How far a Sharpe ratio can be trusted: standard error, and probabilistic and deflated Sharpe ratios
One command in the Portfolio pack, available in the CLI, as an MCP tool and as a JavaScript import: quant sharpe-stats.
What it needs, and why
Every input is checked before anything is calculated. returns is required; the rest refine the answer.
| Input | Required | The question it asks | Why it matters |
|---|---|---|---|
returns | yes | What are the periodic returns, and how many periods make a year? | Every statistic is estimated from the return series. |
trials | optional | How many strategy variants were tried before settling on this one? | The more variants tried, the higher a Sharpe ratio must be to mean anything. |
Requests it answers
- “is this Sharpe ratio significant”
- “could this backtest be luck”
- “deflated Sharpe ratio”
An agent with this pack loaded picks the command from phrasing like that, asks for whatever input is missing, and reports the figures with the method that produced them.
Limits it inherits from wealth-portfolio
- Tax-loss harvesting checks only purchases in the 30 days *before* the sale, across the lots supplied. Confirm separately that nothing substantially identical is bought in the 30 days after, including in IRA and spouse accounts. Replacement candidates track a different index, but whether a fund is "substantially identical" is a judgment for the advisor or tax professional. No replacement is suggested for total-bond funds.
- Backtests use approximate embedded annual returns and capture drawdowns at year-end only.
- Forward-simulation regime assumptions are illustrative, not forecasts; pass your firm's capital market assumptions for client work.
- VaR is parametric normal and only as good as the volatility you supply.
The full pack, its other 10 commands and what it deliberately leaves alone are on the Portfolio page.