InwyResearch

Research log · 2026

We tested ~30 crypto trading ideas. Two held up.

An open research log. Thirteen studies and some thirty strategy ideas, each tested against real market data before any capital was committed — and the two that survived data they had never seen.

Foundations: funding, momentum and trendWeather marketsMention marketsSelling longshotsHyperliquid OutcomesHyperliquid vaultsBehavioural patterns on perpsStock perpsCrypto price targets vs optionsTrend + carryAlternative dataSmaller venuesCopy-trading
Survived unseen dataRejected, or only useful as a toolHover or tap a tile for its name

The two that survived

Both held up on data they had never seen. Short version first; every claim links to its full study.

Trend + carry

Long Bitcoin while a slow trend filter is up. Otherwise the money earns market-making yield instead of sitting idle.

+25%per year since mid-2025−16%worst drawdown
Read the study →

Selling longshots

Prediction-market traders overpay for unlikely outcomes. We sell to them with resting orders, and it paid in every category.

+4.5%per fill on a year of unseen data8 / 8categories positive
Read the study →

What the combination earned

Growth of 100, mid-2023 to late 2026. A slow trend rule plus market-making yield beat each part alone.

Growth of 100, mid-2023 to late 2026Monthly, after fees and funding. Hover for values.
Trend + carryTrend, parked in cashMarket-making vaultBTC buy-and-hold
01002003004005006002023202420252026Trend + carry 553

Why most ideas died

Ideas are cheap; surviving unseen data is not.

~30ideas tested

Each had a structural or behavioural reason to work. Most were killed by a cheap check on real data.

4pre-registered holdouts

Rules written down first, then run once on data nobody had looked at.

2survived

Only these two were built, and first in dry-run, with no money at risk.

How every idea was tested

Guilty until proven innocent: five gates, and most ideas never got past the second.

  1. 1. IdeaA reason an edge should exist: someone is forced, biased, or paid to take the other side.
  2. 2. Cheap kill-checkReal venue data, a control group, medians and “without the best outcome”.
  3. 3. Live book checkIs the price actually tradable in today’s order book, or the mid-point of an empty one?
  4. 4. Pre-registered holdoutRules committed first. Unseen data touched once.
  5. 5. SurvivorOnly now does anything get built.

All thirteen studies

Each page gives the one-line answer first, then the details.

01Foundations: funding, momentum and trendFunding crowding, cross-sectional momentum and BTC trend following on perps.Trend survived02Weather marketsCan public forecasts beat Polymarket’s daily temperature markets?Rejected03Mention marketsDoes retail overpay for “Will X say Y?” markets?Led to a survivor04Selling longshotsIs the longshot bias collectable as a maker, across the whole platform?Survived05Hyperliquid OutcomesIs the new prediction-market venue a blue ocean?Rejected06Hyperliquid vaultsCan data pick, time or protect allocations to trader vaults?One tool07Behavioural patterns on perpsDo retail patterns carry over to perpetual futures?Rejected08Stock perpsDo 24/7 stock perps overreact while Wall Street is closed?Rejected09Crypto price targets vs optionsCan options-implied odds beat Polymarket’s crypto price targets?Failed holdout10Trend + carryWhat if a trend strategy earns yield instead of sitting in cash?Survived11Alternative dataDo sentiment, liquidity or attention data improve the trend rule?Rejected12Smaller venuesIs liquidity provision better where competition is thinner?Rejected13Copy-tradingDo the best traders keep winning?Rejected

What the failures taught