Came across a post on reddit from "woztrades" digging into whether market making can actually be an edge on Kalshi's 15-minute BTC markets. Sharing it because the methodology is more honest than most "look at my returns" posts.
The core question they set out to answer: can post-only resting orders earn enough from spread capture plus maker rebates to survive adverse selection? Instead of cherry-picking one lucky config, they ran 100 systematically varied simulations (via TurbineFi), sweeping quote width, order depth, refresh behavior, position size, price bounds, and loop timing — specifically to see whether results held across the whole parameter region or hinged on one magic setting.
Here's the interesting part, all 100 variants finished green.
Best variant; a post-only spread-capture bot placing resting orders around the BTC 15-min market, only quoting when there's at least a 0.02 spread, 3 order levels, 15-second refresh loop, refresh-on-fill enabled:
- ROI: 1698.40%
- P&L: +$254.76
- Win rate: 99.02%
- Trades: 37,917
- Max drawdown: -$10.35
Weakest variant: more conservative, tighter price band (15c–70c), max 5 contracts:
- ROI: 838.40%
- P&L: +$41.92
- Win rate: 96.57%
- Trades: 9,915
- Max drawdown: $3.46
**Why I think it's worth attention:** the fact that 100/100 finished green isn't the flex — it's the warning. woztrades calls this out themselves. When *every* variant prints, the result is usually being driven by an assumption, not by skill. Here that assumption is maker rebates plus spread capture. The absolute P&L is also tiny ($40–$255), so those huge ROI numbers are leverage on a small base, not life-changing money.
And to their credit, they're upfront about what could break it: tiny execution differences, queue priority, fees, stale quotes, or rebate eligibility could all flip the story fast. Backtests assume your resting order gets filled and earns the rebate, real queues don't always cooperate. This is historical simulation only, not live trading.
**The honest takeaway:** this is a well-structured exploration of an idea, not proof of a money printer. The value is in the framing — sweeping a whole parameter region instead of trusting one run — not in the eye-popping ROI. Anyone tempted to deploy this should assume real-world fills and fees eat most of the edge, and treat the backtest as a hypothesis to stress-test, not a strategy to copy.
Posting it here because it's the kind of thoughtful, caveat-forward analysis the space needs more of. They even asked whether others have had real success with automated MM on these markets — so if you've tried it live, that's a conversation worth having.
Not investment advice. Backtests can be wrong or incomplete.
Credit: original post by **woztrades** on Reddit.
LINK:https://www.reddit.com/r/PredictionsMarkets/comments/1ufrhi4/comment/oty6kkh/?context=3