Stop Guessing Which Sector Is Bleeding — Type the Symptom and See Who Raises a Hand

The last two pieces were about not losing money — how to read the auditor's opinion and where the cash actually went

This one runs the other way. It is about seeing it first

The problem with news: it already happened

The sequence is the same every time.

A company hits a problem → the company writes it up in the quarterly filing → months later an analyst has enough to work with → the research note goes out → the media compresses it into a story → you read the story

By the time it reaches the last box, quarters have gone by and the price has done its work.

That is not the analyst's fault and it is not the media's. It is what synthesis is — synthesis has to wait until there is enough to synthesize, and "enough" means several companies saying the same thing first.

Here is the part worth holding onto: if the precondition for synthesis is "several companies saying the same thing," then by the time the story runs, the raw material has been sitting there for months. Nobody had pulled it together for you.

I hit this the first time building an early-warning system on the credit side of a bank (the long version here) — the numbers that land in the statements are finished business. The language management uses to explain why the numbers look that way arrives first, every time.

The institutional method: listen for the complaints that land together

An institutional analyst covering a sector does not read one company and conclude. They read the whole sector, hunting one thing:

How many companies are complaining about the same thing without having coordinated it

The logic is plain. One company says sales fell because buyers could not get approved — that may be that one company. Weak product, bad location, or an excuse. But when companies 2, 5 and 9 write the same sentence in the same quarter, each drafting its own filing on its own — that is not a company problem. That is an industry problem.

It outweighs news on 2 counts:

1. It comes first — filing deadlines are statutory. Nobody has to show up with a microphone. 2. Lying has a price — every sentence sits in a document filed with the SEC (Thailand's Securities and Exchange Commission). An interview carries no such exposure.

One change to the institutional method — and it is the better version

The institution starts from "I am covering the property sector" and then grinds through the 30 companies in it.

The flaw: you have to guess the right sector before you start. If contractors never crossed your mind, you will never see contractors bleeding.

My version runs backwards: do not pick a sector. Pick a symptom.

Type the symptom you suspect, one phrase — "ยอดปฏิเสธสินเชื่อ" [the banks are turning buyers down], "ต้นทุนค่าระวางเรือ" [freight is eating the margin], "กำลังซื้อชะลอตัว" [the customer stopped spending] — and let the system sweep the management discussion and analysis of the entire market and line the hits up. Then watch who raises a hand.

You do not need to know who is hurting in advance. The list that comes back is the answer, and it usually carries names you never would have opened.

The live case: one phrase, and the whole industry raises its hand

I typed this Thai sentence once — "ยอดปฏิเสธสินเชื่อจากธนาคารสูงขึ้นกระทบยอดขายบ้าน" [bank rejections are climbing and home sales are wearing it] — no sector, no ticker.

This is what came back, ordered by fiscal year:

terminal.boomleverage.com

“ยอดปฏิเสธสินเชื่อจากธนาคารสูงขึ้นกระทบยอดขายบ้าน” [rejections up, home sales wearing it]

One query, no sector filter · results ordered by fiscal year · run 15 Aug 2026

• FY FY2022UV — “increasing mortgage rejection rate…”
• FY FY2023PSH · BLESS — “Housing loan rejection increased by over 60%”
• FY FY2024FPT · KUN · EVER · THANA · BPS — 5 filings, same complaint
• FY FY2025JAK — “…improved homebuyers access to mortgage financing”
• FY FY2026TEKA · NCH — still bleeding below ฿3m

Before you read the results, stop on what just happened

The phrase I typed was "ยอดปฏิเสธสินเชื่อ" [the bank said no].

What came back was "loan rejection rate", "tightening of mortgage loan approvals", "higher rejection rate of housing loans" and "banks being cautious"

The phrase "ยอดปฏิเสธสินเชื่อ" [the bank said no] appears in none of these documents — the filings lodged with the SEC are in English throughout. Ctrl+F that phrase across all 150 PDFs and you get zero hits in every file.

This is not keyword search. It is semantic search.

You type Thai. The system goes and finds the same meaning in English filings across the whole market — in 1.2 seconds

And it is not word-for-word matching. Look at the sentence it pulled from EVER: "…due to high household debt and banks being cautious" — the word rejection is nowhere in that sentence. Different words, different structure, same fact, and the system found it, because what it compares is meaning, not characters (the mechanism behind it, here)

This is why you do not need the English accounting vocabulary, do not have to guess which words a company chose, and do not have to translate anything — type what you suspect, in the language you actually think in.

Then read the results as a timeline

Read it as a timeline and you get what no single research note can give you:

FY2022 — the first voice UV wrote "increasing mortgage rejection rate following the concern over elevated household debt and lower consumer debt repayment capability"

FY2023 — the second voice, and this one has a number PSH said it flat: "Housing loan rejection increased by over 60%, reflecting the weak purchasing power among real demand consumers in the country." · BLESS, same year: "A higher rejection rate of housing loans by financial institutions contributed to the decline in the company's perceived revenues"

FY2024 — it becomes a chorus FPT: "…resulting in rising loan rejection rates and intense competition" · KUN: "tightening of mortgage loan approvals by banks, particularly for homes priced below 5 million baht, which faced a high rejection rate" · EVER, THANA and BPS wrote versions of the same thing

FY2025 — the voice that turns JAK: "The reduction in government policy interest rates since late 2025 improved homebuyers access to mortgage financing, positively impacting sales."

