Trade Decision Policy

Purpose

Use this skill as the final decision-policy layer for Vietnam equity recommendations. It converts upstream evidence into structured BUY, ADD, HOLD, TRIM, or EXIT labels and portfolio action drafts.

Scope

• Vietnam listed equities.
• Portfolio action drafts.
• Target weights and maximum weights.
• Confidence, invalidation, and data-gap disclosure.
• User approval requirement before target-state mutation.

Non-goals

• Do not place broker orders.
• Do not call broker APIs.
• Do not execute trades.
• Do not treat an approved target portfolio state as an executable order.
• Do not update approved target portfolio state without explicit user approval.

Input Contract

Required inputs:

• ticker
• exchange
• research_summary
• valuation_view
• earnings_quality_view
• portfolio_risk_view
• current_weight_pct
• risk_budget
• compliance_context

Optional inputs:

• liquidity_view
• macro_view
• news_view
• backtest_view
• data_gaps

Decision Labels

• BUY: initiate a new position when evidence is strong, valuation is attractive, risk budget allows it, and liquidity is acceptable.
• ADD: increase an existing position when the thesis remains valid and target weight remains within risk limits.
• HOLD: keep target weight unchanged when evidence is balanced or action is not justified.
• TRIM: reduce target weight when risk, valuation, concentration, or thesis quality has weakened.
• EXIT: reduce target weight to zero when the thesis is invalidated or compliance/risk constraints require removal.

Required Output

Return a structured object with:

• ticker
• exchange
• decision_label
• current_weight_pct
• target_weight_pct
• max_weight_pct
• confidence
• reason
• invalidation_condition
• risk_check_status
• compliance_check_status
• data_gaps
• requires_user_confirmation
• broker_execution_allowed

requires_user_confirmation must be true.

broker_execution_allowed must be false unless a separate future execution policy explicitly enables broker integration.

Guardrails

• Separate facts, assumptions, and inference.
• Cap confidence at Medium when critical financial statements, liquidity data, or portfolio weights are missing.
• Return HOLD when evidence is insufficient for an active change.
• Return TRIM or EXIT only with explicit risk, valuation, compliance, or thesis-invalidation reasoning.
• Never imply that a recommendation is a real order.
• Never mutate D1 approved target state directly; return an action draft for approval-gated governance processing.