The Mycelia Signal Volatility Index (MSVI) launched on April 8, 2026 — the first cryptographically signed, pay-per-query composite volatility index for BTC and ETH. Within 24 hours, 96 external IPs had hit the endpoints, including evaluation traffic from Azure and AWS fleets.
One question came up immediately: does it actually work?
We ran two backtests to find out — one against the Oct 2025 crash, one against the Jul–Sep melt-up that preceded it. Then we ran a weight optimisation across 906 combinations. Here's what we found.
MSVI combines five independent signals into a single 0–100 index, updated continuously and signed with Ed25519 on every paid query.
| Component | Weight | Source | Method |
|---|---|---|---|
| Realized Volatility (RV) | 30% | Binance, OKX, Kraken | Parkinson (1980), 30D window |
| Implied Volatility (IV) | 25% | Deribit options book | ATM delta-neutral, OI-weighted |
| Term Structure (TS) | 15% | Deribit options book | 7D/90D IV ratio |
| Funding Rate (FR) | 20% | Binance, Bybit perps | Absolute z-score vs 30D mean |
| Put/Call Ratio (PCR) | 10% | Deribit options book | USD-value OI, z-scored vs 90D |
Every paid response includes a canonical string and an Ed25519 signature verifiable against a published public key. No other crypto volatility index does this.
We reconstructed daily MSVI values from October 1 to December 31, 2025 — the period covering BTC's drop from its all-time high of $126k to a low of ~$84k, a drawdown of approximately 33%.
Four components — RV, IV, TS, and FR — were sourced from live historical APIs: Binance OHLCV, Deribit DVOL, and Binance perpetual funding rates. All freely available.
Historical put/call ratio data at the contract level requires paid access via providers like Tardis.dev or Amberdata. We modelled the PCR component using a curve consistent with documented crash behaviour in prior cycles (Mar 2020, May 2021, Nov 2022) — sharp put buying on the way down, gradual normalisation on recovery. This assumption is clearly flagged in all outputs. Results should be validated with real PCR data before drawing firm conclusions about that component's contribution.
| Date | MSVI | BTC Price | Key driver |
|---|---|---|---|
| Oct 1 | 5.9 | $118,595 | Calm near ATH — all components low |
| Oct 6 | 12.8 | $124,659 | ATH — market euphoric |
| Oct 10 | 43.2 | $112,775 | Crash onset — RV spikes from 1.2 to 64.3 in one day |
| Oct 12 | 60.5 | $114,959 | FR hits 94 — funding rate stress fires |
| Oct 18 | 69.6 | $107,185 | Sustained stress — RV, IV, TS all elevated |
| Nov 22 | 70.7 | $84,740 | Second leg down — IV peaks at 100 |
| Nov 25 | 80.9 | $87,370 | Peak MSVI — FR and TS both max |
| Dec 31 | 30.9 | $87,648 | Normalising — TS collapses as contango returns |
The index moved from 5.9 to 43.2 on October 10 — the first major down day — before BTC had lost 10% from its all-time high.
The peak reading of 80.9 came on November 25, three to four days after BTC's price low of ~$84.7k. This lag is methodologically expected: funding rates and term structure take time to fully reflect spot stress. RV, by contrast, responds within the same daily bar.
By December 31, MSVI had returned to 30.9 as the market stabilised in the $87–92k range — correctly reading the reduction in stress even with BTC still 30% below ATH.
We also ran the index across the final melt-up period — July 1 to September 30, 2025 — as BTC climbed from $105k toward its eventual all-time high.
| Date | MSVI | BTC Price | Key driver |
|---|---|---|---|
| Jul 1 | 51.8 | $105,681 | Elevated RV from prior volatility |
| Jul 11 | 74.1 | $117,528 | FR hits 100 — over-leveraged longs |
| Aug 9 | 28.6 | $116,462 | Calm plateau — IV collapses |
| Sep 5 | 65.2 | $110,660 | TS hits 100 — backwardation during rally |
| Sep 18 | 24.6 | $117,074 | Lowest reading — IV at minimum, calm |
| Sep 30 | 30.9 | $114,049 | Quiet into ATH approach |
Two findings stand out. First, the July funding rate spike to 100 correctly identified over-leveraged long positioning during the melt-up — greed stress, not fear stress. MSVI is designed to capture both. Second, the September term structure spike showed the options market pricing near-term uncertainty even while spot was rising — a signal that proved prescient given the October crash that followed weeks later.
