Home / Guides / What we tested that did not improve the odds on 15-minute Bitcoin contracts
What we tested that did not improve the odds on 15-minute Bitcoin contracts
Most chart ideas feel convincing and do nothing measurable. We tested the common ones against the plain margin model and published what happened, including the nulls.
The standard we used
The baseline uses only two things: the margin and the minutes left. For each new idea we added it to that baseline, refit, and scored both versions on days the model had never seen (the final third of the data). We resampled whole windows, so one noisy window could not fake a result, and we used a wide 99% interval because several ideas were tested at once. An idea counts only if it clearly improves the probabilities.
What we found
| Idea | Result |
|---|---|
| Chart patterns such as double tops and falling steps | Not reliable. Carry names them for reference and never lets them change a call. |
| Price running to a tested support or resistance with the strike just beyond it | No measurable effect. Over 30 days, flagged calls held 85.9%, 90.2% and 95.2% in the three margin bands with enough data, against 85.3%, 91.2% and 95.0% for the same margins without the flag. |
| A tested level sitting between price and the strike | No evidence it helps. |
| Speed of the last 5 minutes toward or away from the strike | No evidence it helps. |
| Which way volume went in the last 5 minutes | No evidence it helps. |
| Volatility speeding up or slowing down | Closest to a signal, but inside the noise. |
| How often price crossed the strike this window | No evidence it helps. |
| Time of day (Asian, European, US hours) | No evidence it helps. |
How good are the calls we do make?
On the test days, 2,871 windows were scored. In 1,734 of them the model reached a 90% call at some point, and 92.6% of those finished on the called side. In 587 windows it reached 90% with 8 or more minutes left, and 90.5% of those held. Calls labelled 90% or better were right 92.6% of the time, so the stated odds were not overstated.
Why Carry still shows the reading aids
Support and resistance, trend, pace and the rest are on the chart because they help you see what the market is doing and decide how much to risk. They do not change the call, and they are not a reason to trade a spot Carry is not confirming.
The honest limits
- "No evidence" is not proof of no effect. It means any effect is too small to see in 90 days.
- Bitcoin behaves differently across periods. We re-run these tests as new data arrives.
- Strikes in the test were proxies built from price bars, so a few windows near the strike may be labelled differently from Kalshi's own.