HK Insider-Signal Picks — Post-Mortem & Playbook v2 (Oct 2026)
Why our best Buy calls worked, why the worst failed, what predicts returns across all 108 picks, and the rules we changed · 8 Oct 2026
1. The winners: mostly not for our reasons
| Pick | Return | What actually drove it | Verdict |
|---|---|---|---|
| 3636 Jinxun Resources Spec Buy · 13 Jul | +44% | Outside fund bought into the post-IPO lock-up washout. Positive profit alert, then H1 profit +71% on LME copper (our named catalyst). Interim dividend; company buybacks. | Right, right reasons |
| 0434 Boyaa Interactive Spec Buy · 20 Jul | +43% | The BTC-NAV discount thesis held. BTC +26%, gaming beat our "declining" view (+17% revenue), and the controlling camp kept buying. Mainly a BTC factor bet. | Right, with beta |
| 3662 Starjoy Wellness Spec Buy · 5 Jun | +38% | 0.15× book. The buyer kept adding (9.9% → 13.1%), a director bought too, and the dividend was reinstated. Results were flat; persistent insider buying drove the re-rate. | Right, right reasons |
| 0065 Grand Ocean Spec Buy · 30 Jul | +26% | Published with fair value (0.25) below the live price (0.27) because we used the stale signal price. The gain came from a coal-mine MOU and a 7-for-1 rights issue, the dilution risk we had flagged. | Luck |
| 1093 CSPC Pharma Spec Buy · 1 Jun | +25% | The founder's HK$283M buy was informative, but our DCF put fair value 10% below the price. The real driver was a US$1.2bn AstraZeneca upfront payment the day after the buy, which our report never mentioned, then H1 profit +139%. | Right stock, wrong reason |
| 1288 Agricultural Bank Spec Buy · 6 Jul | +24% | The whole Big-4 bank sector rose 18–25% on rotation into dividend payers. Our fair value was only +10%. | Sector beta |
| 2288 Rykadan Capital Spec Buy · 22 Jul | +22% | One +21% day on about HK$88k of turnover, with no news. | Noise |
| 1962 Evergreen Products Spec Buy · 6 Jul | +17% | About half of the return is dividend. H1 beat our base case (profit +30%), plus a buyback and family buying. | Partly earned |
What the earned wins shared: a hard, marked-to-market asset anchor (NAV, net cash, commodity); a dated catalyst within about 6 weeks (August interims drove most big up-days); and an insider or controlling camp that kept buying after the signal. Buyer type on its own did not distinguish wins from losses.
2. The losers: five repeatable mistakes
| Pick | Return | What happened | Mistake |
|---|---|---|---|
| 0881 Zhongsheng Spec Buy · 3 Jun | −61% | Buyer was Pzena, a US value fund, not an insider. H1 profit −91%, the dividend was cut, and S&P moved to a negative outlook. The 10.6% yield wasn't covered by earnings. | Value trap; fund read as insider |
| 8308 Gudou Spec Buy · 24 May | −58% | We said the latest financials were Q3 2023, but FY2025 results with a going-concern warning had been on HKEX for 2 months. The CEO's buy was only HK$0.8M. The CFO resigned in August. | Missed filings |
| 2889 PATEO Spec Buy · 26 Jun | −54% | The founder bought during the company's own discounted placing (price support). The stock had run from 102 to 289 on AI hype. We used 5.5× sales on a 12% gross-margin business, against peers at 2–3×. Lock-up overhang. | Placing support; hype run-up |
| 0433 North Mining Spec Buy · 20 Aug | −36% | DCF on peak, partly discontinued revenue. H1 profit was mostly a disposal gain. A stale scanner price (0.158 vs 0.058). A 7.7% share deposit to a small broker afterwards. | Cyclical / one-off earnings |
| 6865 Flat Glass Spec Buy · 27 May | −33% | Founder bought near the low, then H1 swung to a loss (gross margin 8%). Book value treated as a floor while ROE was falling. | Trough-catching on hope |
| 1762 Wanka Online Spec Buy · 31 Jul | −30% | Fundamentals were fine (H1 +48%), but we published 45% above the insider's price. The stock fell straight back to it. | Stale price / run-up |
| 0586 Conch Venture Moderate Buy · 10 Jul | −25% | SOE parent buying. We assumed the holding-company discount would narrow, but the operating EV didn't cover net debt. H1 −35% on weak cement. | "Irrational" discount was rational |
| 2155 Morimatsu Moderate Buy · 7 Jul | −25% | Buyer was Fullgoal, a mutual fund. Profit warning a month later; H1 swung to a loss. Our FX risk was backwards. The "bear case" was 4% below the price. | No next-print check |
Why Moderate Buy went 1 of 11
Every Moderate Buy was a HK$11bn+ company where the "insider" was a fund manager, insurer or SOE parent (Artemis, Fullgoal, Ping An, New China Life, Anhui Conch, Jiangsu Communications), buying 0.3–1.5pp. Conviction was raised for buyer size and respectability rather than information or timing. Upside depended on holding-company or SOTP discounts narrowing, or on peak or one-off earnings (Dongfang Electric, United Labs). Bear cases sat only 0–10% below the price.
