a ?2ej(@sdZddlZddlZddlZddlZddlmZdddddZdZ dZ d d Z d d Z d dZ ddZdddZddZedkredS)a Does the South American home/draw anomaly survive on THOUSANDS of games? (read-only, trades nothing, downloads public CSVs) WHY THIS EXISTS: research/region_calibration_test.py found, on 73 fixtures of our own data, that Pinnacle appears to under-price SA homes (48.2% -> 60.3% actual) and over-price draws (27.4% -> 16.4%). Suggestive, sign-consistent across 5/5 SA leagues -- but 73 games is nothing, and the observed draw rate contradicted the well-known ~26-28% historical base rate for these leagues. This settles it with football-data.co.uk's /new/ archive, which carries PINNACLE CLOSING odds (PSCH/PSCD/PSCA) -- the same anchor our live pipeline uses -- for thousands of Brazilian and Argentine fixtures. CONTROLS: - Mexico + USA as non-SA western-hemisphere comparisons (same data source, same devig, same code path) so any finding is not an artifact of the pipeline. - Per-league and per-season breakdowns: a real structural effect is stable across seasons; a fluke lives in one or two. - Reports our own 73-fixture sample alongside, to show directly how far it sat from the long-run rate. Usage: python research\sa_history_calibration.py N) defaultdict S.AmericacontrolZBRAZARGZMEXZUSAz*https://www.football-data.co.uk/new/{}.csv))ZPSCHZPSCDZPSCA)ZAvgCHZAvgCDZAvgCAc Cs8zt|}|dkr|ndWSttfy2YdS0dS)N?)float TypeError ValueError)xvr ,E:\soccer\research\sa_history_calibration.py_f(s rcs,dd|D}t|fdd|DfS)NcSsg|] }d|qS)rr .0r r r r 1zfair..csg|] }|qSr r rsr r r3r)sum)oimpr rr fair0srcCs6t||d|}|dkr.||||StdS)Nrnan)mathZsqrtr)npZactualZsdr r r z_for6src stjjt|dd}|dd}g}tt |D] dpJd }|dvr^q:d}tD]&}fdd |D}t|rf|}qqf|sq:t|\}} |t d p| d pd ||d |d |d| dq:|S)N<)Ztimeoutz utf-8-sigreplaceZRes)HDAcsg|]}t|qSr )rget)rcrr r rErzfetch..ZLeagueZSeasonrr)lgseasonrespfhpfdpfaover)urllibZrequestZurlopenURLformatreaddecodecsvZ DictReaderioStringIOr%stripupper ODDS_SETSallrappenddict) coderawZtxtoutr,ZoddsZcolsZvalsZpfr0r r'r fetch;s*   rB c st|}|dkr0t||dd|dddS||dd|dg}dD]t\tfdd|D|}tfd d|D}|d |d d d ||d d dt|||ddqLtd|dS)NZ34sz n=Z5dz -- too few))r-r")r.r#)r/r$c3s|]}|VqdS)Nr rr keyr r Yrzline..c3s|]}|dkVqdS)r,Nr rE)letterr r rHZr dz5.1fz->z (zz+6.2f)z )lenprintrr=rjoin)rZlabelindentrpartsrZcntr )rGrIr lineRs & rRc Csi}tD]l}z.t|||<td|dt||dWqtyr}z td|d|WYd}~qd}~00qtdd|D}dd|D}td t|d t|d ttd d D]r}||vrqtt}||D]}||d |qt |dddddD]\}}t||d|q qttdtt}|D]}||d |q\t |ddD]\} }t|d| qtt|} t dd|D| } t dd|D} t dd|D| } t dd|D}tdtdtd| d | d!d"d#| | d!d"d$| d!d"d#|| d!d" td%| | | d!d&d'|| | d!d&d(ttd)td*| | | d+d!d,d-d+| }d+|| }td.||d+d!d,d/td0t d1d|D| d+d!d"d-dS)2Nz fetched z: z usable fixturesz fetch z FAILED: cSs*g|]"\}}t|dkr|D]}|qqS)rFEEDSrr&r r r r r rjrzmain..cSs*g|]"\}}t|dkr|D]}|qqS)rrSrUr r r rkrzCHEADLINE (implied -> actual, Pinnacle closing, proportional devig)zSOUTH AMERICA (Bra+Arg)zCONTROL (Mex+USA)zBY COUNTRY / LEAGUErr*cSst|d S)Nr)rM)Zkvr r r yrzmain..rFrJzBSOUTH AMERICA BY SEASON (a real effect is stable; a fluke is not)r+izseason css|]}|dVqdS)r-Nr rEr r r rHrzmain..css|]}|ddkVqdS)r,r"Nr rEr r r rHrcss|]}|dVqdS)r.Nr rEr r r rHrcss|]}|ddkVqdS)r,r#Nr rEr r r rHrz%VERDICT vs our 73-fixture live samplez; our sample said: homes 48.2 -> 60.3 draws 27.4 -> 16.4rCz games say: homes rKz.1fz -> z draws z long-run home edge: z+.2fzpp draw edge: Zppz* TRADE ECONOMICS implied by the long run:z* buy every home at Pinnacle fair : ROI rz+.1f%z* buy home+away (fade the draw) : ROI z/% (before any vig -- real books charge ~4-7%)z& avg Pinnacle overround in sample: css|]}|dVqdS)r0Nr rEr r r rHr) rTrBrNrM ExceptionitemsrRrlistr=sortedr)datar?eZsaZctlZbyr r*rr+rZiHZcHZiDZcDZnd_impZnd_actr r r main`sn ",   $&  " r___main__)rC)__doc__Zurllib.requestr1r7r6r collectionsrrTr2r;rrrrBrRr___name__r r r r s$  ;