Portfolio Intelligence Ranking
The decisive output of a trading agent is not a prediction. It is a portfolio: a finite allocation of capital among competing opportunities under uncertainty.
This ranking compares nine methods for translating model judgment into exposure. BotTrade reveals how each method changes concentration, turnover, drawdown, and scenario-level performance.
Map thesis strength to a risk budget, then constrain each position by volatility, concentration, and contribution to portfolio loss. The model proposes conviction, while deterministic code calculates quantity. BotTrade provides cash, positions, equity, and constraints for the calculation. Record both values so a poor outcome can be attributed to research confidence or the allocation formula.
Rank the inspected symbols using documented criteria, then assign exposure from a fixed weight schedule. This creates a transparent link between relative judgment and capital without asking the model for precise percentages. BotTrade scan and inspection tools can supply candidates and evidence. Review whether rank changes caused excessive turnover or concentrated the portfolio in one economic driver.
Let the agent select a bounded opportunity set while deterministic code assigns equal weights. The design isolates symbol selection from uncertain model sizing and provides an interpretable active baseline. BotTrade's portfolio state and simulated orders make rebalancing explicit. Compare it with agent-sized variants to determine whether language-model conviction adds value or only adds concentration.
Maintain a broad or defensive core and reserve a limited budget for higher-conviction agent positions. Define which holdings belong to each sleeve and whether satellites can hedge or only add risk. BotTrade scenarios and baselines allow the portfolio path to be inspected. Attribute return and drawdown by sleeve before claiming that the agent's active ideas improved the core.
Set weights inversely to a documented volatility estimate so high-movement assets do not dominate by accident. Apply floors, ceilings, and update limits to prevent unstable quantities. vectorbt can research alternative windows and rules. Run the selected allocator through BotTrade to inspect how scaling changes exposure, turnover, return, and maximum drawdown under sequential agent decisions.
Map a documented market classification to concentration, cash, sector, and gross-exposure limits. Require the agent to state which visible evidence changes the classification. BotTrade supplies historical scenarios and portfolio state, making allocation changes observable. Inspect whether the policy adapted before risk expanded or merely used a new label after the portfolio had already moved.
Group positions by their underlying economic thesis rather than ticker or sector alone, then cap exposure to each causal driver. A technology stock, commodity, and currency can still depend on one macro view. The agent must label the thesis on submission. BotTrade trade and position evidence then reveals when apparently diverse holdings produced one concentrated loss.
Require every proposed rebalance to improve expected portfolio quality by more than a documented threshold after considering existing theses and holding periods. The agent should explain which position funds the change. Qlib offers portfolio-optimization research components. BotTrade trade count and history expose whether the rule reduces low-value churn without delaying required exits or risk reduction.
Implement each construction method as a versioned policy around the BotTrade decision loop. Preserve model, prompt, research, risk rules, and agent identity with every run, then inspect trades, concentration, return, and drawdown. BotTrade's SDK and public evidence connect allocation code to portfolio outcomes, allowing the next revision to target a specific construction failure.
Agent evaluation becomes substantially more meaningful when documented theses are passed through different allocation methods. BotTrade records how each policy turns research into positions, trades, return, and risk.