The quant desk for
prediction markets.

# Does the longshot bias hold outside sports, or is it a football story?
# Loads six product lines and scores each one the same way.

from synthatrade import load, calibration, fees, Backtest
import pandas as pd

SERIES = [
    "sports",               # NFL, NBA, tennis, soccer
    "economics",            # CPI, payrolls, GDP
    "climate",              # temperature and storm thresholds
    "companies",            # earnings and guidance
    "elections",            # races and control
    "culture",              # awards, charts, releases
]

def longshots(book, ceiling=0.20, hours=24):
    # every YES priced under the ceiling, close to resolution
    near = book[book.hours_to_close < hours]
    return near[near.price < ceiling]

rows = []
for name in SERIES:
    book = load(name, since="2023-01")
    sig  = longshots(book)
    if len(sig) < 200:
        continue          # too thin to say anything

    cal = calibration(sig, buckets=10, ci="wilson")
    net = sig.resolved_rate - sig.price - fees(sig.price)

    rows.append(dict(
        series    = name,
        n         = len(sig),
        paid      = sig.price.mean(),
        resolved = sig.resolved_rate.mean(),
        edge      = net.mean(),
        lower    = cal.lower.mean(),
    ))

out = pd.DataFrame(rows).sort_values("edge")
print(out.to_string(index=False))

# and the one that actually survives fees
best = out[out.edge > 0]
Backtest(load(best.series.iloc[-1]), longshots,
         fees="venue").run().calibration_curve()