Adaptive Conditional Probabilities for Short-Horizon Binary Markets
Summary
The article develops a probability-based framework for BTC 15-minute Up/Down contracts. It treats the target as the conditional probability of settling above the round’s official reference price, then compares that estimate with executable contract prices after accounting for order-book depth, fees, and slippage. The probability baseline uses standardized distance from the reference, blended realized volatility across several horizons, a shrunk short-term drift estimate, and lower confidence when early-round information is sparse. The system discounts the center probability for uncertainty before evaluating trades.
A cited public study with 43 features reportedly failed to beat the market-implied probability consistently out of sample, and costs erased any apparent edge. The article’s own example shows how uncertainty adjustments and execution checks can reject a tempting early trade. It also describes persistent-EV entry confirmation, liquidity-aware sizing and exits, and simulated logging. These are design choices rather than proof of profitability: the author calls the system an exploratory research platform and says it should remain focused on simulation and data collection until staged out-of-sample calibration and executable returns support it.
Key ideas
- Model the probability of the contract’s settlement event conditional on current market state, rather than predicting direction alone.
- Keep the probability inputs independent of the contract order book used to evaluate executable prices.
- Blend volatility across horizons and shrink drift so short samples do not create false certainty.
- Discount estimated probabilities for uncertainty, especially early in a market round.
- Evaluate depth, costs, persistent evidence, and position risk before turning a probability edge into an order.
- The described framework remains unvalidated as a profitable strategy and requires out-of-sample testing.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.