RSJ Portfolio Portfolio Optimization

← All posts

2026-10-23

Why Low Correlation Stocks Reduce Portfolio Volatility

Why Low Correlation Stocks Reduce Portfolio Volatility
portfolio optimization modern portfolio theory portfolio correlation portfolio optimization hierachical risk parity minimize volatility minimize semivariance maximize sharpe

Why Low Correlation Stocks Reduce Portfolio Volatility

Ask most investors how to build a calmer portfolio and you’ll get a stock-picking answer: screen for low-volatility names, avoid the high-beta stuff, sleep better. Then those same ten “calm” stocks all sell off together in the same week and the portfolio turns out to be far more volatile than the sum of its parts suggested.

The reason is that portfolio volatility is not an average. A stock’s own volatility tells you how much it moves; it says nothing about how it moves relative to everything else you own. Two stocks can each swing 30% a year and still produce a portfolio that swings 19% — if they move independently. That’s why the question worth asking isn’t “which stocks are volatile?” but “why do low correlation stocks reduce portfolio volatility?”

Short answer: because their fluctuations partially cancel. This article unpacks that with an analogy, the actual arithmetic, the misconceptions that trip people up, and how a statistical optimizer turns the idea into weights you can act on.

The plain-language answer: volatility is not just about individual stocks

Portfolio volatility is the combined fluctuation of everything you hold, measured as the standard deviation of the portfolio’s returns over time. It is a property of the collection, not of any single line item.

Here’s the mechanism in one sentence:

  • When two stocks move together (high correlation), their swings arrive at the same time and add up.
  • When two stocks move independently (low correlation), one zigs while the other zags, and their swings partially cancel.

The cancellation is not magic and it is not free — but it is real, and it is the single largest lever you have over the risk of a diversified portfolio. Low correlation can lower overall volatility without requiring you to give up expected return, because you are not removing assets from the portfolio; you are choosing assets whose risks don’t amplify each other.

That is the punchline most volatility screens miss. Investors hunt for low-volatility stocks. The more useful hunt is for stocks that are volatile in different directions.

A real-world analogy: why mixing ingredients matters

Pour one litre of water at 90 °C into one litre at 10 °C. Neither input is drinkable. The mix is 50 °C — comfortable. You didn’t add anything cold to “cancel” the hot water; you combined two extremes whose deviations point in opposite directions.

A balanced diet works the same way. No single food group is a complete diet, and eating more of your favourite one doesn’t fix the gap. You combine ingredients that cover different deficiencies.

Map that onto a portfolio:

Kitchen Portfolio
Ingredient A stock (or ETF)
How the ingredient reacts to heat How the stock reacts to market events
Two ingredients that respond similarly Two stocks with high correlation
The finished dish Portfolio volatility

The counter-intuitive part of the analogy is the one worth internalising: two volatile but uncorrelated ingredients can produce a smoother combined outcome than one stable ingredient repeated ten times. A portfolio of ten near-identical utility stocks is one ingredient in ten packages. A portfolio of a utility, a miner, a software company and a shipping line is four ingredients, and the dish is steadier.

The goal isn’t to eliminate risk. It’s to combine risks that don’t amplify each other.

How it works under the hood: correlation and portfolio variance

For two assets, portfolio variance is:

σ²p = w₁²σ₁² + w₂²σ₂² + 2·w₁·w₂·ρ₁₂·σ₁·σ₂

Read it left to right: two terms for the individual variances, and one covariance term where ρ is the correlation and σ the standard deviations. That last term is the lever. As ρ falls, the covariance term shrinks — and at ρ = 0 it disappears entirely. At negative correlation it actively subtracts.

