YIELDMAX TIDE — Whitepaper v2.0.0

Noethea Dynamics · September 2026


Abstract

YieldMax TIDE is a deterministic funding-rate harvester for BTC perpetual futures. The strategy holds a delta-neutral book — long spot BTC, short perp BTC in equal notional — and captures the funding payments shorts receive every eight hours. It predicts nothing, contains no indicators, and generates identical results on identical data.

Backtested over 78 months of hourly data (2020-03 to 2026-09), TIDE delivered +55.6% CAGR, Sharpe 2.34, max drawdown −9.2%, and 70 of 78 positive months, with a holdout-period return of +2.13%/month. All numbers are reproducible from two CSVs. Zero randomness in the engine.

The strategy passes 2 of 4 pre-registered lock-battery tests. The two failures are disclosed in full, not hidden. This document treats the failures as load-bearing.


1. The Thesis

Bitcoin perpetual futures have been structurally positive on funding — 80 to 93% of hours in every year from 2020 to 2026. When funding is positive, longs pay shorts. Holding a short perp position therefore earns income from market structure rather than from price direction.

TIDE is the mechanical expression of this observation: hold spot, short perp, collect the payment, repeat.

The thesis is falsifiable. If BTC funding inverts persistently (funding negative more than 50% of hours for a sustained period), the strategy produces no income and should be shut off.


2. Mechanism

Two legs, matched in notional:

Leg Position Purpose
Spot Long BTC Owns the underlying
Perp Short BTC Neutralizes price, receives funding

Price P&L cancels: if BTC rises, spot gains offset perp losses. If BTC falls, the reverse. The net directional exposure is zero.

Income source: funding rate paid every 8 hours (00:00, 08:00, 16:00 UTC) by perpetual longs to shorts. This is the entire alpha.

Residual risk: the two legs do not cancel perfectly in the short term. Spot-perp basis widens during volatility, producing small losses that are the source of the strategy's drawdown.


3. Operations

Parameter Value
Leverage Fixed 5x (exchange cap)
Roll cadence Monthly, with 25% drift guard
Cooldown 8% hourly perp range → drop to 3x for 24h
Funding signal Embargoed 8h (no lookahead)
Margin Portfolio netting (requires portfolio margin)
Roll cost Taker 0.05%, slippage 0.01%
Borrow cost 0.02% per day

Why fixed 5x: adaptive-leverage variants were tested and failed clean holdout. Fixed leverage removes a degree of freedom that was previously used to cherry-pick performance.

Why the cooldown: the 8% hourly-range rule approximates a liquidation guard. At 5x, an 8% adverse move consumes 40% of margin; the cooldown reduces size to 3x for 24 hours when that threshold is hit.


4. Data and Methodology

Backtest window: 2020-03 → 2026-09 (78 months).

Data sources:

Basis modeling: v1 of this strategy used synthetic basis — np.random.normal(0, 5%) hourly noise. A 200-seed sweep revealed that v1's reported numbers were an artifact of the specific random seed used. v1 is dead.

v2 replaces synthetic noise with measured Binance spot-vs-perp hourly basis (mean −1.76 bps, standard deviation 6 bps, extremes ±1.9%, with six hours exceeding 0.5% in six years). The two-leg book is valued tick-for-tick against measured prices.

Reproducibility: every number in this document can be regenerated by running yieldmax_tide.py against the two provided CSVs. There is no random number generator in the engine.


5. Results

Full period (2020-03 → 2026-09, 78 months)

Metric Value
CAGR +55.6%
Sharpe ratio 2.34
Max drawdown −9.2%
Positive months 70 / 78
$10,000 → $175,099
Holdout (2024+) monthly +2.13%
Train (2020–2023) monthly +5.14%

Lock battery

TIDE was evaluated against a pre-registered 4-test battery. The rules were set before the run and were not modified afterward.

Test What it measures Result
Q1 Return / Risk PASS — train +5.14%, holdout +2.13%, DD −9.2% (limit −17%)
Q2 Walk-forward stability FAIL — 5/10 (threshold 8/11)
Q3 2× cost stress FAIL — 1.78%/mo (threshold 2%)
Q4 Regime coverage PASS — 4/5 regimes won

Battery: 2/4. Both failures are analyzed below.


6. Failure Analysis

Q2 — Walk-forward stability

Q2 divides the holdout into 3-month windows and requires each window's z-score (mean ÷ standard deviation) to be less than 2 in absolute value. TIDE fails 5 of 10 windows.

The cause is test design, not strategy instability. Measured basis is so tight that calm windows produce almost no variance. Window 8 (2026) shows a standard deviation of 0.08% with a z-score of 20.6 — the strategy earned ~2%/month consistently, and the test penalizes the consistency.

Q2 was designed for volatile strategies where "stability" means similar returns with healthy variance. For a genuine carry strategy, low variance is the objective, not a defect.

This failure is diagnosed but not waived. A correct remedy is to pre-register a new Q2 (rolling Sharpe, or "positive in ≥ 80% of windows") and run it once. Until then, the fail stands.

