How MacroTraders.ai Works

Processes hundreds of GDELT geopolitical events through a multi-layer relevance filter to extract market-moving signals: central bank speeches, trade policy shifts, OPEC decisions, earnings themes, and supply disruptions.

Source: GDELT Project — pre-filtered by 16 preferred financial domains

Ingests a structured packet of macro indicators tailored to each asset class: rate differentials and trade balances for FX, earnings growth and valuations for equities, real yields and inflation breakevens for rates, supply/demand balances for commodities.

Sources: FRED (Federal Reserve), ECB SDW, Yahoo Finance

Examines recent OHLCV candles, computing momentum indicators (RSI, MACD), identifying candlestick patterns, support/resistance levels, and trend direction. Provides a purely price-action-based view.

Source: Yahoo Finance (delayed historical candles)

Two adversarial researchers — one bullish, one bearish — engage in a multi-round written debate. Each round, they challenge the other's assumptions, cite specific data points, and refine their probability estimates.

2 rounds of structured adversarial debate per cycle

Evaluates the quality of both arguments, identifies logical fallacies, weighs the strength of evidence cited, and produces a final probability distribution between bull and bear scenarios.

Outputs: bull/bear probabilities + moderator notes

Merges all analyst outputs and debate conclusions into a unified signal: recommended direction, key bullish and bearish factors, risk factors, and a confidence-weighted probability.

Combines 3 analyst views + debate outcome into one signal

Sets portfolio-level risk parameters: alpha (0.3–1.5, controlling risk concentration vs equal-risk) and utilization rate (0–1, fraction of available risk budget to deploy). Cannot override conviction scores.

The only LLM-driven step in the execution chain

Computes the available VaR budget (VaR_disp) by scaling a $100K base allocation based on cumulative P&L. As drawdown increases toward the $1M hard stop-loss, a drawdown scalar progressively reduces the budget to zero — enforcing automatic de-risking.

Formula: VaR_disp = base_VaR × (1 − drawdown%). Hard stop = team eliminated

Converts conviction scores into target notionals using risk-parity optimization. The alpha parameter controls risk concentration: at 0.3 all assets get equal risk budgets, at 1.5 high-conviction assets dominate. The optimizer finds portfolio weights that match the target risk distribution under a covariance matrix.

VaR_eff = VaR_disp × utilization_rate. Notional = VaR_eff / portfolio_volatility

Compares the target portfolio from the optimizer with current live positions and generates executable delta orders. Applies market-convention rounding: FX pairs round to 100,000-unit lots, ETFs and equities round to whole shares. Sub-threshold deltas are filtered as HOLD.

Action types: OPEN, INCREASE, REDUCE, CLOSE, REVERSE, HOLD