit-down, 中新赛克 (Sinovatio, 002912) −9.98% limit-down, 奇安信-U (QI-ANXIN-U, 688561) −3.70%. Cybersecurity Awareness Week does not end until 9/20, yet the branch faded early, inside its own event window.

7. Tone for tomorrow: the broad rally has been delivered, but theme concentration is insufficient. The largest theme cluster in the limit-up pool is only 6 communications-equipment names (6.7%), and first-time limit-ups account for 86.5%; the electronics sector's +RMB 22.703 billion is day 1 of a single-day reversal after 5 straight days of net outflow from 9/11–9/15, not a trend confirmation.

---

1. Market Overview

Pattern: the whole market gapped down, then rose one-way all day. Three of the four major indices opened lower and all closed sharply higher; open→close was positive for every one of them and larger than the full-day change. STAR 50 at +4.14% led by a wide margin, a single-day dominance of the tech-growth style.

Sentiment indicators (vs 9/15)

Sentiment temperature: euphoric (broad-rally type, not theme-led type). Failed-limit rate 11.0%, promotion rate 37.5% and gap-up hold rate 72.9% all strengthened together — a tape with excellent follow-through. But the first-time limit-up share of 86.5% and the extreme industry dispersion of the limit-up pool (largest cluster 6 names = 6.7%) say this is a liquidity-driven broad rally, not the money-making effect of one theme spreading. Those two things imply completely different things for tomorrow; see §6.

Performance today of yesterday's 32 limit-ups: 23 up / 9 down, average +3.65%, average open→close +0.83%, 12 hit the limit again (promotion rate 37.5%).

---

2. Reconciliation Against the Pre-Market List

① The 5 Picks

Hit rate

⚠️ The single most important reconciliation conclusion: all of the excess came from the overnight gap. The picks' day excess was +0.66pp, the open→close excess −0.05pp. The 0.71pp difference between the two is exactly the gap-open contribution, and that gap had already happened by the time the list was finished — it was not tradable. Executed at today's open as written, this list's return relative to the market is 0.

② The 25-Name Pass List

Pass calls correct: 13/25 = 52% (Pass group average day change +1.72%, excess only +0.28pp).

⚠️ Split the Pass list by rationale and the conclusion flips — this is the one thing to remember from today:

Pass rationale type	n	Avg. day change	Excess	Verdict
"Logic / fundamentals do not hold" (label denied by the company, profit unverifiable, loss-making)	13	+0.46%	−0.98pp	✅ effective
"Price level too high / sentiment metrics" (elevated price, poor limit board, hot money, small size)	12	+3.09%	+1.65pp	❌ ineffective

Table scrolls horizontally

"Logic does not hold" Passes underperformed the market by 0.98pp; "price level too high" Passes beat the market by 1.65pp. Passing on a name because "it is already high" amounts to systematically throwing away the strongest names — 闽东电力 (6 consecutive limit-ups), 锡华科技, 通鼎互联, 华瓷股份 and 超声电子, five names that hit or nearly hit the limit, all came out of the "price level too high" category.

③ Self-Critique in One Line

What was called right were the "falsification" judgments; what was called wrong were the "ranking" judgments.

• Called right: every case that used a primary disclosure to negate a label (天融信's 0 institutional seats, 意华's interim report mentioning servers 0 times, 双星's own statement that gross margin is negative) delivered today, and delivered very cleanly (two limit-downs). 金安国纪's absolute-threshold verification point, "heavy volume with a stalled price means distribution", is the only criterion in the whole report that precisely predicted a loss in advance.
• Called wrong: every case that ranked branches and individual stocks in order was wrong today. The branch ranking 1/2/3 had an actual strength order of 3≈2 >> 1; among individual names, 拓荆科技, given the smallest position, performed best, and 北方华创, rated weakest on fundamentals, came second.
• How to change the next pre-market report (three items, in priority order):
  1. For any criterion of the form "at least N names up" or "sector index turns positive", the whole-market base rate must be computed before it is written down. Today the CCL primary criterion had a base pass rate of 89.8%; it could not possibly have falsified anything. A criterion with a base rate above 60% should never be written into the list.
  2. Stop using single-day fund flow to rank direction. Measured today, Spearman = +0.055 (n=399), consecutive days = +0.071, both are zero. Pre-market had just used empirical evidence to break the gross-margin ranker (−0.415), yet put the new yardstick straight into service without subjecting it to the same test. Swapping yardsticks requires measuring the new one first.
  3. "Price level too high" may only cap position size; it must not become a Pass rationale. This is the second consecutive time the same conclusion has come out of reconciliation (last time the group gap was −0.21% vs +6.18%; today it is −0.98pp vs +1.65pp). The next pre-market Pass list must have two columns: "falsification-type Pass" goes into Pass, while "position-type Pass" only cuts position size and stays in the watch pool.

