---
title: "Risk estimation"
description: "Risk estimation is how an allocation method measures volatility and correlation from history: the plain sample estimate, or one of seven corrected ones."
canonical_url: "https://fincanva.com/docs/strategies/risk-estimation"
last_updated: "2026-09-24"
md_url: "https://fincanva.com/docs/strategies/risk-estimation.md"
---

# Risk estimation

Risk estimation is the choice of how an [allocation method](/docs/strategies/allocation-and-allocation-method) measures, from past returns, how much each instrument moves and how much each pair moves together — the covariance matrix that methods such as [Risk Parity](/docs/strategies/risk-parity) and [MPT](/docs/strategies/modern-portfolio-theory) are built on. The app offers eight estimators: the classic **Sample** estimate, and seven that correct it, either to reduce estimation noise or to react faster to changing markets. The control is labelled **Risk estimation**, with the hint "How the method measures volatility and correlations from history."

**Also seen as:** covariance estimator, covariance estimation, shrinkage estimator

## Why does the risk estimate need correcting?

Because the sample estimate is noisy when there are many instruments and not much history. With 30 instruments, the covariance matrix holds 465 separate numbers — 30 volatilities and 435 correlations — and a 12-month window gives about 252 days to estimate them from. Some correlations then come out unusually low by pure chance, and an optimiser will lean hard on exactly those. A corrected estimate pulls such accidents back toward something more plausible, which makes the weights steadier from one rebalance to the next. The price is a small, deliberate bias.

## Which estimators can you choose?

The list names each one and says what it does, in the app's own words:

| Estimator | What it does |
|---|---|
| **Sample** | "The classic estimate, with no corrections." |
| **Ledoit-Wolf · identity** | "Reduces noise more firmly, treating every instrument the same way." |
| **Ledoit-Wolf · constant correlation** (Recommended) | "Reduces noise by pulling correlations toward their average." |
| **Exponentially weighted moving average (EWMA)** | "Gives more weight to recent days: reacts sooner when volatility changes." |
| **Manual shrinkage** | "You choose how much to correct, from 0 to 1." |
| **Marchenko-Pastur** | "Separates the signal from the noise and keeps only the signal." |
| **Nonlinear Ledoit-Wolf** | "A tailored correction for each part of the estimate." |
| **Regime-conditional (HMM)** | "Uses crisis correlations when the market is in crisis." |

In prose: the first five are the basic estimators, and the last three sit in the list's **Advanced** group.

- **Ledoit-Wolf** estimators blend the sample estimate with a simple target, and work out from the data how much to blend. The *identity* target treats every instrument alike; the *constant correlation* target keeps each instrument's own volatility and pulls every pairwise correlation toward the average one, which usually suits a basket of similar instruments such as stocks. It suits a deliberately mixed basket — stocks, bonds and gold together — less well, because there two very different levels of correlation are real.
- **EWMA** weighs recent days more heavily, so the estimate follows a change in volatility within weeks instead of averaging it away over the whole window. It also reacts to noise, which can raise trading.
- **Manual shrinkage** is the constant-correlation blend with the amount set by you rather than by the data.
- **Marchenko-Pastur** uses random-matrix theory to tell which parts of the correlation structure are indistinguishable from noise, flattens those, and keeps the rest — such as the market-wide and sector-wide co-movement. It can discard weak structure that was real.
- **Nonlinear Ledoit-Wolf** corrects each part of the estimate by its own amount rather than by one blend for everything. It helps most with large baskets and short windows, and adds almost nothing when history is long compared with the number of instruments.
- **Regime-conditional (HMM)** estimates one covariance from calm periods and one from turbulent periods, and leans on the turbulent one when the market looks turbulent today, so the method sees crisis-level correlations while a crisis lasts.

## How do shrinkage and EWMA work?

Shrinkage blends two estimates:

$$
\hat{\Sigma} = \delta \, F + (1 - \delta) \, S
$$

where: $S$ is the sample covariance, $F$ is the target (for constant correlation: every instrument keeps its own volatility, every pair gets the average correlation), and $\delta$ is the blend between 0 and 1. Ledoit-Wolf works $\delta$ out from the data; **Manual shrinkage** lets you set it as **Shrinkage intensity** — "0 = no correction, 1 = all correlations equal."

EWMA updates each variance with a **Decay factor** $\lambda$:

$$
\sigma^2_t = \lambda \, \sigma^2_{t-1} + (1 - \lambda) \, r^2_{t-1}
$$

where: $\sigma^2_t$ is today's variance estimate, $r_{t-1}$ yesterday's return, and $\lambda$ how much of the old estimate survives each day — "The closer to 1, the more distant data counts." At the default 0.94 the estimate's memory is about 1 ÷ (1 − 0.94) ≈ 17 trading days.

