Sub-$1 RMSE was achieved over specific test periods, not consistently 24/7.
The next challenge is execution: can the timing advantage survive Polymarket spreads, fees, and order latency?
Still testing. Curious if anyone else is researching this.
Built a public Chainlink BTC/USD prediction dashboard.
Using data from 7 exchanges, it estimates the index ~1s before the official feed. Sub-$1 RMSE over some test periods.
Built for Polymarket research.
https://t.co/AZ4QIahAVn
#Polymarket#Chainlink
For Polymarket 5-min markets, the settlement price is based on a 60-second TWAP.
The dashboard tracks both the reconstructed Chainlink price and TWAP, including Price to Beat and projected settlement.
The goal is to understand the edge before trying to trade it.
The model reconstructs the Chainlink BTC/USD index using real-time bid/ask data from 7 exchanges.
The challenge isn't just matching the price, but getting the timing right.
Even small differences in feed latency can affect accuracy.
@Probcall Not yet. So far I've been focused on reconstructing the Chainlink index as accurately and early as possible. Next step is testing actual execution on Polymarket, including latency, spreads, and fees, to see if the edge is really tradable.
Anyone still trading Polymarket 5-min markets?
Finally built a model that predicts the Chainlink BTC/USD index 1+ second ahead using live data from 7 exchanges.
Achieved sub-$1 RMSE over certain test periods.
Now testing whether that edge is actually tradable on Polymarket.
@64s Weekends usually mean thinner liquidity, making prices easier to manipulate. I wouldn't be surprised if that's what's happening here—possibly by traders who understand Chainlink's index methodology better than we do.
Just a hypothesis for now, but worth investigating.
@64s Noticed the same anomaly: Chainlink jumps $10–20 while CEX top-of-book prices barely move. I suspect order book depth/liquidity may affect the index, not just bid/ask midpoints.
Started collecting L2 books and trades across exchanges to test it. Curious what you think.
@karlbooklover Great write-up! I'm building a similar 7-exchange BTC/USD model. Volume weights + 1s smoothing + rolling Chainlink bias still give me ~$2.3 RMSE on held-out data.
Were your weights fitted or volume-based? Any per-exchange lag/smoothing? How often did you update bias?
@64s Impressive BTC dashboard! I've spent days trying to replicate your 7-feed forecast. Price moves track closely, but my RMSE vs Chainlink is higher.
Not asking for the formula—are weights fitted or liquidity-based? Is temporal smoothing involved? Any other inputs?