A rigorous, cost-aware search for genuinely harvestable alpha across crypto, equities, FX, commodities, and volatility / options. Each paper pairs a falsifiable hypothesis with a fully reproducible, net-of-cost backtest and a fixed statistical-rigor bar — purged / embargoed walk-forward validation, deflated Sharpe, probability of backtest overfitting (PBO via CSCV), non-overlapping t-statistics, a mandatory prediction-diversity check, and a benchmark against both the trivial baseline and a +0.3-Sharpe mean-reversion floor. Honest negative results are first-class outcomes. The program builds directly on its predecessor, JEPA-Trader, whose single genuine positive — the volatility risk premium — is the lead for Paper 1.
Is JEPA-Trader's one genuine win (Deribit DVOL beating trailing realized vol) harvestable net of costs across crypto and equities? Finding: the premium is real and statistically robust (BTC t=4.45, deflated Sharpe 0.88, PBO 0) but not a tradable edge — it halves under realistic volatility measurement, has decayed to ~zero since 2024, and (like equity option-writing since 2010) earns no excess over buy-and-hold. Read →
Is perpetual-funding carry harvestable net of costs and the liquidation tail? Finding: the funding-level harvest is a decayed, artifact-Sharpe trade that no longer beats cash — but the cross-sectional carry factor (long-low / short-high funding) is the program's first real net-of-cost, out-of-sample-positive edge (OOS Sharpe ~0.4, deflated Sharpe 0.63), albeit modest and tail-heavy. Read →
Does crypto stat-arb survive realistic costs out-of-sample? Finding: no — a clean honest negative. Daily PCA-residual reversion on liquid coins has no edge even gross; the apparent hourly edge is a stale-price artifact in illiquid coins that dies at their real 50–100bp spreads; and 210 in-sample-selected cointegration pairs lose out-of-sample. Read →
How much of the published factor zoo survives out-of-sample, net of costs? Finding: little. 212 OSAP predictors decay 53% post-publication; in the last decade the median has a Sharpe of 0.22 and only 3% are statistically strong; realistic turnover costs erase most of the rest. Only low-turnover quality (and the market) survive. Read →
Can a non-HFT capture microstructure edge? Finding: standalone market-making is structurally unavailable (SPY's half-spread is 0.09bp — a sub-0.1bp game) and order-flow imbalance predicts the present, not the future — but microstructure pays twice: a micro-price execution overlay cuts ~13% of slippage, and vol-conditional reversal (the liquidity-provision premium) earns Sharpe 0.42→1.16 as VIX rises. Read →
The most-documented anomalies in finance, on newly-sourced data. Finding: FX carry is real but modest and decaying (0.64→0.2, crash-prone); FX value steadier (~0.46); energy roll yield has no net edge. The finale tallies all six papers: every edge that survived honest costs is a modest risk premium (~0.4 Sharpe) — never the spectacular backtest alpha. Read →
Combining the survivors into one diversified book. Finding: the four long-history sleeves (FX carry/value, equity quality, liquidity provision) are near-uncorrelated, so risk parity lifts an average 0.40 Sharpe to a combined 0.53 — above the QSPIX 0.41 ceiling (1.33× diversification). The free lunch is real — but the premia are decaying (2020s ≈ 0.13). Read →