## Which methods use risk estimation?

Only the methods that are built on a covariance estimate show the control; every other method ignores it.

- **Inside a strategy:** [Risk Parity](/docs/strategies/risk-parity), [MPT (Markowitz)](/docs/strategies/modern-portfolio-theory), [Hierarchical Risk Parity](/docs/strategies/hierarchical-risk-parity), [Hierarchical Equal Risk Contribution](/docs/strategies/hierarchical-equal-risk-contribution), [Nested Clustered Optimization](/docs/strategies/nested-clustered-optimization) and [Maximum Diversification](/docs/strategies/maximum-diversification).
- **Inside a Combined:** Risk Parity, MPT (Markowitz) and Maximum Diversification.

The tail-focused methods — [Minimum CVaR](/docs/strategies/minimum-cvar), [Minimum MAD](/docs/strategies/minimum-mad), [Conditional Drawdown at Risk](/docs/strategies/conditional-drawdown-at-risk), [Entropic Value at Risk](/docs/strategies/entropic-value-at-risk), [Robust worst case](/docs/strategies/robust-worst-case) and [Stochastic programming](/docs/strategies/stochastic-programming) — work on the returns themselves and never show it. In MPT the choice affects the risk side only: expected returns are left to MPT itself — the plain averages of the window, or the Black-Litterman estimate when that option is on.

## How does Fincanva handle it?

- **A method you choose now starts on Ledoit-Wolf · constant correlation**, the estimator the list marks **Recommended**. A strategy saved before this choice existed keeps **Sample**, so its results do not change by themselves.
- The Risk-On and Risk-Off profiles each keep their own estimator.
- **An instrument whose price stayed flat for the whole window has no measured risk, and most estimates leave it that way.** With Sample, Ledoit-Wolf · constant correlation, EWMA, Manual shrinkage, Marchenko-Pastur or Regime-conditional (HMM), such an instrument stops the backtest when the method is [Hierarchical Risk Parity](/docs/strategies/hierarchical-risk-parity), [Hierarchical Equal Risk Contribution](/docs/strategies/hierarchical-equal-risk-contribution), [Nested Clustered Optimization](/docs/strategies/nested-clustered-optimization) or [Maximum Diversification](/docs/strategies/maximum-diversification). Ledoit-Wolf · identity and Nonlinear Ledoit-Wolf give it a small positive risk instead, so the backtest runs — Nonlinear Ledoit-Wolf only once the window holds about 30 trading days, below which it behaves as Sample.
- The estimate is taken over the [calculation window](/docs/strategies/calculation-window) (**In-sample**), so the window and the estimator work together.
- **Decay factor** (EWMA only) accepts 0.90 to 0.99 and starts at 0.94.
- **Shrinkage intensity** (Manual shrinkage only) is required and accepts 0 to 1.
- **Two advanced estimators need enough history, and quietly fall back to Sample without it.** Regime-conditional (HMM) needs about 126 trading days — an in-sample period of at least 6 months — and also falls back when calm and turbulent periods cannot be told apart. Nonlinear Ledoit-Wolf needs about 30 trading days, so at least 2 months. When the window is too short the app warns: "With this history the estimate has no effect", with an action to lengthen the in-sample period.

## Which plan includes the advanced estimators?

- **The five basic estimators come with the method.** Any plan that includes a method which uses risk estimation includes Sample, both Ledoit-Wolf estimators, EWMA and Manual shrinkage with it — see each method's page for the plan that includes it.
- **Marchenko-Pastur, Nonlinear Ledoit-Wolf and Regime-conditional (HMM) start at Ultimate.** On Free, Starter and Advanced they stay in the list, tagged with the plan that includes them; choosing one opens a dialog naming that plan and leaves the current estimator in place. Ultimate and Professional allow them. See [what each plan includes](/docs/account-security/what-each-plan-includes).

## What does it look like in practice?

A strategy holds 25 stocks on a 12-month window. Most pairs show a correlation around 0.5, but the sample estimate puts one pair at 0.05 — very likely an accident of this particular year.

- **Sample** keeps the 0.05, and MPT leans on that pair as if it were a near-perfect diversifier.
- **Manual shrinkage at 0.4** moves it to 0.6 × 0.05 + 0.4 × 0.5 = **0.23**: still below the crowd, but no longer an outlier worth betting the portfolio on. Every other pair moves toward 0.5 in the same proportion, and each stock's own volatility is left as measured.
- **Ledoit-Wolf · constant correlation** does the same blend but picks the amount from the data — more when the estimate is noisier, for example with more stocks or a shorter window.

Fincanva is for education and illustration only. It is not personalised financial advice, and past or simulated results do not predict future ones. [Read the Terms Addendum](https://fincanva.com/terms/addendum#section-3)