You can see the size of the effect with two stocks that are each 30% volatile, held 50/50:

import numpy as np

s1, s2 = 0.30, 0.30      # annualised volatility of each stock
w1, w2 = 0.50, 0.50      # equal weights

for rho in (0.9, 0.5, 0.1, -0.2):
    var = (w1 * s1)**2 + (w2 * s2)**2 + 2 * w1 * w2 * rho * s1 * s2
    print(f"rho={rho:+.1f}  portfolio vol={np.sqrt(var):.2%}")
rho=+0.9  portfolio vol=29.24%
rho=+0.5  portfolio vol=25.98%
rho=+0.1  portfolio vol=22.25%
rho=-0.2  portfolio vol=18.97%

Nothing about the two stocks changed. Only their relationship did — and the portfolio got a third less volatile.

Scaled to n assets the same structure holds: portfolio variance is a weighted sum over every individual variance and every pairwise covariance. With many holdings, the covariance terms outnumber the variance terms, which is exactly why correlation dominates the risk of a diversified book. Computing it directly is straightforward once you have a covariance matrix:

import numpy as np

# returns: DataFrame of daily returns, one column per ticker
cov = returns.cov() * 252          # annualise the covariance matrix
w = np.array([0.4, 0.3, 0.2, 0.1]) # weights, must sum to 1

port_var = w.T @ cov.values @ w
print(f"annualised portfolio volatility: {np.sqrt(port_var):.2%}")

The catch is that cov is estimated, not known. A raw sample covariance matrix is noisy — a few hundred daily observations to estimate thousands of pairwise relationships. In practice you use an estimator: sample covariance, a shrinkage estimator that pulls noisy entries toward a structured target, or a factor model that attributes co-movement to a handful of common exposures. Different estimators give different weights.

Modern portfolio theory takes this machinery and inverts the question: instead of computing the risk of weights you already hold, it searches for the weights that minimise risk for a given return target — or maximise return per unit of risk. The covariance matrix is the input; the allocation is the output.

For a hands-on walk through reading a correlation matrix and acting on it, see How to Reduce Portfolio Volatility Using Correlation Analysis. This article stays on the “why”.

Why it matters: practical benefits and common misconceptions

Lower volatility is not an abstract statistic. It shows up as smaller drawdowns and a smoother equity curve, which is what keeps you holding the portfolio through a bad quarter instead of capitulating at the bottom. Smoother returns also make risk limits, position sizing and rebalancing rules easier to obey.

Three misconceptions are worth naming, because each one quietly undoes the benefit.

“Low correlation always means low risk.” No. Correlation is estimated from history and it drifts. In stressed markets, correlations across risk assets tend to rise together — the diversification that worked in calm periods thins out exactly when you need it. Low correlation reduces volatility on average, not in every regime.

“You only need low-volatility stocks.” This is the screen-and-hope approach. Ten sleepy stocks in the same sector are highly correlated; the portfolio inherits the sector’s risk with none of the offset. Conversely, two genuinely volatile stocks with weak correlation can produce a steadier portfolio than either alone, as the table above shows.

“Diversification across many stocks is enough.” Count is not diversification. Fifty holdings that all react to the same rate decision, the same oil price, the same dollar move are one bet in fifty wrappers. Without checking correlations you may be doubling down on the same risk while feeling diversified.

Statistical optimization avoids all three by modelling the relationships explicitly instead of inferring them from labels like sector or market cap.

From theory to practice: how RSJ Portfolio applies low-correlation allocation

RSJ Portfolio is a Windows desktop application that builds a statistically optimized stock portfolio from historic end-of-day prices. It analyses the return correlation between candidate stocks and allocates capital toward those that correlate weakly — or whose downside correlates weakly, which matters if you care more about losses clustering than about daily wiggle.

Under the hood, the allocation uses modern portfolio theory. What you choose on top of it:

  • Objective — minimum volatility, maximum Sharpe ratio, minimum semi variance, hierarchical risk parity, or conditional value at risk.
  • Risk estimator — one of eight selectable risk models, which determines how the covariance structure is estimated from the price history.
  • Weight constraints — bounds on how the optimizer may allocate, so the output stays investable rather than a theoretical corner solution.

The workflow is managed end to end: build a candidate list, point the app at a data provider, set the local currency and date range, run the optimizer, and review the recommended allocation. The matrix algebra happens inside the app — doing it by hand from raw price series is possible, and it is the hard way.