3. What predicts returns across all 108 picks
Each pre-report signal was measured against the return vs the Hang Seng from the publish date to 8 Oct.
| Signal at report time | Flagged picks | Clean picks | Use in v2 |
|---|---|---|---|
| Live price >20% away from the insider's purchase price | −11.5% (n=22) moved down: −20.6% (n=8) · up: −6.2% (n=14) | +3.8% | Hold cap unless explained |
| Up >25% in the 20 days before the report | −9.1% (n=13) | +2.0% | Spec Buy max |
| Buyer is a fund / insurer / SOE parent | −4.1% (n=20) | +1.7% | Spec Buy max |
| >100% above the 52-week low | −5.2% (n=18) | +1.8% | Hold if also run-up |
| >40% below the 52-week high ("falling knife") | +3.2% (n=47) | −1.3% | Not predictive; blocks Moderate Buy only |
| Token buy (<HK$5M or <0.2% of market cap) | +1.0% (n=33) | +0.5% | Not predictive; blocks Moderate Buy only |
| Micro / penny / GEM / illiquid | +0.4% (n=42) | +0.9% | Not predictive; position-size note |
Also: the 57 candidates we skipped trailed the published ones (median −3.9% vs +1.4% vs the HSI), so the skip step works. But several skipped control changes and takeovers rose sharply (9978 +160% after a mandatory offer; 8131 +82%).
4. What changed in the pipeline (Playbook v2, live from 9 Oct 2026)
- Automatic market check before research. Every candidate now carries the live price, move since the insider bought, 20/60-day momentum, 52-week range, liquidity, market cap, trade size vs market cap, buyer type, and warning flags. Valuation uses the live price, never the stale signal price.
- Red-flag sweep before valuing:
- 120 days of HKEX filings checked for going-concern, default and audit issues;
- placings, rights issues or convertible bonds within ±30 days, lock-up expiry within 90 days, and large share deposits to brokers;
- a web search for major deals (the CSPC lesson);
- a next-print test: if results are due within 60 days, the base case can't need more than 30% improvement on the latest period.
- Earnings discipline: cyclicals valued on mid-cycle margins and a trough-earnings bear case; one-offs stripped out; price-to-sales only above 30% gross margin; book value is not a floor when ROE is falling; holding-company discounts must be tested against net debt.
- Hard rating rules:
- No Buy if base-case fair value is below the live price.
- Speculative Buy needs at least +25% upside on the base case.
- Moderate Buy needs all of: a founder, executive director or controller buying ≥HK$10M or ≥1% of market cap; no flags; a dated catalyst; and a bear case no worse than −20%.
- Avoid/Sell needs a concrete reason.
- Each report shows which caps fired, a catalyst calendar, and "what would prove us wrong".
- Control changes and takeovers are researched as special situations (offer price, spread, deal risk) instead of skipped.
- Scorecard every 6 weeks, comparing v1 and v2 picks.
Rough replay of the mechanical caps on the past Buys: the 10 that would have been capped returned −5.5% vs the HSI against −0.4% for the rest. That's a modest in-sample improvement. The bigger gains should come from the qualitative checks (filings, placings, next print, major news), which would have caught 8308, 2889, 0881, 2155 and CSPC's missing catalyst, but those can't be replayed mechanically.
Method: entry = close on the publish date, Yahoo Finance adjusted (dividend names include dividends), vs ^HSI over the same dates. Post-mortem facts come from HKEX filings, company results and news; items we couldn't verify are omitted or labelled as inference in the working notes. Not investment advice.
No redactions applied.
Published via wiis. Originals are encrypted at rest; this page reflects the redacted version. report abuse.
Discussion
No comments yet.