Heavy Tails in Correlated Sums

Introduce the Central Limit Theorem and then ask why correlated sums can escape its conclusion.

• The CLT guarantees a Gaussian limit for independent summands.
• Real-world sums (returns, signals, network traffic) are often correlated.
• Driving question: why can correlation make the limit's tails heavier than Gaussian?

Ask the learner to choose the mechanism they suspect produces the heavy tail before seeing evidence.

• Make one independent choice about the dominant cause

Show a simulated correlated sum vs. a matched Gaussian, with a log-scale tail panel highlighting excess probability at large |x|.

• Center looks broadly bell-shaped.
• Log-y tail plot shows the correlated sum sits well above the Gaussian.
• Extreme events appear roughly an order of magnitude more often than Gaussian predicts.

Let the learner toggle correlation strength and watch the tail probability of a simulated sum change in real time.

• Adjust correlation coefficient
• Compare tail probability P(|S| > threshold) against Gaussian
• See how stronger correlation inflates the tail

Explain that correlation lets a single large shock influence many subsequent terms, so extremes cluster instead of cancel.

• Independence makes large positive and negative steps cancel on average.
• Positive correlation lets a big step pull future steps in the same direction.
• Rare large excursions accumulate coherently, dominating the tail.
• The limit law inherits these rare-but-aligned events and becomes heavy-tailed.

Clarify the boundary: weak or short-range correlation still yields Gaussian-like tails; only persistent or strong correlation produces heavy tails.

• Short-range dependence typically preserves a Gaussian-like limit.
• Mild correlation changes variance, not tail heaviness.
• Heavy tails require correlation structure that aligns rare shocks.

Present a new scenario and ask the learner to predict whether its tail will be Gaussian or heavy, then reveal the simulation.

• Reason about correlation structure in a new setting
• Predict tail behavior
• Compare against a simulation to confirm or revise

Restate the answer directly: correlation aligns rare large steps so they accumulate rather than cancel, and the limit inherits heavy tails.

• Independence → cancellation → Gaussian tail.
• Strong correlation → alignment of extremes → heavy tail.
• The limit's tail is heavier than Gaussian precisely because extremes reinforce instead of cancel.