Quant Memo

Visualizers

Central Limit Theorem

Draw samples from a lumpy, skewed distribution and watch their averages magically form a bell curve. Statistics' greatest hit, made visual.

Source distribution (all deliberately non-normal)
1values averaged per trial50

The source, one draw at a time (n = 1)

A hard floor at 0 and a long right tail, most draws are tiny, a rare few are large.
Preparing simulation…

Sampling distribution of the mean (n = 5)

Preparing simulation…

The x-axis is fixed across n. Slide n up and watch the same averages pile into a tighter, taller, more symmetric bell, that shrinking width is the standard error σ/√n. The black curve is the theoretical normal the CLT predicts.

Population mean μ
-
stays put
Mean of sample means
-
≈ μ, always
Population σ
-
spread of one draw
Std error σ/√5
-
theory: shrinks with n
Observed spread
-
matches σ/√n

No matter how weird the original distribution, flat, wildly skewed, two-humped, or a lumpy die, the averages of many samples pile up into a smooth bell curve. Set n = 1 and the bell chart matches the source above exactly; raise n and the skew and lumpiness melt away into a clean, symmetric normal.

The bell also clusters more tightly the bigger each sample is: its width is the standard error σ/√n, which falls as n grows. That is why the normal distribution shows up everywhere, and why averaging more observations reduces noise, quadrupling n only halves the error, so precision is expensive.

Learn how it works

Five worked examples. Read a couple before you dive in, try to answer first, then reveal the solution.

One sample looks like the source

Choose the exponential source (a lopsided, front-loaded shape), set sample size n = 1, and run many trials. Look at the histogram of the means.

Show solution

With n = 1, each 'mean' is just a single draw, so the histogram is a copy of the source shape, piled up on the left with a long tail to the right.

  • Averaging one number does nothing; you're just re-drawing the raw distribution.
  • Skewed source in, skewed picture out.

This is the starting point. Remember how lopsided it looks, because the next step is going to transform it.

The bell appears

Keep the exponential source but raise the sample size to n = 30. Run the trials again and watch the histogram change shape.

Show solution

The lopsided pile magically turns into a symmetric bell curve, centered on the true average.

Each trial now averages 30 draws. The high draws and low draws inside each group tend to cancel, so the group averages cluster tightly and symmetrically around the middle.

This is the Central Limit Theorem: average enough independent things and their average is bell-shaped, no matter how weird the original ingredients were.

Bigger samples, narrower bell

Compare n = 30 against n = 120 on the same source. Watch the width of the bell.

Show solution

The bell stays centered in the same place but gets noticeably narrower as n grows.

The spread shrinks in proportion to 1 ÷ √n:

  • Go from n = 30 to n = 120 (4× the sample) and the width roughly halves (√4 = 2).
  • To cut the spread in half you need the data, not 2×.

That square-root rule is why more data helps, but with diminishing returns, each extra bit of precision costs more than the last.

Even a two-humped source

Switch to the bimodal source, a distribution with two separate humps, nothing like a bell, and set n = 30.

Show solution

The two humps of the source vanish from the picture of the means, which settles into a single clean bell.

  • The averaging blends draws from both humps within each sample, and the group averages land in between.
  • It doesn't matter that the raw shape had two peaks, or was jagged, or was skewed.

The punchline of the CLT: the source can be almost anything, and the distribution of its averages still marches toward the same familiar bell.

Why averaging kills noise

Use any lumpy source (dice or uniform) and step n up from 1 to 5 to 30, watching the histogram tighten each time.

Show solution

This visual explains two things at once:

  • Why averaging reduces noise. Each individual measurement is jumpy, but their average is far steadier, the tight bell shows the average barely moves from trial to trial.
  • Why the normal distribution is everywhere. So many real quantities (test scores, measurement errors, daily returns) are secretly sums or averages of many small effects, and the CLT drags all of them toward the same bell.

It's the deep reason the bell curve shows up across finance and science even when nothing underneath looks like a bell.

What you'll learn

Why averages of almost anything become normally distributed, the reason the bell curve shows up everywhere in statistics and finance.