④ Line-by-Line Settlement of the Pre-Market Verification Points

Meta-conclusion on the verification points: of the 10 criteria, only 金安国纪's actually pointed to a loss in advance; 2 passed but lost money (华正 and the CCL primary criterion), 1 had a wrong baseline (the semiconductor "3rd day"), and 1 had a threshold sitting on the noise (生益 98.72 vs 100). "Pass rate" is not a measure of criterion quality — a criterion has to test the variable that will lose me money.

---

3. Validating Today's Leading Themes

① Did the Pre-Market Report Identify the Strongest Theme? — No, and It Got Both Ends Wrong

Actual strength order: optical modules / CPO ≈ semiconductor equipment / materials >> copper-clad laminate. Pre-market order: copper-clad laminate (1) > semiconductor equipment (2) > optical modules (3, explicitly marked "do not chase").

The two ends were completely swapped. Pre-market ranked CCL first and committed 3 of the 5 pick slots to it, yet CCL was the only one of the three branches to underperform the market; optical modules, which pre-market explicitly described as "only an oversold stabilization, do not chase", took the largest concept-level fund inflow in the whole market today (+RMB 15.720 billion).

Why the ranking was wrong — pre-market wrote down the reason itself: it downgraded optical modules on the basis that "中际旭创 had main-force net outflow of RMB 1.109 billion on 9/15 (the largest in the sample)". Today 中际旭创 had main-force net inflow of +RMB 1.402 billion and the stock was +5.07%. In the same group, 新易盛 went −RMB 1.082 billion → +RMB 920 million and 天孚通信 −RMB 61 million → +RMB 447 million. A whole branch was downgraded on one single-day fund-flow extreme, and that variable was measured today as zero-correlated.

⚠️ But this must not be taken too far. My initial inference was "single-day fund flow is a contrarian indicator" — the full-sample test rejected that claim: the top-decile 9/15 net-inflow group was +2.36% today, the bottom-decile net-outflow group +2.49%, a gap of only −0.13pp, Spearman +0.055. The correct statement is "single-day fund flow has no predictive power for next-day direction", not "it runs in reverse". The beautiful contrast of 华正新材 (+RMB 707 million → fell) and 中际旭创 (−RMB 1.109 billion → rose) is an n=2 coincidence, not a rule.

② The Surprise Theme Pre-Market Missed: PCB Manufacturing (Components)

All 4 "component" names in the limit-up pool are PCB manufacturers: 澳弘电子 (605058, SSE main board ±10%, 4 consecutive limit-ups, second-highest of the day), 威尔高 (301251, ChiNext ±20%), 奥士康 (002913, SZSE main board ±10%), 依顿电子 (603328, SSE main board ±10%). Copper-clad laminate (CCL), meanwhile, had 0 limit-ups.

Pre-market Passed on the entire "generic / small-to-medium-batch PCB" line, on the grounds that the interim reports showed "revenue growth with profit decline" plus the companies' own statements about cost squeeze. Today that group was +3.88%, excess +2.43pp, with 崇达技术 at +7.94%. Pre-market had itself already written that "this is the highest-risk judgment in this list" — and that self-warning was realized today.

Industry implication: the pre-market bet logic was "raw-material price hikes ⇒ buy upstream CCL, avoid downstream PCB". The answer the market gave today was the opposite: 4 limit-ups among downstream PCB manufacturers, 0 among CCL. Combined with pre-market's own statistic that "CCL has underperformed generic boards by 46.7pp over the past 40 days", this is no longer one day of noise but 41 consecutive days in the same direction.

---

4. Limit-Up Ladder and the High-Board Names

89 limit-ups | 4 limit-downs | 11 failed limit-ups | failed-limit rate 11.0% | 77 first-time limit-ups (86.5%)

Consecutive-Limit Ladder

Quality of the high boards: the highest board is the worst, the second-highest is the cleanest.

