Pricing Kalshi 15 Minute BTC Markets with the Vol Smile
How I used the options market's volatility smile to price Kalshi's 15 minute BTC Up/Down contracts, and what happened when the model met live markets.
By HYHY.APP ·
Where did the idea come from?
Back in April, I had a thought. The 15 minute BTC Up/Down market on Kalshi is really just a short-term options market. If I could observe the implied volatility the options market is pricing, could I use it to price the contract?
The theory was that a model built on the vol smile should lead a market that is still pricing mostly on sentiment and momentum.
How did I test it?
I started by researching data availability and decided to proceed. I priced a few points by hand, and the results were surprisingly good. That sent me deep down the rabbit hole.
The model uses the vol smile as its anchor, which is more precise than relying on a single volatility number. On top of that, it adjusts for three things:
- Short-term order flow
- Spot price trend
- The psychology you see in retail-driven markets, such as momentum chasing and anchoring near round numbers
What does it look like when the model leads?
Here is Kalshi's 15 minute BTC market on Sep 11 at 11:00 am.
The blue line is the model price, the green line is the market, and the orange line is spot. The model pushed toward resolution while the market was still catching up, and spot had not even confirmed the move yet.
The distance from the target was not extreme compared with how most other session markets moved. Even so, the model called the finality of the outcome ahead of the market.
This supported my original idea. The options market does embed information that you cannot get from spot and price action alone.
What does it look like when the market is already efficient?
It does not always lead, and that is by design. When the options market is relatively quiet, the model price should hug the market.
Here is Kalshi's 15 minute BTC market on Sunday, Sep 6 at 4:15 am.
This was a different session with different behavior. The market was already pricing efficiently, so the model confirmed it rather than led it.
What comes next?
I have been running the pricer since June and have observed it continuously since. The next step was building an execution bot for myself, and that turned out to be an adventure of its own. More on that soon.