Home Leaderboard Dashboard

Hong Kong Trades Are Now More Accurate Than Ever: We Plugged the Engine Straight Into the Source That Settles the Market

We just shipped the biggest accuracy upgrade in WeatherBot's history for one city — and it is a game-changer. Hong Kong temperature markets on Polymarket settle on one number: the official daily maximum recorded by the Hong Kong Observatory (HKO). So we wired our engine directly into HKO. It is now the dominant forecast model for Hong Kong, it feeds a live hour-by-hour reading of the exact gauge that decides every market, and it powers a brand-new resolution-day edge that is about as close to a sure thing as this game gets. Here is everything we added, why it wins, and the math behind the profit.

Every forecast model we have ever used — GFS, ECMWF, UKMO, NWS — predicts the temperature for a grid cell floating somewhere near a city. That is fine when nothing better exists. But for Hong Kong, something better does exist, and it is the same thing Polymarket uses to resolve the market: the official gauge at the Hong Kong Observatory headquarters. If you are forecasting a grid cell but the market pays out on a specific instrument, you are predicting the wrong number. We closed that gap completely.

What we added — three upgrades in one

This release is three connected upgrades that all draw from the same official source. Together they make Hong Kong the most precisely-modeled city on the entire platform.

50%
HKO weight in the HK consensus
the dominant model
~90%
HKO day 1-3 forecast accuracy
official, published figure
Hourly
Live running-max polling
on resolution day

1. HKO is now the dominant forecast model for Hong Kong. We pull HKO's official 9-day forecast and give it a 50% weight in the consensus — more than all three global models combined. ECMWF, UKMO and GFS stay in as a sanity check, with GFS deliberately down-weighted because it runs warm in the subtropics. The HKO weight also tapers with how far out the market resolves, matching HKO's published accuracy curve (about 90% at 1-3 days, 85% at 4-7 days, 80% at 8-9 days).

2. A live, hour-by-hour read of the exact resolution gauge. On the day a market resolves, we poll the HKO headquarters temperature every few minutes and track the running daily maximum. This is the literal number the market will settle on — not a forecast of it.

3. A resolution-day edge engine. Because the daily maximum can only ever rise, the moment the live reading crosses a "X° or higher" threshold, that outcome is mathematically locked to win. The bot buys it while the market is still mispricing it. After the afternoon peak passes, a second play buys the bucket the day's high actually landed in.

Why HKO dominates the other models

This is not about HKO being a "slightly better" forecaster. It has a structural advantage no global model can ever match: it is the gauge that defines the answer. A global model has to forecast both the weather and the offset between its grid cell and the official station. HKO only has to forecast its own station — and on resolution day, it doesn't forecast at all, it simply reports.

Global models
GFS · ECMWF · UKMO
HKO
Hong Kong Observatory
What it predictsA grid cell near the cityThe exact HQ gauge the market settles on
Station errorUnknown, persistent offsetZero — it is the resolution gauge
Day 1-3 accuracyGood~90% (official)
Live intraday readingNoYes — updates hourly
Weight for Hong KongSecondary (sanity check)Dominant (50%)

The edge: near-riskless money on resolution day

Here is the part that gets us excited. A daily maximum temperature can only go up as the day heats up — it can never go back down. So the instant the live HKO reading hits, say, 33°, the market outcome "33° or higher" is locked to resolve YES. It is decided. If slower traders are still pricing that outcome at 70¢, the bot buys it and waits for it to settle at $1.00.

A winning share pays out $1.00, so if you buy a locked outcome at price P, your profit per share is $1 − P and your return on capital is (1 − P) / P. The cheaper the locked outcome is still trading, the bigger the edge:

Buy price of a locked outcome Profit / share Return on capital
35¢65¢+186%
50¢50¢+100%
65¢35¢+54%
75¢25¢+33%
83¢ (our ceiling)17¢+20.5%

We cap the buy price at 83¢ — above that, the gap to $1.00 no longer clears the cost of trading. And we set a 35¢ floor as a safety switch: if an outcome looks locked but is somehow trading that cheap, we trust the market over our own read and skip it, because it probably means a data mismatch. Even at the worst price we will pay, a locked Hong Kong trade returns about 20% on capital — and the bot always grabs the cheapest locked outcome first, so real fills are usually far better.

The second play: after the peak has passed

Hong Kong's temperature peaks in the early-to-mid afternoon, then falls. Once it is evening and the live reading has dropped a clear margin off the day's high, the maximum is effectively locked in. At that point the bot buys the temperature bucket the day actually landed in — the classic "it is not getting any hotter today" trade — while everything above it becomes a near-certain NO. It is the mirror image of the locked-YES play, and it captures a whole second window of mispricing every single day.

See it live on your dashboard

HKO is now a first-class data source in the WeatherBot dashboard, sitting right alongside GFS, ECMWF, UKMO, NWS and NCEI — with its own live status, a per-market temperature readout for Hong Kong, and a real-time display of the official running maximum as the day unfolds. When you open the app on a Hong Kong resolution day, you are watching the same number the market will settle on, tick by tick.

Near-zero
Risk on a locked trade
the outcome is already decided
Daily windows of edge
morning lock + afternoon settle

Hong Kong has always been one of WeatherBot's top-earning markets. With this upgrade, it is now also our most accurate — modeled directly off the source of truth, watched live to the hour, and traded with an edge that is, on locked outcomes, about as certain as prediction markets allow. If you trade Hong Kong weather, today is a very good day.

As always: WeatherBot is a tool, not financial advice. Markets carry risk, "locked" assumes clean resolution, and past performance does not guarantee future results. Trade responsibly.

Read next

WeatherBot by the Numbers: 9,592 Traders, $2.87M in Member Profit, and Exactly What It Costs to Run

Read article →

We Banned 213 Accounts for Engine Abuse — Here Is Exactly What Happened, and Why

Read article →

Introducing 20TradesStop — the set-and-forget mode we'd actually pick ourselves

Read article →
← All articles