Finding, not used

The variance risk premium

The volatility the options market charges systematically exceeds the one that actually happens — and not only in crypto.

The question: Do people overpay for protection? And is it a crypto phenomenon or a general one?

What was found

Across 1,963 days, Deribit's implied volatility exceeded subsequent realized by 9.96 points on average, positive 73.8% of the time, with a 90% block-bootstrap confidence interval of [+7.59, +12.90] — which excludes zero. And it holds in the holdout: +6.55 points, positive 72% of the time. This isn't an odd crypto anomaly: the S&P 500, Nasdaq and crude oil also pay for selling volatility. Gold doesn't. It has an ordinary, well-known economic explanation: whoever sells insurance charges more than its expected cost, because the risk they take on isn't symmetric.

Bitcoin+6.55Ethereum+2.87S&P 500+2.39Nasdaq 100+1.24Crude oil+5.17Gold-1.72
The premium in the 2023+ holdout, in volatility points, by market. It exists almost everywhere; gold is the exception.analysis/scripts/phase20_eth_replication_ytd_2026-09-07.py

Try it yourself

Why a positive premium in almost every month isn't enough: one bad month eats the year. Drag the three pieces.

Year total = Good months × Average premium + The bad month
Year total (volatility points)

How many good months just to cover the bad one:

With the unfiltered book's real numbers — +2.39 a month and a worst month of −55 — it takes 23 good months to cover one bad one. That is exactly the problem the filter on the previous page tries to solve.

Illustrative example numbers for practice — not real data.

How it was tested, step by step

  1. For each day, the 30-day implied volatility quoted by the options market is taken (the DVOL index in Bitcoin's case).
  2. Those 30 days are allowed to pass and the volatility that actually occurred is computed.
  3. The premium is the subtraction. Its mean, its share of positive days and a BLOCK bootstrap confidence interval are measured — not single days, because consecutive days share information and resampling them separately would understate uncertainty.
  4. It's repeated in markets with decades of history and a quoted implied-volatility index, each with its own data, to see whether the phenomenon is general.
What this does NOT say. That the premium exists doesn't automatically mean collecting it is a good idea: the project also measured that ETH's premium on its own is NOT significant in the holdout (the interval includes zero, 63% winners), and that the ETH−BTC spread runs the opposite way from what was expected (−3.67 points, positive in only 39% of months) — an attempt to improve the finding with a spread between the two failed, which strengthens it by elimination. And the fundamental point: collecting this premium means being short a tail. It's "recorded and not pursued": the clearest hint of where to look next, and it demands a kind of execution this project doesn't have.

Tracking

This finding does have a real number that gets published and kept updated.

Data last updated on 2026-09-09.

← All research