a j@sdZddlZddlZddlZddlZddlZddlZejddejddddl Z ddl m Z m Z mZddZdd Zd d Zd d ZedkredS)aIs the volatility effect just illiquidity wearing a disguise? Volatile legs fill 20.5% against 7.4% for calm ones (mm_volatility_clean.py). But the leagues that top the volatility table -- CN1, NWSL, CAN, PE1 -- are plausibly the THIN ones, and a thin book moves further per trade simply because less volume is needed to shift it. Volume independently predicts fills at AUC ~0.66, and a book nobody trades cannot complete however much it swings. So the +13pp may be measuring books that are cheap to move and impossible to finish. This splits the two: volatility from live_data, measured strictly BEFORE the first quote volume from the Kalshi market at post time (volume_24h / open interest) fill real trade prints, queue-aware and reports the 2x2. If volatility only works in the low-volume cell, it is illiquidity and worthless. If it works within BOTH volume halves, it is a real and separable signal. NzE:\soccer\mainzE:\soccer\research) load_legs load_movementCACHEc Cs(z t|WSttfy"YdS0dS)N)float TypeError ValueError)xr &E:\soccer\research\mm_vol_vs_volume.py_fs r cCsi}tddddz}t|D]^}|d}|r||vr:q|dpFdsNqt|dt|d t|d d ||<qWd n1s0Y|S) zE{leg_ticker: (volume_24h, open_interest, volume)} at its first quote.z&E:\soccer\research\mm_paper_quotes.csvutf-8replace)encodingerrorsZ leg_tickerZask_sizeZ volume_24hZ open_interestZvolume)v24oivolN)opencsvZ DictReadergetstripr )outZfhrtr r r load_volume#s      *rc CsV|s|ddStdd|D}|ddt|dd|dd d |t|d d S) Nz>18z - -css|]}|dVqdSfilledNr .0rr r r 7zcell.. z>5z legs z>4z filled Y@z>6.1f%)sumlen)rowsZlabelfr r r cell4s  r)c s t}t|}t}tttdd}g|D]4\}}||vs2||vrNq2|d|d}||vrnq2||}|ddur|dn|d}|durq2d\} } ||D]\} } } t | }|sqD|d}|j du|j dukr|j r| n |j |j d }||krqD| |d d kr| | 7} | |d krd } qDq||d|t| |ddq2tdkrtdtddSdtddDt}ttd|ddt}tt|fddd}ddt|Dtt|fddd}d dt|Dtfd!dt|D}d"d#||||d"}td$|d%td&td't|d(d)t|d*krd+nt|d,krd-nd.d/td0dDtd1dDtd2d3d(d4d5d/td6i}d7fd8dfd9fd:dffD]n\}d;fddffD]B\}fd?d@D}||||f<tt||dA|qNq&tdBdCD]}|d7|f|d9|f} }t| dDkst|dDkrqdtdEd| Dt| }dtdFd|Dt|}td'|dGdH|dIdJ|dIdK||dLdM qtdNdOD]}||d;f||d=f}}t|dDksZt|dDkrqZdtdPd|Dt|}dtdQd|Dt|}td'|dRdS|dIdT|dIdK||dLdM qZdS)UNr )r|tsrr)gFko)tzinfoZour_askg& .>ZqueueTsdr)r.liqrrPzonly z legs join -- too fewr#css|]}|dVqdSrr rr r r r br!zmain..z= legs with movement, volume AND queue-aware fills; base fill z.1fz% cs |dSNr.r ir'r r hr!zmain..)keycSsi|]\}}||qSr r rkr3r r r ir!zmain..cs |dSNr/r r2r4r r r5jr!cSsi|]\}}||qSr r r7r r r r9kr!c3s"|]}||dVqdS)Nr )rr3)rk_lqrk_sdr r r lr!z#Spearman rho(volatility, volume) = z+.3fzE strongly NEGATIVE would mean volatile == thin, i.e. the same thing.z z.2fz is g?z'strong -- they are largely one variableg333333?Zmoderatezweak -- they are separable css|]}|dVqdS)r.Nr rr r r r sr!css|]}|dVqdS)r/Nr rr r r r tr!zsplitting at median volatility dzc and median 24h volume z,.0fz=== the 2x2 ===calmcs |dkSr1r rsd_medr r r5zr!VOLATILEcs |dkSr1r rCrDr r r5{r!thincs |dkSr:r rClq_medr r r5|r!LIQUIDcs |dkSr:r rCrHr r r5}r!cs g|]}|r|r|qSr r r)lselvselr r ~r!zmain..z + z< === does volatility still help WITHIN each volume half? ===)rGrJ css|]}|dVqdSrr rr r r r r!css|]}|dVqdSrr rr r r r r!z>7z: calm z>5.1fz% -> volatile z% z+.1fZppz@ === and does volume still help WITHIN each volatility half? ===)rBrFcss|]}|dVqdSrr rr r r r r!css|]}|dVqdSrr rr r r r r!z>8z: thin z% -> liquid )rrrjsonloadrritemsZ isoformatSZ_ptr-Z astimezoner appendintr&printr%sortedrange enumerateabs statisticsZmedianr)) ZlegsZmvZvolsZtradesrvr6Zvvr/ZcumrZts_spcr+r,basenZrsZrs2Zd2ZrhoZquadZvlabZllabZselZrcrvZt_Zl_ZrtZrlr )rIrKr<r=r'rErLr main<s      &      ra__main__)__doc__r collectionsrOosrZsyspathinsertZ mm_settlerRZmm_volatility_cleanrrrr rr)ra__name__r r r r s0Z