• 闽东电力 (000993), 6 boards: 20 failed limits, 21.8% turnover rate, seal order / turnover of only 0.024. Pre-market Passed on it for having "the worst seal quality in the pool" — the factual description was entirely correct, yet it still produced the highest board of the day. This is the textbook form of "chip metrics size a position, they do not set direction" — the quality judgment was right, the direction judgment wrong.
• 澳弘电子 (605058), 4 boards: only 1 failed limit, 7.6% turnover rate, seal order / turnover 0.147 — the only name in the ladder that is strong on both height and quality. And it belongs to the PCB manufacturing branch that pre-market misjudged today.

Heavyweight Leaders (recognizable names with large turnover and significant gains)

The heavyweight leaders are entirely concentrated in optical modules and semiconductors, and three of the top four are optical modules. This is the hardest evidence for where today's leading theme sits — the largest turnover went to the line that pre-market told readers "do not chase".

Broken Boards / Limit-Downs

• 4 limit-downs: 天融信 (002212, −10.04%), 桂林旅游 (Guilin Tourism, 000978, −10.03%), 国芳集团 (Guofang Group, 601086, −10.01%), 中新赛克 (002912, −9.98%). Two are cybersecurity, two are retail / tourism — the pre-market decision to avoid both directions was entirely correct.
• Of the 11 failed limit-ups, 2 are wind power (吉鑫科技, 天顺风能), both gapped up, failed at the limit and fell back sharply (open→close −8.42% / −6.35%).

---

5. Fund Flows

⚠️ This section contains no Dragon-Tiger List or northbound data: the Dragon-Tiger List is published after 18:00 (this report was generated at 15:30 and did not obtain it); northbound daily net buying has not been disclosed since 2024-08-19.

Top Industry Main-Force Net Inflows

Top Concept Main-Force Net Inflows

Main-Force Net Outflows

Per-Stock Fund Flow (picks and key names, unit: RMB 100 million)

Two fund-flow patterns worth remembering:

1. The electronics sector's +RMB 22.703 billion is a single-day reversal after 5 straight days of net outflow from 9/11–9/15 (−99.93 / −53.71 / −4.41 …). The size is enormous but it is only day 1, and it cannot be treated as a trend confirmation.
2. 通鼎互联 (002491) had the largest net inflow market-wide on 9/15 at +RMB 1.955 billion; on 9/16 it had RMB 53 million of net outflow and still hit the limit. Money and price decoupled, consistent with the pre-market characterization that "the source of profit cannot be verified" — a limit-up does not mean the logic has been confirmed.

---

6. Outlook for Tomorrow

① Theme Continuation Assessment

⚠️ One reservation about the overall tone: today was a broad-rally tape (75.1% advancing, 86.5% first-time limit-up share, largest theme cluster in the limit-up pool only 6.7%). On a broad-rally day, branch strength readings are contaminated by overall beta — how much of semiconductors' +4.82% was the theme and how much was the market cannot be separated today. The "continues" judgments in the table above therefore rest entirely on breadth and fund flow, not on percentage gains.

② Key Names to Watch Tomorrow

Note on verification-point design (a direct product of this reconciliation): for every line above, the base rate was computed first, or it is stated why the threshold is not automatically met. Today's lesson is that the CCL primary criterion "passed" with a base rate of 89.8%, which amounts to no judgment at all. In addition, "at least N names up" criteria are deliberately no longer used; they are replaced by forms such as "decliners ≤ K" and "same sign for a 2nd straight day" that will not pass automatically on a broad-rally day.

③ What to Avoid Tomorrow

1. The entire cybersecurity line — 天融信 (002212), 中新赛克 (002912), 奇安信-U (688561). Two limit-downs, sector fund flow in consecutive net outflow, and the branch died before its event (Awareness Week, ending 9/20) was even over. Once an event-driven branch fades early, there is no "event delivery" rebound rationale.
2. Wind power / electric power — 大金重工 (002487), 天顺风能 (002531), 吉鑫科技 (601218). Fund flow turned negative after 4 positive days; all three gapped up and then fell back sharply today (open→close −4.72% / −6.35% / −8.42%), a textbook gap-up distribution.
3. 闽东电力 (000993, SZSE main board ±10%) — a lone 6-board name, the highest of the day but the worst in quality: 20 failed limits, seal order / turnover 0.024. In the same branch (electric power) only 3 names hit the limit today, with no second high board to take the baton — this is a lone-runner move.
4. Chasing the copper-clad laminate chain — 华正新材 (603186) saw main-force flow flip from +RMB 707 million to −RMB 595 million today and gapped up then fell back; 金安国纪 (002636) was the only pick that fell and the pre-market distribution criterion has already been confirmed.
5. Any stock selection based on "single-day main-force net inflow rankings" — measured today, Spearman = +0.055 (n=399); it has no predictive power for next-day direction.
6. 9/17 lockup expiry pressure: about RMB 7.61 billion of shares unlock, of which roughly RMB 6.63 billion is a single name (from the pre-market §8⑤ calendar; the stock's name was not obtained because the API returned no code/name column).

