Przejdź do głównej zawartości

Position Sizing & Portfolio

"Jak duży powinien być ten trade?" to question które decides czy edge compounds lub blows up. Institutions answer to z volatility targeting i Kelly criterion, i build book z risk parity rather niż equal dollars. cMind brings oba do retail — deterministyczne math na strategy's return series, z plain-English recommendation.

Otwórz cBots → Position Sizing (/quant/sizing).

Single-strategy sizing​

Biorąc strategy's returns (lub equity curve), target annual volatility, Kelly fraction i leverage cap, sizer reports:

  • Realized annual volatility — strategy's own volatility, annualizowany przez square-root-of-time rule.
  • Volatility-target sizing — exposure które makes realized volatility meet Twój target (target ÷ realized vol), capped na Twój leverage limit. Lower-vol strategies earn more size.
  • Full Kelly — growth-optimal fraction f* = μ / σ² (mean over variance z returns).
  • Fractional Kelly — f* scaled przez Twój Kelly fraction. Half-Kelly (0.5) to common safe choice; full Kelly to famously too aggressive dla real, uncertain edges.
  • Recommended exposure — smaller (safer) z volatility-target i fractional-Kelly sizings, capped. Strategy z no positive edge (full Kelly ≤ 0) to sized do zero.
POST /api/quant/sizing
{ "returns": [...], "targetVolatility": 0.10, "kellyFraction": 0.5, "leverageCap": 3 }

Portfolio allocation​

Give to dwa lub więcej strategies (aligned return series) i builds book przez inverse-volatility risk parity — każdy strategy weighted przez 1 / volatility, normalized — więc risk, nie dollars, to shared evenly. To także returns:

  • correlation matrix across Twoje strategies (spot ones które są secretly same bet);
  • projected portfolio volatility na te weights, z sample covariance;
  • leverage factor które scales całą book toward Twój target volatility (capped).
POST /api/quant/portfolio
{ "strategies": [[...], [...]], "targetVolatility": 0.10, "leverageCap": 3 }

Dlaczego jest niezawodny​

Wszystko to to pure, deterministyczne domain code (Core.Portfolio) z żadną infrastrukturą dependency i no external calls — unit-tested dla vol-target scaling, Kelly formula, equal-risk property z inverse-volatility weights, i correlation matrix. Advisory domyślnie: numbers to recommendation, nigdy automatic order.