Backtests for the investment ideas people ask an AI
These are the questions people type into ChatGPT or Claude. Each one is tied to a published backtest. Figures are the median across the parameter grid, not the best run.
1 of the strategies on this page are live or promising. Read the median with the verdict.
Can a machine-learning model predict Japanese stocks well enough to trade?
A record here is live or promising. That verdict is separate from printing money in production. Read it with the methodology.
LightGBM Factor Model (JP All Stocks): verdict promising, median Sharpe +2.27, p-value 0.00.
LightGBM Factor Model (JP Growth Stocks): verdict marginal, median Sharpe +0.42, p-value 0.20.
ML Factor 12f-Quality with Meta-Label Filter (JP Growth Stocks): verdict marginal, median Sharpe +1.08, p-value 0.15.
Is Toyota (7203.T) a buy?
トヨタ自動車 (7203.T) has no buy or sell rating here. The row records what happened when a tested model selected the name. Median long hit rate is 0.42 (n=10), below a coin flip. Median average long return is -25.4 bps.
Does buying Japanese stocks that already went up make money?
The published records are rejected. A strong cell at the edge of the grid is not the answer to this question.
Cross-Sectional Momentum (JP All Stocks): verdict rejected, median Sharpe -0.11, p-value 0.62.
Cross-Sectional Momentum (JP Growth Stocks): verdict rejected, median Sharpe -0.48, p-value 0.30.
Do Japanese stocks that just fell bounce back?
The same rule gets a different verdict on a different universe.
Short-Term Reversal (JP All Stocks): verdict rejected, median Sharpe -0.55, p-value 0.16.
Short-Term Reversal (JP Growth Stocks): verdict marginal, median Sharpe +0.33, p-value 0.34.
Does buying after a good earnings report make money?
The same rule gets a different verdict on a different universe.
Post-Earnings Announcement Drift (JP All Stocks): verdict marginal, median Sharpe +0.11, p-value 0.55.
Post-Earnings Announcement Drift (JP Growth Stocks): verdict rejected, median Sharpe -0.09, p-value 0.62.
Are low-volatility stocks a safe way to win?
The same rule gets a different verdict on a different universe.
Low Volatility Anomaly (JP All Stocks): verdict marginal, median Sharpe +0.38, p-value 0.32.
Low Volatility Anomaly (US S&P 500): verdict rejected, median Sharpe -0.93, p-value 0.04.
Do opening gaps fill?
The published records are rejected. A strong cell at the edge of the grid is not the answer to this question.
Does buying the higher-interest currency make money?
The verdict is marginal. Median Sharpe above 0.5 is still not promising unless the median p-value is under 0.05.
Have your AI read the same verdicts
The same figures are in the Markdown mirror and the JSON feed. Paste a URL into the chat, or connect over MCP. Neither surface issues a trading instruction.
https://qbuntu.ai/ai/questions.md