④ Input Notes for Tomorrow Morning's Pre-Market List

1. Rank branches on "breadth", not on "single-day fund flow amount". Today's two strongest branches already had the cleanest breadth readings pre-market (equipment 23 up / 2 down), yet pre-market ranked that one #2; CCL was ranked #1 on the strength of a fund-flow figure (华正 +RMB 707 million), and that variable was proven today to have no predictive power. In the next issue, please make "industry decliner count" the first column of the branch ranking.
2. The Pass list must be split into two columns: "falsification-type Pass" (primary disclosure negates the label, profit unverifiable) — excess −0.98pp today, keep passing; "position-type Pass" (elevated price, poor limit board, hot money, small size) — excess +1.65pp today, change to "cut position size but stay in the watch pool"; they must not be excluded. This is the second consecutive time the same conclusion has emerged.
3. Every verification point must have its whole-market base rate written beside it, and criteria with a base rate >60% must not be written into the list.
4. Verification-point baselines must be checked on the spot, never written from memory. Today's "semiconductor 3rd straight day" was wrong (9/14 was −RMB 7.100 billion, so it was actually day 2).
5. Sentiment is elevated today (89 limit-ups, 11.0% failed-limit rate, 37.5% promotion rate), so if tomorrow opens higher again, watch especially for the "gap-up hold rate" falling back from today's 72.9% — this indicator has swung from 17.4% to 72.9% within two days, is extremely volatile, and its persistence cannot be assumed.
6. The memory branch needs to be reassessed: the two-day observation that "the good news is not landing on the three leaders" was correct, but the branch as a whole was +4.21% with RMB 13.211 billion of inflow, and the cost of labeling the entire branch "avoid" has now shown up. Recommend switching to a two-part treatment: "avoid the leaders, watch the branch".

---

⚠️ Data-collection failures and open items for this issue (not sent to clients)

