VRP Short Volatility

Every directional strategy we tested — momentum, mean-reversion, breakout, trend, and every combination across timeframes and coins — reduced to market beta once we subtracted the market’s own return. Win rate could be dialed anywhere from 20% to 87% by changing the exit, but expectancy stayed negative. (See why backtests fail.)

VRP Short Volatility was different. It was the one edge that survived a beta-strip — because it does not predict direction at all. It harvests the volatility risk premium: option buyers systematically overpay for protection, so the seller of that protection earns a premium over time.

RETIRED (2026-07-22). The premium below is measured and still real. What failed is the vehicle — a 7DTE iron condor on real Deribit bid/ask earns a negative Sharpe before cost. The 2026-06 case for this strategy rested on a proxy model that our own audit refuted the same month. Read the status section for the numbers and the file:line of each verdict. This page stays up because the failure is the finding.

1. Sell a defined-risk iron condor on ETH (OKX; one ETH contract is 0.1 ETH against 0.01 BTC for a BTC contract, so ETH sizes about 3.2× finer — see the status note below) — sell an out-of-the-money call spread and put spread simultaneously.
2. Collect the premium upfront. As long as price stays inside the range until expiry, you keep it.
3. Defined risk — the long wings cap the maximum loss. No naked exposure, ever.
4. Weekly cycle — roll each week. That was the intended compounding: many small, uncorrelated bets. On real quotes it did not survive the round-trip cost (see status).

You are not betting up or down. You are the insurance company — you win when the world is calmer than the premium implied.

Implied volatility (what option buyers pay) consistently runs above realized volatility (what actually happens). On BTC over 5 years of Deribit DVOL data, IV averaged ~61% vs realized ~53% — a +8.4 percentage-point premium, positive 79% of the time. That gap is the insurance premium. Sellers of volatility collect it; buyers pay it for peace of mind.

This return is uncorrelated with market direction (correlation to BTC ≈ +0.03) — but our own audit calls that figure a tautology rather than evidence: a symmetric straddle is uncorrelated with direction by construction. The exposure it actually carries is volatility (correlation −0.93 against the realised move). So this is a different risk being taken on, not a direction-free free lunch (research/findings/vrp_audit_bias_20260620.md:23,48).

• Proxy figures (2026-06) — superseded: short-straddle Sharpe ≈ 1.6 and, as a defined-risk condor capped at 10% loss, Sharpe ≈ 2.0. Our own adversarial math audit refuted the 2.0 in the same month: the cap truncated losses without paying for the long wings, so it was a free cap. Re-derived with a true condor payoff and realistic cost, the central estimate is Sharpe 0.92 (research/findings/vrp_audit_math_20260620.md:5,25).
• Real-quote verdict (2026-07-22, OOS-confirmed 2026-07-23): NO-GO. On real Deribit bid/ask snapshots a 7DTE iron condor earns a negative Sharpe before cost (ATM fly −0.09, 1SD condor −0.29) and −0.49 / −0.68 after spread and fees, cumulative −34% / −31% (research/findings/20260722_vrp_condor_realfills.md:3-8). The 2026-06 proxy was overstated by payoff-form bias. A 2026-08-01 pre-screen of a two-leg strangle found the fee objection does not carry over, but it is still not a GO (research/findings/strangle_cost_prescreen_20260801.md:3), and no VRP finding has been filed since.
• Correction (2026-09-06): this section used to say “Gate reached (2026-08-14) … deployment decision pending results,” and the status field above used to read testing. Both were already wrong when written — the real-quote backtest had run on 2026-07-22 and returned NO-GO. The status is now retired (internal status killed). No real capital, and on this evidence the 7DTE condor is not the vehicle.
• Removed (2026-09-06): winRate: 79. That number was never a win rate. It is the share of periods in which the premium was positive (IV above realised) from research/findings/vrp_options_FOUND_20260620.md:17 — a property of the premium series, not of any trade. Rendered as “79% win rate” it promised something no measurement ever made.
• Runs OKX-native on ETH options — an OKX coin-margined ETH option contract is 0.1 ETH (ctVal 1 ETH × ctMult 0.1, read from OKX /api/v5/public/instruments on 2026-09-05). At ETH $2,457 that is about $246 of notional, so a condor capped at about 10% of notional risks about $25 — roughly 0.5% of a $5k account. A BTC contract is 0.01 BTC, about $797 at BTC $79,657 — only about 3.2× larger, so ETH is the finer size rather than the only usable one. The dollar amounts are a 2026-09-05 snapshot and nothing updates them automatically; the contract sizes do not move.
• Correction (2026-09-05) — the bullet above used to read one ETH contract is ~$1,700 notional, so a defined-risk condor risks ~2–3% of a small account. (BTC options on OKX are ~37× larger, too big for small size.) Both numbers were wrong when they were written, not merely stale: $1,700 counted a whole ETH, but a contract is 0.1 ETH; and 37× was the BTC/ETH price ratio on 2026-06-22 (BTC $62,490 / ETH $1,660 = 37.7), not a ratio of contract sizes. The “2–3% of a small account” figure inherited the 10× error. (On OKX’s USD-margined _UM options both contracts are 0.01 coin, and there the size ratio really is the price ratio — the likeliest origin of the 37×.)
• Defined-risk only — iron condors with capped max loss, weekly loss limits, and a consecutive-loss circuit breaker. Never naked.

Because deleting it would be the dishonest move. This page carried the superseded proxy figure and a “decision still to come” line for six weeks after our own backtest had returned NO-GO — the failure worth publishing is not only the strategy’s, it is the lag between a verdict and the screen. The premium in the thesis section is still measured. What we could not do is harvest it through a weekly iron condor at real bid/ask, and no substitute vehicle has cleared a pre-screen yet. If one ever does, it will be a new entry with its own verdict — not this one revived.