We ran a grid search across 906 weight combinations — every permutation of the five components in 10% increments summing to 100% — and scored each against Pearson correlation with BTC drawdown from ATH.
| Rank | RV | IV | TS | FR | PCR* | Correlation |
|---|---|---|---|---|---|---|
| #1 | 30% | 60% | 0% | 10% | 0% | 0.593 |
| #2 | 20% | 70% | 0% | 10% | 0% | 0.591 |
| #3 | 40% | 50% | 0% | 10% | 0% | 0.587 |
| Live | 30% | 25% | 15% | 20% | 10% | 0.349 |
| Rank | RV | IV | TS | FR | PCR* | Correlation |
|---|---|---|---|---|---|---|
| #1 | 0% | 10% | 0% | 80% | 10% | 0.313 |
| #2 | 0% | 0% | 10% | 80% | 10% | 0.311 |
| #3 | 0% | 10% | 0% | 70% | 20% | 0.308 |
| Live | 30% | 25% | 15% | 20% | 10% | -0.111 |
The two optimal configurations are nearly opposite. The crash period rewards IV-heavy weights — implied volatility tracks fear well. The melt-up period rewards FR-heavy weights — funding rate tracks greed well. No single configuration dominates both.
* PCR note: The zero-PCR configurations dominating the top of both optimiser outputs may partly reflect noise from our modelled assumption. This finding warrants caution and should be re-run with real Deribit options OI data.
The optimiser found better-performing weights for each regime. We chose not to use them.
A regime-specific index is a worse index. If we tuned MSVI to maximise crash correlation, it would underperform during calm markets and melt-ups. If we tuned it for greed detection, it would miss crash signals. The value of a general-purpose volatility index is precisely that it doesn't need to know which regime it's in.
Two 92-day windows are not enough. Any weight optimisation from a single market event is likely overfit. The Oct 2025 crash was driven by a specific macro catalyst — the US-China trade re-escalation on October 10. A different crash with different dynamics might favour different weights.
The PCR assumption contaminates the results. The zero-PCR configurations dominating the top of both optimiser outputs may partly be avoiding noise from our modelled assumption rather than genuinely reflecting PCR's contribution. This finding alone warrants caution in interpreting the optimiser results.
The current weights have a principled basis. RV at 30% reflects that realised volatility is the most direct measure of actual market stress. IV at 25% captures forward-looking fear. FR at 20% captures leverage stress in perpetuals — the dominant trading venue in crypto. TS at 15% provides regime context. PCR at 10% provides options positioning sentiment. These weights reflect the relative information content of each signal, not curve-fitting to historical data.
Multi-component composition is essential. No single signal would have produced MSVI's behaviour across both periods. RV drives the crash onset signal. IV drives the sustained fear signal. FR drives the greed signal during the melt-up. TS provides the regime context. A single-source index would have missed at least one of these.
The index correctly distinguishes fear from greed. During the July melt-up, MSVI elevated on FR stress — over-leveraged longs — not on RV or IV. This is the right behaviour. A pure realised volatility index would not have flagged the leverage risk building during the rally.
The peak lag is a feature, not a bug. MSVI peaked at 80.9 on November 25, three days after BTC's price low. FR and TS are slower-moving signals — they reflect sustained positioning, not instantaneous price moves. A risk model using MSVI would have remained in elevated-stress mode through the recovery, which is the appropriate posture when BTC has fallen 33% and funding rates are still dislocated.
Real PCR data. The put/call ratio component was modelled. We plan to source historical Deribit options OI data via Tardis.dev to replace the assumption and re-run both backtests with verified PCR values.
Longer history. Two 92-day windows across adjacent market regimes is a starting point, not a conclusion. We plan to extend the backtest to cover multiple cycles — including the Nov 2022 FTX collapse and the Mar 2020 COVID crash — once historical options data is available.
Both will be published when complete.
Free preview endpoint — no payment or account required. Paid queries return a cryptographically signed Ed25519 attestation verifiable against our published public key.
Mycelia Signal is an independent oracle publishing cryptographically signed price, economic, and volatility data. All endpoints are pay-per-query with no subscriptions or accounts required. myceliasignal.com