Polymarket Bot Tutorial · Chapter 27 of 32
Weather and climate prediction bots on Polymarket: hurricane landfall markets, daily max temperature, El Nino/La Nina (ENSO), NOAA and NWS data sources, and how to convert weather data to trading signals.
What this chapter covers
Weather markets on Polymarket are an underrated category. They have clean public data sources, slow price discovery, and infrequent active traders. The edge for a bot is real but the markets are usually thin. This chapter covers hurricane, temperature, and ENSO markets.
- Weather as a tradeable signal
- Hurricane markets: NHC data
- Daily max temperature: NWS data
- ENSO (El Nino/La Nina) cycles
- Latency: weather updates are slow (good for retail)
- Risk: forecast model error tails
- Code: pull NOAA hurricane data and adjust position
Weather as a tradeable signal
Weather markets are well-served by free, authoritative data sources (NOAA, NWS, NHC) and resolve on objective measurements rather than judgment. That makes them ideal for systematic strategies - the edge is in data interpretation, not in racing humans to news.
The downside: volumes are modest. A hurricane market might do $500k-2M lifetime; a city temperature market $50-200k. Strategies that work at scale on politics or sports don't transfer to weather - the dollar size of your edge is bounded by the market's total liquidity.
Bot pattern that fits: small, diversified positions across many weather markets, hold to resolution. Slow-paced; weather isn't a day-trading market.
Hurricane markets: NHC data
Hurricane season (Atlantic: Jun-Nov) creates Polymarket markets on landfall location, intensity, and named-storm counts. Data: National Hurricane Center (NHC) public advisories every 6 hours during active storms, every 3 hours when a hurricane is <72h from landfall.
Strategy: when NHC's forecast cone implies a specific landfall probability that the market disagrees with, take the side closer to NHC's official forecast. The NHC is the source-of-truth that the market will eventually converge to.
Caveat: long-tail risk. Hurricanes occasionally do things the forecast didn't expect. Size positions assuming the NHC is right 80% of the time, not 100%.
Daily max temperature: NWS data
Polymarket lists daily-temperature-threshold markets for select US cities. "Will NYC reach 95°F on Aug 15?" Data: National Weather Service forecasts updated 2-3 times daily; observations after the fact.
The market typically prices the NWS forecast probability with some noise. The edge: NWS forecasts have biases (typically conservative on extreme heat events). A bot that knows the bias direction for a city/season takes the side the NWS systematically underestimates.
Constraints: low volume ($50-100k typical), small position sizes, hold-to-resolution. Cycle: enter morning-of, resolve evening.
ENSO (El Nino/La Nina) cycles
El Niño / La Niña forecast markets have multi-month horizons and clean data (NOAA monthly ENSO updates). The Polymarket implied probability often lags the NOAA forecast confidence by 1-2 weeks after each monthly update.
Bot pattern: read NOAA's update on release day, take the side that matches NOAA's forecast adjustment, hold for 1-2 weeks until the market catches up. Multiple updates per season offer multiple entry points.
Volume is modest ($100-500k per cycle) but the strategy is slow enough that pure-quant retail can compete against the limited bot competition in this niche.
Latency: weather updates are slow (good for retail)
Weather data updates are minutes-to-hours, not sub-second. This is a meaningful retail advantage: the latency arbs that dominate sports and crypto markets don't apply here.
A retail bot reading NOAA's 8am update at 8:15am can place an FOK at the new fair value before the slower traders in the market have even seen the update. The 15-minute latency budget is generous compared to the 2-second budget on news arb.
The trade-off: thin volume means even a fast bot can only deploy small positions per market. The breadth-not-depth pattern (chapter 21) applies even more strongly to weather.
Risk: forecast model error tails
Weather forecasts have known error bars. NHC publishes their hurricane forecast errors annually - landfall location averages 100-200 miles error at 72-hour lead time. NWS temperature forecasts average 2-4°F error at 7-day lead time.
Implication for sizing: never bet "the forecast is right" with high confidence. Size positions assuming the forecast is right 70-80% of the time. A bot that takes the forecast as gospel loses on the 20-30% of trades where the model was off.
The hurricane category is particularly tail-heavy. A Cat 5 making landfall in a forecast-low-probability location is a positive infinity loss for a confidently-short position. Cap exposure on any single hurricane to 10% of weather allocation.
Code: pull NOAA hurricane data and adjust position
Reference: poll NHC advisory feed during hurricane season, alert on forecast cone changes.
import requests, feedparser
NHC_RSS = "https://www.nhc.noaa.gov/index-at.xml"
def poll_nhc():
while True:
feed = feedparser.parse(NHC_RSS)
for entry in feed.entries:
storm_id = entry.id
advisory = parse_advisory(entry.summary)
prev = load_last_advisory(storm_id)
if advisory["track"] != prev.get("track"):
alert(f"track update for {storm_id}: {advisory['track']}")
save_advisory(storm_id, advisory)
time.sleep(900) # 15 minPolymarket landfall markets are best matched manually to NHC's storm IDs at season start; automating the matching is fragile because Polymarket's market titles don't follow NHC's naming consistently.