The features page documents every method, risk model and export; the concept chapter of the user guide covers the statistical background. For how the estimator choice changes results, see the sibling post Portfolio Optimizer with Selectable Risk Models: A Comparison.

Putting it into practice: steps to build a lower-volatility portfolio

Step 0 — Install and verify. Download the Windows installer from the downloads page and check the published SHA-512 against the file before running it:

Get-FileHash .\RSJPortfolio-Setup-<version>.exe -Algorithm SHA512

Step 1 — Assemble a candidate list. Use the built-in symbol search to add the stocks you’re actually interested in. More candidates give the optimizer more room to find weak correlations; a list of five doesn’t offer much. Details in the selection chapter.

Step 2 — Configure the data settings. Set the data provider, your local currency, the date range and the cleaning options. Correlations are only as good as the price series behind them — stale prints, splits and currency mismatches distort them. The data page explains the cleaning pipeline and the local cache; the configuration chapter walks through each setting.

Step 3 — Choose an objective and a risk model. If you’re unsure, start from the recommended starting parameters in the getting-started guide. Minimum volatility is the natural first run when the question you’re asking is “how do I reduce portfolio volatility?”

Step 4 — Run the optimizer and read the result. Review which stocks received weight and which were dropped, and how the allocation shifts when you change the objective or the estimator. A single run is a data point; two runs that disagree tell you something about how stable the estimates are. The recommendation chapter covers how to read the output.

Step 5 — Validate, then monitor. Don’t trade the first output. Check the assumptions, the date range and the sensitivity of the weights first — the checklist is in How to Validate Portfolio Optimizer Output Before You Trade.

Correlations drift, so keep an eye on them:

# rolling 90-day correlation between two holdings — watch it drift
roll = returns["AAA"].rolling(90).corr(returns["BBB"])
print(roll.tail())

When the pairs you relied on start moving together, refresh the date range and re-optimize.

FAQ

Does low correlation always reduce portfolio volatility?

In general, yes: when stocks are less correlated, their price movements tend to offset each other, which lowers the portfolio’s overall volatility compared to a portfolio of highly correlated stocks with the same individual volatilities. However, the magnitude of the effect depends on the specific correlations and the weights; in extreme market events, correlations can rise and reduce the benefit.

Can I reduce volatility with just a few low-correlation stocks?

It is possible, but a small set of stocks may not provide enough diversification to reliably lower volatility. A larger candidate list gives the optimizer more flexibility to find combinations with weak correlations. Weight constraints can also limit how much the optimizer can exploit low correlations.

How do I measure correlation between stocks?

Correlation is typically measured using historical return data over a chosen period. You can compute it manually from price series — the hard way — but tools like RSJ Portfolio automate the process: they ingest end-of-day prices, clean the data, and estimate correlations as part of the optimization workflow.

What if correlations change over time?

Correlations are not static; they can shift due to market conditions, sector rotations, or macroeconomic events. That’s why it’s important to re-run the optimization periodically and validate the output before trading. RSJ Portfolio lets you update the date range and re-optimize with fresh data.

Conclusion: correlation is the lever for smarter risk control

Low correlation stocks reduce portfolio volatility because their fluctuations offset instead of compounding. That’s the whole idea: portfolio risk is built from variances and covariances, and the covariances are where the leverage lives. Chasing low-volatility names one at a time misses it; counting holdings without checking how they move together misses it too.

Statistical optimization makes the idea actionable by modelling those pairwise relationships explicitly and searching for the allocation that best satisfies an objective you choose. RSJ Portfolio packages that as a Windows application with selectable objectives, eight risk models and configurable weight constraints, driven from end-of-day price data — no matrix algebra on your side.

It’s licensed per single user; the pricing page covers what the license includes and the checkout, and the getting-started guide takes you from install to first optimized allocation.

Grab the installer, add the stocks you were already considering, and run it once with minimum volatility. The weights you get back will tell you more about your portfolio’s correlation structure than any volatility screen will.

Related posts

EU label: AI-generated content