Q3 — 2× cost stress

Q3 reruns the strategy with all costs doubled:

Under 2× costs, TIDE produces 1.78%/month — 0.22 percentage points below the 2% floor.

This failure is real. At 5x leverage, costs are amplified fivefold. Borrow alone at 0.02%/day on 4× equity equals ~2.4%/month in base cost; at 0.04%/day it becomes ~4.8%. The edge survives, but with less margin.

The base case (1× costs) remains profitable. The 2× case is a stress test, not a forecast.


7. Risk Disclosures

TIDE is a live-edge candidate, not a finished product. The following gaps remain:

1. Exchange mismatch. Funding is sourced from Binance; perp prices are sourced from Bybit. The ±20% robustness of results to this mismatch was checked, but exact cross-venue execution introduces small unmodeled frictions.

2. No ADL modeling. Auto-deleveraging during extreme stress events (FTX, LUNA) is not simulated. ADL can forcibly close positions at adverse prices. Real 2022 dislocations may have triggered ADL that the backtest does not capture.

3. Borrow rate assumption. The model uses a fixed 0.02%/day. Real borrow rates spike during volatility — occasionally to 0.05% or 0.10%/day. At 5x leverage, a sustained borrow spike would reduce returns below the base case.

4. Capacity. The mechanism scales structurally (funding accrues on notional regardless of size), but empirical capacity is not verified. At large size, market impact on both legs would compress the edge.

5. 10x liquidates. The strategy is calibrated to 5x. A 10x variant liquidates on the same data. 5x is two tiers below the liquidation cliff; it should not be treated as a starting point for higher leverage.

6. Paper trade required. The only sterile confirmation available is three to six months on testnet with real order flow. Live funding, real basis, and actual fills may differ from the model in ways the backtest cannot capture.


8. Deployment

Recommended path:

  1. Paper trade on Bybit testnet for 3–6 months. Confirm live funding and basis match the model within ~10%.
  2. Live at $100 minimum. Bybit, Binance, or Bitget all support 5x spot margin within their portfolio-margin products.
  3. Scale only after a profitable quarter at test size.
  4. Kill switch: the live executor halts all trading if equity draws down 15% from peak. Manual review is required before resuming.

Live executor: yieldmax_tide_live.py (paper / testnet / live modes, hard-wired kill switch, minimum $100, all actions logged to tide_live_log.csv).


9. Prior Work (Dead Lines)

TIDE v2 is the only strategy in this workspace that is currently runnable. The following lines were tested, failed, and are documented for completeness:

Line Verdict
HARVEST (adaptive formula) 3/10 holdout — failed clean holdout
Bidirectional 7x Full-period cherry-pick — dead
DEMON (50/50 vol harvester) 1/4 — DD −49.6%
Grid experiment 1/4 — Sharpe 0.55
Alt funding screen (DOGE, XRP, ADA, AVAX, LINK, TON) None qualifies
Hybrid TIDE+DEMON 2/4 — Q3 cannot pass
XS-N1 / XS-ALL (cross-sectional funding) 3/4 and 1/4 — tail too large
SPREAD (cross-venue funding) 0/4 — sub-scale
HL-only carry 2/4 — richer venue, larger tail
HLP transcription (live vault) ~1/4 — fails the same walls
Margin-differential, BitMEX spread, digit contracts Dead pre-build

Engines for dead lines have been removed. Verdicts stand as recorded.


10. Why This Paper Is Short on Claims

Most quant research documents present a strategy at its best. TIDE could have been presented the same way — dropping the Q2/Q3 fails, quoting the backtest maximum, calling it locked.

This paper does not do that, for three reasons.

First, the previous version of this strategy was caught cheating. v1's numbers were the output of one random seed that happened to land favorably. A 200-seed sweep showed the strategy's reported performance was not a measurement. That v1 was killed, not shipped.

Second, the lock battery is the product. A strategy that passes 4/4 by refusing to disclose its misses is worth less than one that passes 2/4 honestly. The battery's value comes from the fact that it can fail, and does.

Third, capital that trusts a false number is capital that is lost. The purpose of this document is not to raise money on optimism. It is to describe what TIDE actually is — its mechanism, its verified performance, its known failure modes — so that anyone deploying it does so with correct expectations.

TIDE is deployable at the numbers disclosed above. It is not deployable at any other number.


11. Files and Reproduction

File Purpose
yieldmax_tide.py Backtest engine, deterministic
yieldmax_tide_live.py Live executor (paper / testnet / live)
tide_plot.py Equity curve generator
btc_hourly_2020.csv Hourly BTC perpetual price + funding
binance_spot_hourly.csv Hourly Binance spot price
tide_equity.csv Output: equity curve
tide_monthly.csv Output: monthly returns
RESEARCH_LOG.md Full research log, dead lines, disclosures

Reproduce:

python yieldmax_tide.py
python tide_plot.py

No API keys required for backtest. Live executor requires exchange keys (testnet keys first).


12. Contact

Noethea Dynamics — research lab.

For questions about this document, replication attempts, or issues found in the methodology, contact the lab directly.


This document is a research note, not investment advice. Past backtested performance does not guarantee future results. All figures are reproducible from the provided data files.

JotBird Logo
Published with JotBird