alexnguyen25/patchoracle-skyblock-mayor-cells

SkyBlock Mayor Election Event Study — 42 pre-registered cells

What: the complete adjudicated output of a pre-registered event study asking whether Hypixel SkyBlock mayor elections move the price of thematically-linked bazaar item baskets. One row per (basket × horizon × scope) cell — all 42, including the 33 that failed.

Why publish the failures. The gate, the baskets and the horizons were fixed before any outcome was computed, and 33 of 42 cells were CUT. A table of only the 9 survivors would look like a strong result and would be a multiple-comparisons artifact. The denominator is the finding.

Size

42 rows. 9 PASS, 33 CUT. Of the 9 passes, 2 are executable long-only (SkyBlock's bazaar has no shorting, and post-office horizons require holding through the term).

How the gate worked

• Baskets were pre-registered, not selected post-hoc. Picking items because they look related to a winning mayor is exactly the construction this design exists to avoid.
• 2,000 permutations, mayor-label permutation null and a random-basket null, FWER-corrected across all 42 cells. p_bh_fdr is reported alongside for readers who prefer FDR.
• A candidacy control separates winning from running: the fishing-bait basket is +12.37% when Marina wins and +0.50% (t=0.53) when she is merely on the ballot and loses.

Fields

Column	Meaning
basket	pre-registered item basket (e.g. marina_bait, diana_goldsink)
mayor	the mayor whose win the cell is conditioned on
horizon	H_PRE1 (93-hour voting window), H_P1, H_P5
horizon_meaning	plain-English expansion of the above
scope	cell scope as pre-registered
direction	registered sign of the hypothesised effect (+1 up-leg, −1 down-leg)
n_events	number of historical elections contributing
mean_abn_pct	mean market-adjusted % move, winsorized ±50
t_stat	t statistic
p_permute_raw, p_permute_fwer	permutation p, raw and family-wise corrected
p_random_basket	p against the random-basket null
p_bh_fdr	Benjamini–Hochberg FDR
verdict	PASS / CUT under the pre-committed gate
executable_long_only	whether the cell is actually tradeable (see below)
g18_tradeability_note	the correction that applies to this cell

⚠️ Statistics are sound; tradeability is not. Read this before acting on any row.

The two executable cells were subsequently audited and most of their apparent profitability is bid-ask spread capture, not mayor alpha:

• Instant execution (crossing the book) nets −27.5%.
• Resting orders net +40.3% — but the identical strategy on random windows with no election at all nets +38.4%. So ~92% of it is market-making on wide-spread items. marina_bait faces a 44% median spread against a +12.37% abnormal return.
• The mayor-specific increment is +2.96pp ± 1.44pp against a 1.25% bazaar tax. With an MDE of 4.0pp at 53 Marina terms, that is BLOCKED — not established, and not refuted either.
• A previously published "100% fill rate, 0.00pp adverse selection" figure was a tautology (its threshold was a member of the set being minimised over) and is retracted. The real entry fill rate is 52.4% with −2.77pp adverse selection.
• Volume-weighting was suspected twice of inflating this signal and did not: equal- vs volume-weighted differ by 0.64pp against a 0.96pp bootstrap SE, and no observation has zero bid volume.

mean_abn_pct is a measured market reaction, not a strategy return.

Caveats

• 53 Marina / 58–63 Diana terms — right at the threshold where a cluster bootstrap turns mildly anti-conservative, so intervals are if anything slightly narrower than truth.
• Returns are winsorized at ±50%, and that clip is load-bearing: the patient round trip saturates it on 72.6% of observations, so levels computed under it are clip-dependent. The share attributable to spread (~92%) is robust across clips of 50/100/200; the level is not.
• Prices come from CoflNet SkyApi. This release contains only derived per-cell aggregates, never their raw series.

License & attribution

CC-BY-SA 4.0. Credit: Hypixel SkyBlock Wiki (Fandom), CoflNet SkyApi, Hypixel API. Not affiliated with Hypixel. Built by the PatchOracle project.

Companion datasets

• patchoracle-skyblock-patch-events — 2,078 patch→price events, the main dataset.
• patchoracle-skyblock-patches — the 396-patch corpus metadata.