Author
Order book imbalance predicts price moves because a market maker who knows the true price posts it that way to manage her own inventory risk, not because it leaks information.
Crypto VPIN of 0.47 sounds toxic until you derive the estimator's null value: 0.500. And the Roll measure that drives every prediction is mostly a volatility proxy. One label of five survives.
FX edge sits in the graph: EM stocks lead by a month, G10 loops pay daily, and FX futures Sharpe jumps from 0.16 to 0.66 once other asset classes are wired in.
A belief-vol kernel needs prices that sum to $1 and carry no predictable drift. Polymarket prices sum to $0.60, $39.6M walked out, and a $2 shove still moves a market 60 days later.
KMZ's virtue of complexity survives only under a zero-intercept restriction and a per-draw scoring rule. Fix either and it reverses: 15 predictors beat 12,000, and buy-and-hold beats both.
Four optimizers, 31 seeds, one BTC rule set. The useful output is not which one won, it is that Differential Evolution's population stayed 82% scattered after 1000 generations.
Lotter splits strategy building into model-based and data-mined, then shows a random walk you cannot tell from EUR/USD. His own reality-check slide has a p-value near 0.085.
Two stocks end the month flat, one by recovering, one by fading. The shape between the endpoints predicts next month: low-convexity stocks beat high-convexity by 0.84%/mo, and no factor explains it.
Value, momentum, volatility, and sentiment timing all lost to a plain equal-weight factor basket in China. Theory says timing is huge; estimation error eats it. Trust the plateau, not the peak.
Dual momentum on GLD vs IBIT posts 79.91%/yr at 8 weeks, Sharpe 1.64, DD still -44%. That lookback won a 10-spec in-sample grid. A 20% vol cap leaves 12% at Sharpe 1.37.
Cointegration tests stationarity, not profit. This method optimizes a basket's price swing inside a band directly, finds ten-asset stat-arbs, and a moving band keeps them alive longer out of sample.
Three ML trading papers report 92% and 98.7%. All three fail on their own printed numbers. Six forensic tests, one honest counter-example, and why a do-nothing model beats the headline.