FY2026 — not finished, and more specific TEKA: "The mass-market residential segment and mid- to lower-end condominium market were materially affected by elevated loan rejection rates" · NCH: "particularly in the residential segment priced below THB 3 million"

Three things one query just handed you for free

1. A timeline The problem shows up as a single voice (FY2022), widens into a chorus (FY2024), then starts to reverse (FY2025) — two fiscal years separate the first voice from the chorus. That gap is what the filing readers see and the news readers do not.

2. The perimeter of the damage Every one of them names the same price band — below ฿3m (EVER, NCH) and below ฿5m (KUN). The problem is not spread across the sector. It is concentrated at the bottom of the market. That turns "is property in trouble" into "which company sells houses at what price" — a question you can answer, and trade.

3. The turn JAK in FY2025 is the first to say it is improving — and it came out of the exact same query. You do not need a new question to find the inflection. It surfaces in the same result set.

What this is not: this is not a recommendation to buy or sell any property name, and the example above is a set of search results, not a census — the system returns the 100 closest matches, not every match in the market. A company that does not appear is not a company that stayed quiet.

Why retail cannot do this by hand

Try running the above manually.

First you have to know how many property companies are listed, then pull the Form 56-1 (One Report) for every one of them, five years back — call it 30 companies, so 150 files, 200–300 pages each. Then open them one at a time, find the management discussion and analysis section, read for any mention of loan rejections, and log who said it in which year.

And after all 150 files, you still miss TEKA and JAK — if you did not count those two as property names at the outset, you never opened their files.

That is the real cost of starting from a sector: you only see inside the box you drew first.

The one phrase I typed has no box. It sweeps the whole market and lets the data name the list.

The tool I built to do this

Query ideas — with measured counts of how many companies each one sweeps

I ran these for real on the production server on 15 Aug 2026. The "Companies surfaced" figure is the count of distinct companies in the first 100 results — type them in as they are.

Read this table correctly: the right-hand number is not the total number of companies in the market discussing that topic. It is the count of distinct companies inside the 100 closest matches — a floor, not a ceiling. I report it this way because it is what was actually measured, not because it sounds better.

Four rules that make it work

1. Type a symptom, not an industry name — "ยอดปฏิเสธสินเชื่อ" beats "หุ้นอสังหาฯ", because the second one only returns companies that happened to write the word real estate.

2. Read the list of names before you read the text — the names you did not expect are the most valuable part of the output.

3. Sort by year, not by score — what you are hunting is one voice → a chorus → the turn, and that only reads in chronological order.

4. Open the source every time you find something — context can flip a whole sentence. A company discussing loan rejections may be telling you it is the one benefiting.

Limits — what this tool does not do

1. No sector filter in the retail packs, and that is a decision, not an oversight — on the search page you choose between the whole market and SET50, and nothing else. Sector-by-sector sweeps with dataset export sit in the institutional pack, because that is a different job at a different resource level · for the kind of question in this article, not filtering is the advantage, for the reason above

2. It answers "who said it," not "is it true" — a company can blame external conditions to bury a problem of its own. Five companies complaining about the same thing means the industry has its attention there. It does not mean everyone's explanation is correct.

3. Search results are not a census — 100 results per query, hard ceiling.

4. Coverage — 916 companies · FY2021–FY2026 · the pre-FY2021 archive is still being built in the back end and is not open to search yet.

5. Not investment advice — the companies named in this article are here because they were the real output of a real query, shown to demonstrate how to read, not to point at a security.

Run your own query

Open the Boom Leverage Terminal and type the symptom you suspect is running through the economy right now — plain Thai, no translation, no ticker required.

The question I would run first: the thing you suspect is happening but have not seen written up anywhere — then count how many companies have already put it in a filing.

Log in today and the DELTA pack is free through 30 Sep 2026 — a full four years of history (deep enough to see the timeline in the property table above) · 150 credits/day · no card required. Billing is switched off entirely for the duration, and when it ends the account reverts to the free pack on its own with nothing charged — why I am giving it away, and the full terms · for a straight pack comparison, read the pack selection guide · teams and institutions that need full-sector sweeps and dataset export, talk to us on the Enterprise page or contact@boomleverage.com

Further reading: mining MD&A for alpha as alternative data · how semantic search differs from Ctrl+F · screening your portfolio and exporting to Excel

Disclaimer: this article is for education and to explain how to read documents that listed companies disclose publicly. It is not investment advice, it does not point at any security, and it guarantees no outcome. Every company statement is quoted from filings lodged with the SEC, with the fiscal year given so it can be checked · a company appearing in a search result is not an indication that the company has a problem — many of these are ordinary descriptions of industry conditions, and some companies discuss the same subject as something they benefit from · the company counts and response times were measured on 15 Aug 2026 against an index built 14 Aug 2026, and both change as new documents enter the archive · investing carries risk. Past results do not guarantee future performance