UPQ · Market data
Historical bars, option chains and Greeks for the agent and broker simulator.
Market data guide ↗01 / Task design
Trading tasks
Three tasks cover portfolio management, earnings announcements and same-day expiry. Each includes a recorded example.
02 / Agent design
Analysis and execution
The analysis agent proposes a trade; the execution agent selects contracts and quantities. AVSR summarizes volatility, while QOCR groups contracts by expiry and moneyness.
Loading records…
03 / QFinZero infrastructure
QFinZero infrastructure
LiveOption uses QFinZero for market data, news and events, and simulated brokerage.
LiveOption
Tasks and agents
Task rules, model inputs, decisions and evaluation.
QFinZeroService hub · REST / Python APIs
UPQMarket dataESPNews & eventsPMBPaper brokerMarket and account data go to the agent. Order proposals are passed to the simulator, which records fills and account changes.
ESP · News and events
News and earnings calendars queried as of the simulation time.
News and events reference ↗PMB · Broker simulation
Accounts, orders, fills and margin checks for simulated trading.
Trading guide ↗04 / Results
Evaluation results
Returns, drawdowns and trading behaviour across eight models. Select a model for its metrics; table columns can be sorted.
Loading benchmark figures…
Some models received different inputs, including news. Account for these differences when comparing results.
Supporting resultsResult tablesExpand for metrics and sorting.
Click a column header to sort. Mean returns include 95% bootstrap intervals from 10,000 resamples.
Input differences GPT-5.6-Terra and Gemini-3-Flash used a separate gateway without news headlines in the earnings input. The other six models received headlines. This difference limits model comparisons.