Multi-Agent Macro Analysis for a Leveraged ETF Basket
Code Lumibot
Summary
This LumiBot example uses a team of agents to build a basket from leveraged long and inverse ETFs, with SHV as a cash-like fallback. Separate agents assess growth, inflation and rates, and debt and liquidity. A fourth challenges their conclusions, and a trading agent reconciles the views and places orders. The instructions call for at least three positions in ordinary conditions, favor diversified 3x exposure when evidence supports risk-taking, use 2x exposure to reduce risk, and require explicit downside or hedge evidence for inverse ETFs.
The implementation sets limits on macro-data and news tool use and requires fresh portfolio, position, and exact-symbol price checks before each order. It includes a backtest configuration with a stated trading fee and SPY benchmark, but provides no backtest output or live trading results. The file demonstrates agent roles and workflow constraints rather than establishing that the approach is profitable. Its universe is concentrated in leveraged products, so the code’s portfolio rules do not by themselves establish suitable sizing or control of leverage-related losses.
Key ideas
- Specialist agents assess growth, inflation and rates, and debt and liquidity before recommending leveraged ETFs.
- A disagreement agent challenges the specialists, and a trader agent reconciles their views before trading.
- The strategy calls for diversified holdings and distinguishes between 3x, 2x, inverse, and cash-like exposure.
- Each order requires fresh account, position, and exact-symbol price checks in the same agent run.
- The code configures a backtest but supplies no performance results.
Tags
Full text
# ai_trading_team_ray_dalio_idea_meritocracy_leveraged.py
```py
"""Ray Dalio / Bridgewater-inspired idea-meritocracy AI trading team example.
Leveraged ETF data-on variant. Uses LumiBot's default built-in tools, including
FRED/ALFRED macro tools when FRED_API_KEY is supplied, Alpaca News when
ALPACA_NEWS_API_KEY / ALPACA_NEWS_API_SECRET are supplied, SEC tools, market
state, account state, and order tools. The trade universe is 2x/3x leveraged
ETFs plus SHV/cash as a rare escape hatch.
"""
from lumibot.credentials import IS_BACKTESTING
from lumibot.entities import Asset, TradingFee
from lumibot.strategies.strategy import Strategy
from lumibot.traders import Trader
ORDER_READINESS_RULE = (
"Immediately before every buy or sell order, call account_portfolio, "
"account_positions, and market_last_price for the exact ordered symbol in "
"this same agent run, then call orders_submit_order. LumiBot rejects blind "
"orders with ORDER_READINESS_REQUIRED when those readiness calls are missing."
)
DATA_USAGE_RULE = (
"Use official LumiBot built-in tools by their real names: get_fred_snapshot, "
"get_fred_latest, get_fred_series, and list_fred_series for macro evidence; "
"alpaca_news for recent market, sector, rates, and ETF-proxy headlines; SEC "
"tools only when sector or company fundamentals are relevant. Keep tool use bounded: "
"at most one FRED snapshot or short series request and one Alpaca News call per agent run; "
"set Alpaca News limit <= 5; do not paginate or repeatedly re-check the same evidence. "
"Do not rely on a custom public CSV FRED helper."
)
LEVERAGE_RULE = (
"This is a leveraged ETF strategy. Use only symbols from the leveraged "
"universe plus SHV/cash-like exposure. Stay biased toward diversified 3x "
"exposure when evidence supports risk-taking; use 2x as a risk-down choice; "
"use inverse leveraged ETFs only with explicit downside or hedge evidence. "
"Hold at least three positions in normal conditions, and do not make a single "
"all-in bet. SHV/cash may count as one position when the model refuses every "
"reasonable leveraged setup."
)
class AITradingTeamRayDalioIdeaMeritocracyStrategy(Strategy):
parameters = {
"universe": [
"TQQQ", "QLD", "SQQQ", "QID", "UPRO", "SSO", "SPXU", "SDS",
"UDOW", "DDM", "SDOW", "DXD", "TNA", "UWM", "TZA", "TWM",
"FNGU", "FNGD", "TECL", "TECS", "SOXL", "SOXS", "FAS", "FAZ",
"CURE", "RXD", "LABU", "LABD", "ERX", "ERY", "GUSH", "DRIP",
"TMF", "UBT", "TBT", "TTT", "UGL", "GLL", "AGQ", "ZSL",
"UCO", "SCO", "NUGT", "DUST", "YINN", "YANG", "EDC", "EDZ",
"EURL", "EUO", "YCS", "DRN", "SRS", "SHV",
],
"min_positions": 3,
}
def initialize(self):
self.sleeptime = "1D"
self.agents.create(
name="growth_agent",
model="openai/gpt-6-luna",
reasoning_effort="high",
allow_trading=False,
system_prompt=(
"Argue which leveraged ETFs win if growth improves. First call get_fred_snapshot "
"or get_fred_latest for growth, liquidity, rates, and credit context, then call "
"alpaca_news for broad market and ETF-proxy headlines. Use exact symbols from the "
"universe only. " + DATA_USAGE_RULE + " " + LEVERAGE_RULE
),
)
self.agents.create(
name="inflation_agent",
model="openai/gpt-6-luna",
reasoning_effort="high",
allow_trading=False,
system_prompt=(
"Argue which leveraged ETFs win or lose if inflation and rates surprise. First call "
