Follow the question. Examine the evidence.

What this shows. The targeting model has more error than a constant benchmark that knows the true average effect. This test does not support a claim that the model reliably identifies who benefits more.

What this shows. With a budget of 1,000 in each market, the model-based policy averages 64.50 extra trips versus 68.50 for uniform allocation. These are simulated policy outcomes, not field-experiment estimates.

What this shows. Strict as-of uses only information known at the time. Fixed mask keeps the original release availability but imports 5,464 later revisions. Unrestricted latest values also introduce 863 future releases. The two counts can overlap; do not add them.

What this shows. In the final GDP holdout, Elastic Net moves from fourth to first when later revisions replace the original values. AR(1) moves from first to second. Both predictor values and scoring targets are revised, while release availability stays fixed. Eight forecasts (July 2024–April 2026) are too few to establish general model superiority.

What this shows. A one-percentage-point larger rate gap is associated with an 18.3% lower conditional hazard in this model (1 − 0.817). This is a relative association, not an 18.3-point fall in monthly probability. No confidence interval is displayed; this is not a causal estimate.

What this shows. Each point is a rate-gap group, placed at its mean gap. The sample overrepresents loans that exit; these unweighted shares are not population monthly probabilities. The downward pattern has reversals and differs from the adjusted model above.

What this shows. The pre-duty effect stays close to +0.025 log points in all windows, rather than showing a large supplier-side price cut. Duty-inclusive effects stay near +0.15. Unit values are imperfect price proxies.

What this shows. Thin vertical lines show reported 95% intervals. The omitted reference month is −3. Pre-duty values show no sustained offsetting decline; duty-inclusive values rise after treatment. This figure uses the report’s rounded estimates.

What this shows. Each dot is one scenario’s gross edge per unit of turnover, sorted from lowest to highest; 110 of 144 are above zero and the best reaches 2.39 bp. The dashed line marks the fixed 4 bp taker fee. Overlapping scenarios must not be summed.

What this shows. At least 22 of 36 scenarios are gross-positive at each horizon, and 35 of 36 at the 100-event horizon. This is one four-hour exploratory capture, not a test of live profitability or cross-day reliability.

What this shows. Against the unpooled own mean, every model achieves positive OOS R² (+0.83% to +1.16%), but the non-timing regional shrinkage mean alone reaches +1.20% (t = 9.05). Measured against the shrinkage mean, all learned models have negative OOS R².

What this shows. Because learned timing rules turn over 41%–61% per month versus 6.1% for the shrinkage-mean tilt, the shrinkage-mean portfolio achieves the highest net Sharpe (1.91) ahead of static equal weight (1.48) and GBRT (1.26).

What this shows. After residualizing on [1, p, p²], illiquidity (PM-LIQ, IC = −0.072), low volatility (PM-VOL, −0.069), 30-day reversal (PM-MOM, −0.032), and time-to-resolution (PM-TTR, −0.028) remain statistically significant out of sample.

What this shows. Percentage returns near zero, extreme price bounds, same-bar bounce, pre-settlement drift, and stale non-executable quotes inflate the apparent 30-day reversal IC from −0.440 to −0.054 raw (−0.032 price-neutral).

What this shows. Low-price range (+1.63), funding-rate contrarian carry (+1.42), and retail long/short positioning (+1.39) hold up across the 730-day Binance panel and OKX cross-validation, while cross-sectional momentum (−0.38) fails to transfer. The cost formula behind these published Sharpes charged about 2.5 bp per side instead of the stated 5 bp; read them as upper bounds until the rerun.

What this shows. Tracked out of sample without re-fitting after the May 2026 cutoff, the equal-weight composite gained +4.57% in June 2026 before losing −6.93% in July and −3.97% in August (−6.55% cumulative), underscoring regime sensitivity.