Trading Master

I started this hackathon thinking the hard part would be predicting whether the market was going up or down. Once I started working with options, I realized that was only the beginning. You can be right about direction and still make a bad trade because the structure is wrong, the option is too expensive, the quote is stale, or the risk simply is not worth taking. That changed the way I built Trading Master. A market view does not automatically become a trade. The system still checks the evidence, the options structure, the economics, and finally the risk. If something does not make sense, it can simply say NO_TRADE. One moment really captured the project for me. In a real Alpaca market-data run, the model was bearish on SPY and the system found a bear put debit spread. But before anything could be submitted, it found that the option quote was stale. The risk layer rejected the trade and no order was placed. That was exactly the point: having a market view is not the same thing as having a trade worth taking. I also spent a lot of time on the research itself. My strongest training result reached about 58.8% directional precision with a positive average signed move, but it came from only 17 signals. That was not enough to meet the standard I had set beforehand, so I did not call it validated. I also built a counterfactual layer to understand what actually happened after a decision: was the forecast wrong, was the option structure wrong, was execution the problem, or was refusing the trade actually the better decision? For the hackathon, Trading Master is paper-only, uses defined-risk option structures, and integrates Alpaca market data and Alpaca MCP. The biggest thing I learned is that prediction can be useful without being enough. A forecast is not a trade.