"get_fred_snapshot or get_fred_latest for CPI, inflation expectations, Treasury yields, "
"and policy-rate context, then call alpaca_news for rates, commodities, and market headlines. "
"Use exact symbols from the universe only. " + DATA_USAGE_RULE + " " + LEVERAGE_RULE
),
)
self.agents.create(
name="debt_liquidity_agent",
model="openai/gpt-6-luna",
reasoning_effort="high",
allow_trading=False,
system_prompt=(
"Argue from debt, liquidity, currency, credit, and policy pressure. First call FRED tools "
"such as get_fred_snapshot/get_fred_latest, then call alpaca_news for market stress and ETF-proxy headlines. "
"Use exact symbols from the universe only. " + DATA_USAGE_RULE + " " + LEVERAGE_RULE
),
)
self.agents.create(
name="thoughtful_disagreement",
model="openai/gpt-6-luna",
reasoning_effort="high",
allow_trading=False,
system_prompt=(
"Challenge all views with thoughtful disagreement. Identify the best diversified leveraged basket after stress testing. "
"Check whether upstream agents actually used get_fred_snapshot/get_fred_latest and alpaca_news. Do not re-call tools unless upstream evidence is entirely absent; if you must, make only one short FRED call and one Alpaca News call with limit <= 5. "
+ DATA_USAGE_RULE + " " + LEVERAGE_RULE
),
)
self.agents.create(
name="trader",
model="openai/gpt-6-luna",
reasoning_effort="high",
allow_trading=True,
system_prompt=(
"Build a Ray Dalio-style idea-meritocracy leveraged ETF basket. This is not All Weather and not a one-ETF momentum bet. "
"Use the specialists' disagreement process, reconcile overrides, and size at least three positions in normal conditions. "
"Favor 3x ETFs for high-conviction sleeves, use 2x when conviction or drawdown risk is lower, and use inverse ETFs only as carefully justified hedge or downside exposure. "
"SHV/cash-like exposure may count as one position only when evidence is too weak for full leveraged risk. "
+ DATA_USAGE_RULE + " " + LEVERAGE_RULE + " " + ORDER_READINESS_RULE
),
)
def on_trading_iteration(self):
context = {
"date": self.get_datetime().date().isoformat(),
"universe": self.parameters["universe"],
"min_positions": self.parameters["min_positions"],
"data_expectation": "Use get_fred_snapshot/get_fred_latest and alpaca_news commonly; smoke tests inspect agent_detail for actual tool calls.",
"leverage_expectation": "Use only leveraged ETFs plus SHV/cash escape, hold at least three positions, and bias toward diversified 3x exposure.",
"data_tool_validation_run_id": "2026-07-08-fresh-alpaca-news-fred-smoke-v1",
}
growth = self.agents["growth_agent"].run(
task_prompt=(
"Call get_fred_snapshot or get_fred_latest and alpaca_news, then rank the strongest leveraged ETFs from a growth-regime view. "
"Prefer diversified 3x exposure when evidence supports it."
),
context=context,
)
inflation = self.agents["inflation_agent"].run(
task_prompt=(
"Call get_fred_snapshot or get_fred_latest and alpaca_news, then rank the strongest leveraged ETFs from an inflation-and-rates view. "
"Explain any inverse or 2x risk-down preference."
),
context=context,
)
liquidity = self.agents["debt_liquidity_agent"].run(
task_prompt=(
"Call get_fred_snapshot or get_fred_latest and alpaca_news, then rank the strongest leveraged ETFs from a debt-and-liquidity view. "
"Explain whether liquidity argues for 3x risk, 2x risk-down, inverse exposure, or SHV ballast."
),
context=context,
)
disagreement = self.agents["thoughtful_disagreement"].run(
task_prompt="Challenge the growth, inflation, and liquidity views. Prefer a diversified leveraged basket of at least three symbols unless risk evidence argues for SHV/cash-like ballast.",
context={**context, "growth": growth.summary, "inflation": inflation.summary, "liquidity": liquidity.summary},
)
self.agents["trader"].run(
task_prompt=(
"Rebalance into a diversified leveraged ETF basket of at least three positions. Bias toward 3x ETFs, use 2x only as a risk-down choice, and use inverse ETFs only with explicit hedge/downside evidence. "
"Before each order, call account_portfolio, account_positions, and market_last_price for the exact ordered symbol in this same run. "
"Explain which specialist advice you accepted or rejected and cite FRED/Alpaca evidence used."
),
context={**context, "growth": growth.summary, "inflation": inflation.summary, "liquidity": liquidity.summary, "disagreement": disagreement.summary},
)
if __name__ == "__main__":
quote_asset = Asset("USD", Asset.AssetType.FOREX)
params = AITradingTeamRayDalioIdeaMeritocracyStrategy.parameters
if IS_BACKTESTING:
trading_fee = TradingFee(percent_fee=0.001)
AITradingTeamRayDalioIdeaMeritocracyStrategy.backtest(
datasource_class=None,
benchmark_asset=Asset("SPY", Asset.AssetType.STOCK),
buy_trading_fees=[trading_fee],
sell_trading_fees=[trading_fee],
quote_asset=quote_asset,
parameters=params,
)
else:
trader = Trader()
strategy = AITradingTeamRayDalioIdeaMeritocracyStrategy(
quote_asset=quote_asset,
parameters=params,
)
trader.add_strategy(strategy)
trader.run_all()
```Shown in full with attribution under the source's licence. Licence: GPL-3.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.