a Zj @sdZddlZddlZddlZddlmZejjdddeedZ dd Z ed d Z e D]\Z e e d ddd 4ZeeD]Zeddkrqe edZedusedkrqedpdZdD]Ze edeZe ededZe ededZe edeZe edededefZedurerererdekrjekrjdkrnnede edeeefereed<qqWdq^1s0Yq^e D]Zedjddqd \ZZgZgZ gZ!d!Z"dZ#dZ$e %D]\Z&Zedreds(qede&d"dkr@d#nd$kZ'e(d%d&e)edDdZ*e*durpqede*\Z+Z,ZZZeed'Z-eedZ.ee-e-dZ/eeedZ0ee.e e/e!e0deeZ1e"e're1nd(7Z"e#d7Z#e$e'7Z$qd)d*Z2e2e\Z3Z4Z5e2e \Z6Z7Z7e2e!\Z8Z7Z7e9d+ed,d-ed,d.e#d/e$d0de$e#d,d1 e9d2e3d3d4e4d3d5e5d3e9d6e6d7d8e9d9e8d7d:e9d;e6e8d3d<e9d=de"e#d>d?e"d>d@e#dAe9e"dkrdBndCdS)Da6 Is the Kalshi bid-ask spread small enough that our edge survives paying it? At each signal moment (first PRE snapshot where clv_kalshi in [5,12]% for a side), measure: bid-ask spread (cents), edge vs MID, edge vs ASK (what we pay), and the spread tax = edge_mid - edge_ask. Settle outcomes to get at-ask ROI. N) defaultdictzutf-8replace)encodingerrorszE:\soccer\live_data\*.csvc Cs0zt||dWSttfy*YdS0dS)N)floatget TypeError ValueError)rkr ,E:\soccer\research\analysis\_spread_check.pyfval srcCs gddS)N)rowsftrr r r r rrr)newlinerrZstatusZPREelapsedr)homeawayZ clv_kalshi_Zkalshi_Z_askZ_bidZ pin_fair_Z game_daterrrdT)reverse)g@g(@gHAccs8|]0\}\}}}}}t|kr(tkrnq|VqdS)N)FLOORCAP).0imcbapr r r )rr'cCs6t|}t|}t||||d|t|dfS)Nr(g?)sortedlensumint)xr r r rstats5sr/zsignals (first clv in [z.0f,z]): z , winners z (z%) zbid-ask spread (cents): mean z.2fz median z p90 zedge vs MID : mean z+.2f%zedge vs ASK : mean z&% (this is what the signal/CLV uses)zspread tax : z- pts of edge lost to crossing half the spreadz at-ask flat ROI: z+.1fz% (z units over z bets)z=> edge survives the spreadz=> spread eats the edge):__doc__ZcsvZglobsys collectionsrstdout reconfigurer*filesrZseriesfpopenfZ DictReaderr rZelrZsideZclvZaskZbidZpfdappendvaluessortrrZspreadsZedge_midZedge_askZunitsnZnwinitemskeywinnext enumerateZeir"r#ZmidZsprZemZeaZbor/ZmsZmedsZp90sZmm_Zmaprintr r r rsf    D2    6 (