Skip to content
All library documents

AI Value Stock Selection with Research, Skeptic Review, and Long-Term Holding

Article Lumibot

Summary

This document describes an AI-assisted value strategy organized as three roles. A research agent reviews company reports and current prices to identify a small group of businesses with steady profits, durable competitive advantages, and prices judged reasonable. A skeptic agent challenges each choice for valuation, weakening business quality, debt, or inconsistent figures. A trading agent holds the surviving stocks in roughly equal allocations and sells when the skeptic no longer supports a holding because the business has deteriorated or its price has become excessive.

The document reports a brief backtest over January 5–16, 2026, using daily prices and company reports. It says the bot held four named stocks, finished up 1.7%, and that SPY gained 1% over the period; cash stayed above a stated minimum. This short sample does not establish long-run effectiveness, and the document explicitly cautions against treating it as a forecast. The selection judgments depend on AI review of reports and prices, while the described evidence does not establish robustness across market regimes or universes.

Key ideas

  • A research agent screens companies for steady profits, durable advantages, and reasonable prices.
  • A skeptic agent reviews picks for excessive valuation, weakening business quality, debt, and inconsistent figures.
  • The trading role holds surviving names in roughly equal allocations and sells when the review turns negative.
  • The document reports a short backtest with four holdings and a small gain relative to SPY.
  • A brief backtest cannot establish future returns or robustness across different conditions.

Tags

EquitiesFactor investingMachine learningPortfolio constructionBacktesting
Full text
# agents example warren buffett ai stock picker


Warren Buffett AI Stock Picker
==============================

.. meta::
   :description: An AI stock picker that thinks like Warren Buffett. It reads company filings, hunts for great companies at fair prices, and holds them. Free Python code for LumiBot.

.. image:: ../docs/assets/ai-agent-workflows/warren-buffett-ai-stock-picker.png
   :alt: Research agent reads company reports and prices, skeptic agent attacks each pick, trading agent owns the picks that survive, then buys and holds
   :width: 100%

This bot picks stocks the way Warren Buffett describes in his Berkshire Hathaway letters: buy wonderful businesses at fair prices and hold them for a long time. Why copy that? From 1965 through 2025, Berkshire's stock grew 19.7% a year, versus 10.5% for the S&P 500 (`Berkshire 2025 letter <https://www.berkshirehathaway.com/letters/2025ltr.pdf>`__).

How it works
------------

1. **Research agent** reads each company's latest financial reports and checks its price today. It picks the 3 to 5 companies with the best mix of steady profits, a lasting edge over rivals, and a fair price.
2. **Skeptic agent**, like Buffett's partner Charlie Munger, attacks each pick: a price that is too high, a shrinking edge, too much debt, or numbers that do not add up. It keeps only the picks that survive.
3. **Trading agent** owns the survivors, split about evenly. It holds for the long run and only sells when the skeptic drops a stock because the business got worse or the price got far too high.

Change ``universe`` to pick from different companies.

Run it on BotSpot
-----------------

Run this bot on `BotSpot <https://botspot.trade/marketplace?utm_source=documentation&utm_medium=docs&utm_campaign=lumibot_ai_examples&utm_content=agents_example_warren_buffett_ai_stock_picker>`_ without installing anything. BotSpot runs LumiBot in the cloud, backtests it, and connects it to your broker.

Backtest tear sheet
-------------------

GPT-6 Luna, January 5 to 16, 2026, Yahoo daily prices and company reports, $100,000 start. After the skeptic's review the bot bought AXP, GOOGL, KO, and PG and held them, ending at $101,660 (+1.7%) while SPY rose 1%. Cash never went below $5,145.

.. image:: ../docs/assets/ai-bot-backtests/warren-buffett-ai-stock-picker.png
   :alt: Backtest tear sheet for the Warren Buffett AI Stock Picker
   :width: 100%
   :target: tearsheets/warren-buffett-ai-stock-picker.html

`Open the full tear sheet <tearsheets/warren-buffett-ai-stock-picker.html>`__. A short backtest shows the bot works as written. It is not a promise of future returns.

The code
--------

The whole bot is one short file. The prompts are plain English, and they are the strategy.

.. literalinclude:: ../lumibot/example_strategies/ai_trading_team_warren_buffett_value.py
   :language: python

Run it yourself
---------------

.. code-block:: bash

   pip install lumibot
   python -m lumibot.example_strategies.ai_trading_team_warren_buffett_value

Put these in your ``.env`` file: ``OPENAI_API_KEY``, and your broker keys (for example ``ALPACA_API_KEY``, ``ALPACA_API_SECRET``, and ``ALPACA_IS_PAPER=true`` for paper trading). With ``IS_BACKTESTING=false`` the bot trades. With ``IS_BACKTESTING=true`` it backtests instead; set ``BACKTESTING_START`` and ``BACKTESTING_END`` to pick the dates, and start with a week or two, because every AI call costs a little.

Good to know
------------

* Inspired by Warren Buffett's public shareholder letters. Not affiliated with or endorsed by Warren Buffett or Berkshire Hathaway.

See :doc:`agents_examples` for more AI trading bots and :doc:`strategy_run_modes` for backtest and live runs.

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.