1. Eastmoney's push2.eastmoney.com went from working to 502 during this generation run: on the opening probe, clist returned normally (total=5915); about 10 minutes later the same URL returned 502 across the board, and parameter bisection (six variants across pz/np/invt/fields) confirmed this was not a parameter problem but the endpoint itself going down. After switching to push2delay.eastmoney.com, the full pull succeeded with 5915/5915 and no pagination gap (total==got was explicitly validated, given the prior record of "Eastmoney clist pagination one-sided bias"). 88./19./48.push2 and push2his all returned RemoteDisconnected during that window.
2. ⚠️ New trap (contradicts the pre-market record; the note must be updated): the pre-market internal record said "daykline only returns the last 1 day under lmt=0". This issue proves that is a host difference, not a parameter problem — push2delay and push2 return only 1 day at any lmt (including 30), while push2his.eastmoney.com returns 121 days normally under lmt=0. Multi-day fund flows for both industries and individual stocks must go through push2his; this issue's consecutive net-inflow day counts and Spearman tests all depend on that one channel.
3. ⚠️ A silent limit-down-pool failure was intercepted: getTopicDTPool with sort=fbt:asc (the limit-up pool's parameter) returned tc=4 but an empty pool, the classic "tc is non-zero but pool is empty" silent failure. Checking the akshare source confirmed that the DT pool must use sort=fund:asc; after the fix, 4=4 closes. All three pools have been validated with tc == len(pool).
4. Dragon-Tiger List not obtained: stock_lhb_detail_em returned TypeError: 'NoneType' object is not subscriptable at 15:37; this is a scheduling problem, not an API failure (the Dragon-Tiger List is published after 18:00). §5 has been explicitly left blank, and no seat-level conclusion was used. To include the Dragon-Tiger List, this slot would have to move to after 18:30.
5. No primary source obtained for attribution: three WebSearch runs returned nothing whose publication date could be verified as 2026-09-16 (what came back were semiconductor market pieces from April/May/June/July, with undated headlines pinned to the top of the results — exactly the "recurring column headline hides the publication date" pattern). This report therefore assigns no catalyst at all to today's semiconductor / optical module surge, and only describes the structure (materials +6.14%, equipment +5.03%, CPO first in fund flow). The 9/15 《电子信息制造业"十五五"规划》 (the electronic information manufacturing industry's 15th Five-Year Plan) explicitly includes "high-density memory chips" and "high-end memory", and today is its day 2, but I have no evidence of causation, so it was not written into the body as attribution.
6. ⚠️ The single most important self-interception this issue: from a 15-name sample I saw the strikingly clean contrast of 华正新材 (9/15 +RMB 707 million → fell today) and 中际旭创 (9/15 −RMB 1.109 billion → +5.07% today), and almost wrote "single-day fund flow is a contrarian indicator". After expanding to the full 400-name sample, Spearman = +0.055 and the TOP-vs-BOTTOM group gap was only −0.13pp, which rejected the contrarian-indicator claim; the correct conclusion is "no predictive power". This is the second instance of "a monotonic subsample is a coincidence", and this time I nearly committed it myself. §3① of the body explicitly writes out this reversal process.
7. Caliber and cross-validation record:
   • Turnover and percentage change for 8 key names were compared across Tencent qt.gtimg.cn and Eastmoney push2delay, differing by 0.00%.
   • The conclusion that 生益科技's RMB 9.872 billion vs the RMB 10 billion threshold is a "1.3% miss = fail" was confirmed across both sources, since it affects the verification-point settlement.
   • Ex-rights check: today's previous close for all 5 picks was compared one by one against the 9/15 close and all matched (150.25 / 638.16 / 83.99 / 251.57 / 649.00), so there is no ex-rights contamination. The pre-market §8⑤ note flagged 30 ex-rights names today; a check confirmed none of them are within the reconciliation scope.
   • Advancer/decliner counts are based on 5,551 valid samples (of 5,915 total, 364 were suspended or had no quote). The Spearman test sample is the top 400 names by turnover (minimum RMB 990 million), of which 399 were retrieved; the last date is 2026-09-16 for every one of them, with no stale data.
8. Not obtained / explicitly left blank: (a) the Dragon-Tiger List and seat data (scheduling); (b) northbound flows (disclosure discontinued, not a failure); (c) a primary catalyst for today's leading theme; (d) the codes and names of the 9/17 lockup-expiry stocks (the API does not return that column, and nothing was fabricated); (e) attribution for limit-up industries such as automation equipment (北自科技, 2 boards), packaging and printing (4 names), plastics (4 names), non-baijiu beverages (3 names) and household products (3 names) was not obtained in any case and is honestly left blank; (f) roughly half of today's 89 limit-ups have no industry attribution — name-by-name attribution on a broad-rally day is too costly, so this issue only attributed the branches that entered the theme judgment.
9. No sub-agents used: this issue is pure data reconciliation plus statistical testing, every conclusion is reproducible by local scripts, and the scripts remain in work_ashare0916r/ (pull1–10.py, an1–4.py). risk-auditor was not run — given the pre-market record's lesson that "QA started too late", note here that this issue's main conclusions (hit rate, Spearman, base rates, branch strength) are all deterministic computations already validated across two sources, but the tomorrow-facing judgments in §6 have not passed independent QA, which is this issue's largest un-gated item.

---

⚠️ Risk disclaimer: this recap is post-close information review and observation only and does not constitute investment advice. Data may differ in timeliness or caliber; please defer to exchange and company disclosures, and do not use this directly as a basis for trading.

Data sources: exchange-caliber limit-up / limit-down / failed-limit pools (Eastmoney push2ex, validated with tc == len(pool): ZT 89, DT 4, ZB 11) | whole-market snapshot and industry/concept fund flows (Eastmoney push2delay, 5,915/5,915 with no pagination gap) | multi-day fund-flow series (Eastmoney push2his, last date asserted as 2026-09-16) | individual quotes, opening prices and standard volume ratios (Tencent Finance qt.gtimg.cn, 2026-09-16 close-of-session settlement values) | pre-market list reconciliation baseline: this repository's reports/a-share/2026-09-16.md