Working papersFull dataReproducible

Research

Seven working papers on the boundaries of statistical edge in systematic trading. Pick one.

What we found

Retail memecoin traders get back less than any casino game pays, about 75 for every 100 staked.*

Within-strategy Monte-Carlo permutation testing adds no predictive value to forward strategy selection.†

A long-only, realistically-costed trader has no tradeable edge in DEX-only coins.‡

*Across 44.1 million retail positions held by 6.69 million pump.fun wallets from 8 August to 18 September 2026, retail got back 0.754 per unit staked against 0.947 at American roulette, the worst of six casino games, and only 14.1% of wallets ended ahead. Tests registered before any outcome was computed. Published on SSRN (7676458).

†Lift indistinguishable from zero across 437,911 strategy configurations and 6.63 billion within-strategy permutations on nine instruments spanning crypto, forex and commodities. An earlier negative-lift result was traced to a floating-point summation-order artefact, not a real effect.

‡For a long-only trader using only price-based signals, with realistic fees and slippage, across 27 decentralized-exchange-only chains. Working paper published on SSRN (6858778).

The work, in the open

7
Working papers
all on SSRN
6K+
SSRN downloads
and counting
119.4k
Lines of code
seven papers + the framework
100%
Reproducible
data + code open

SSRN 7676458crypto vs casino odds

Is Retail Crypto Trading Better Than Gambling?

Retail memecoin traders on pump.fun got back 0.754 per unit staked over six weeks in 2026, less than at American roulette or any other casino game tested, and only 14.1% of their wallets ended ahead. Rebuilt from on-chain records with tests registered in advance, Hyperliquid and Polymarket retail paid out at casino rates.

Read the study
Memecoins 0.754Polymarket 0.984Hyperliquid 0.9999casino games0.751.00
Return per unit staked · retail vs 6 games

SSRN 7461901crypto maker adverse selection

Quoting the Touch Does Not Pay Its Adverse Selection

A simulated maker quoting at the back of the queue on Bybit, Binance USD-M and Hyperliquid pays more in post-fill drift than it earns in half-spread, across 9,983 coin-days and 125.8 million fills. Because all three venues publish the true aggressor side, the paper also prices what Lee–Ready and the tick rule cost when that flag is missing.

Read the study
BybitBinanceHyperliquid
3 crypto venues · net < 0 at the touch

SSRN 7157638Aave v3 liquidation risk

Liquidation Without Loss

A 5% de-peg flags $1.38B of one Aave-v3 slice as liquidation-eligible, and the protocol bad debt on the same path is $0. The paper separates eligible debt from executed volume from realized loss on one live block of 17,419 borrowers, derives the depth at which volume could become loss, and sizes the channels that actually convert it.

Read the study
Eligible debt, executed repayment, and a zero bar for protocol bad debt.AAVE V3 · 5% DE-PEGeligibleexecuted$0lostLT(1+b) = 0.9595 < 1

SSRN 7022599maker adverse selection

Adverse Selection Consumes the Touch

How much of a market maker's earned spread is consumed by adverse selection, measured fill by fill on true-aggressor-signed CME futures across 13 roots and 353,387 maker fills, with interactive in-browser figures. Published on SSRN.

Read the study
capture ≈ −adverse
13 CME roots · net ≈ 0 after adverse selection

Cross-market studySSRN 6955879

The Reach of the World Cup Distraction Effect

The World Cup is a clean, planet-scale attention shock, yet the global derivative venues that now carry most trading barely register it. The headline evidence that they do is an artifact of measurement. Across 11 instruments and five market structures, 158 tests turn up nothing that survives correction.

Read the study
attention 8× normalspotperpeqcmdfxall
attention spikes, volume flat

SSRN 6858778DEX-only

No Edge Without Information

Whether a long-only, realistically-costed trader has any price-based edge in decentralized-exchange-only coins, across 4,990 pairs on 27 chains, with interactive in-browser demos. Working paper published on SSRN.

Read the study
cost drag
27 chains, long-only net of costs

SSRN 6636018permutation testing

Predictive Value of Within-Strategy Permutation Tests for Forward Selection

Whether acting on within-strategy Monte Carlo permutation tests improves forward strategy selection, across 437,911 strategy configurations and over 6 billion permutations on nine instruments spanning three asset classes.

Read the study
0
bootstrap lift, centered on zero

Paper in progressopen source

Quant Research Framework

A backtester built twice, a Python reference and a Rust port that agree to within a thousandth, with walk-forward optimisation, regime segmentation, and robustness stress testing built in.

Open the framework
PythonRust
same backtester, two languages · Δ < 1e-3