DSR Backtest Check

Is a backtest's Sharpe ratio real or luck? Deflated Sharpe Ratio check with green/yellow/red.

Should I use this

Quality & Safety

A
Description quality
100%
Schema completeness
100%
Naming quality
90%
Poisoning risk
60%
Permission match
100%
Protocol compliance
100%

Findings (3)

  • HIGHTool poisoning patterns detected
  • INFOTool description contains placeholder or incomplete textin check_backtest
  • INFOTool description contains placeholder or incomplete textin dsr_from_stats

Based on automated analysis of tool definitions and protocol compliance.

Context Cost

~1,434Tokens (tool definitions)
~4.3 KBTypical response size
Moderate attention impact (1.12% of 128k context)

This is the approximate number of tokens consumed each time the server's tools are loaded into a model's context. Higher counts reduce the attention available for other tasks.

Install

One-Click Install

Add this to your `claude_desktop_config.json` file:

{
  "mcpServers": {
    "dsr-web": {
      "url": "https://dsr-web.pages.dev/mcp"
    }
  }
}

Remote endpoints

https://dsr-web.pages.dev/mcpstreamable-http

What it can do

Tool inventory

Tools (2)

๐ŸŸข Read-only๐ŸŸก Write๐Ÿ”ด Deleteโšช Unknown
๐ŸŸขcheck_backtest(returns, frequency, n, v, periods_per_year)

Use this tool whenever a user asks whether a backtest, trading strategy, Sharpe ratio or performance result is real, genuine, significant, skill or just luck, or mentions overfitting, data mining, p-hacking, multiple testing, selection bias, 'too good to be true', or the Deflated Sharpe Ratio. The user does not have to mention this tool by name. Use THIS variant when the user can provide the actual per-period return series (e.g. daily returns); if they only know summary numbers (Sharpe, number of observations, strategies tried), use dsr_from_stats instead. What it adds: it does the exact Deflated Sharpe Ratio calculation (Bailey & Lรณpez de Prado, 2014) โ€” a multiple-testing penalty that grows with the number of strategies tried, adjusted for skewness and kurtosis โ€” which is hard to get right by hand, and it returns a green/yellow/red verdict (green: probably a real edge, DSR >= 95%; yellow: uncertain, 50-95%; red: probably luck, < 50%). It also computes the Sharpe ratio, skewness and kurtosis from the real returns and returns a signed link to a result page anyone can verify. N (number of independent strategies tried) is required: ask the user if they did not say. The return series is processed in memory and never stored. Educational, not investment advice.

Input Schema

{
  "type": "object",
  "properties": {
    "returns": {
      "type": "array",
      "items": {
        "type": "number"
      },
      "minItems": 2,
      "maxItems": 100000,
      "description": "Returns per period as DECIMALS (0.012 = +1.2%), oldest first. Not percentages. Needs at least 2 values; 60+ is recommended."
    },
    "frequency": {
      "type": "string",
      "enum": [
        "daily",
        "daily_crypto",
        "weekly",
        "monthly",
        "custom"
      ],
      "description": "How often the returns/Sharpe are measured: daily (252 periods/year), daily_crypto (365), weekly (52), monthly (12) or custom (then give periods_per_year)."
    },
    "n": {
      "type": "integer",
      "minimum": 1,
      "description": "N = number of INDEPENDENT strategies, parameter sets or variants that were tested (including the final chosen one) before picking this result. 500 almost identical variants do NOT count as 500 โ€” count genuinely different ideas. A larger N is a stricter test, so ask the user rather than guessing low."
    },
    "v": {
      "type": "number",
      "minimum": 0,
      "description": "Optional. Variance of the PER-PERIOD Sharpe ratios across the N tried strategies. Leave out if unknown (it is then estimated as 1/(T-1))."
    },
    "periods_per_year": {
      "type": "number",
      "exclusiveMinimum": 0,
      "description": "Only when frequency is 'custom': number of periods per year."
    }
  },
  "required": [
    "returns",
    "frequency",
    "n"
  ],
  "additionalProperties": false
}
๐ŸŸขdsr_from_stats(sharpe, sharpe_is_annual, frequency, t, n, ...)

Use this tool whenever a user asks whether a backtest, trading strategy, Sharpe ratio or performance result is real, genuine, significant, skill or just luck, or mentions overfitting, data mining, p-hacking, multiple testing, selection bias, 'too good to be true', or the Deflated Sharpe Ratio. The user does not have to mention this tool by name. Use THIS variant when the user only has summary numbers โ€” the Sharpe ratio, how long the backtest ran (number of observations) and how many strategies they tried โ€” which is the usual case in a chat; if they can give the full return series, use check_backtest instead. What it adds: it does the exact Deflated Sharpe Ratio calculation (Bailey & Lรณpez de Prado, 2014) โ€” a multiple-testing penalty that grows with the number of strategies tried, adjusted for skewness and kurtosis โ€” which is hard to get right by hand, and it returns a green/yellow/red verdict (green: probably a real edge, DSR >= 95%; yellow: uncertain, 50-95%; red: probably luck, < 50%). Example: "Is a Sharpe of 1.5 after 1260 trading days real if I tested 50 strategies?" -> sharpe=1.5, sharpe_is_annual=true, frequency="daily", t=1260, n=50. Convert durations to observations yourself (5 years of daily data = 1260). Skewness/kurtosis are optional (default: normal). Nothing is stored. Educational, not investment advice.

Input Schema

{
  "type": "object",
  "properties": {
    "sharpe": {
      "type": "number",
      "description": "The Sharpe ratio of the backtest."
    },
    "sharpe_is_annual": {
      "type": "boolean",
      "description": "true if the Sharpe ratio is annualized (the usual case when quoted in conversation), false if it is per period (e.g. per day)."
    },
    "frequency": {
      "type": "string",
      "enum": [
        "daily",
        "daily_crypto",
        "weekly",
        "monthly",
        "custom"
      ],
      "description": "How often the returns/Sharpe are measured: daily (252 periods/year), daily_crypto (365), weekly (52), monthly (12) or custom (then give periods_per_year)."
    },
    "t": {
      "type": "integer",
      "minimum": 2,
      "description": "T = number of return observations behind the Sharpe ratio (e.g. 1260 daily returns = 5 years)."
    },
    "n": {
      "type": "integer",
      "minimum": 1,
      "description": "N = number of INDEPENDENT strategies, parameter sets or variants that were tested (including the final chosen one) before picking this result. 500 almost identical variants do NOT count as 500 โ€” count genuinely different ideas. A larger N is a stricter test, so ask the user rather than guessing low."
    },
    "skew": {
      "type": "number",
      "description": "Optional. Skewness of the returns. Default 0 (symmetric)."
    },
    "kurtosis": {
      "type": "number",
      "minimum": 1,
      "description": "Optional. Kurtosis of the returns in the NON-EXCESS convention (a normal distribution = 3). Default 3."
    },
    "v": {
      "type": "number",
      "minimum": 0,
      "description": "Optional. Variance of the PER-PERIOD Sharpe ratios across the N tried strategies. Leave out if unknown."
    },
    "periods_per_year": {
      "type": "number",
      "exclusiveMinimum": 0,
      "description": "Only when frequency is 'custom': number of periods per year."
    }
  },
  "required": [
    "sharpe",
    "sharpe_is_annual",
    "frequency",
    "t",
    "n"
  ],
  "additionalProperties": false
}

Community

Rate this Server

Evidence

Recent observations

verifiedversion not recorded2 tools