Quant Research and Historical Analysis

Pull historical prediction market trade, position, and PnL data for quant research and backtesting with the Bravado Data API.

The problem

Quantitative research on Polymarket requires clean historical fill data, position snapshots at arbitrary points in time, and PnL series that reconcile with wallet state. Reconstructing this from on-chain logs is possible but expensive: parsing conditional-token events, joining against order-book snapshots, and handling neg-risk mechanics is a several-week engineering project.

What Bravado provides

• Full historical fills for any wallet via GET /traders/{address}/trades with cursor pagination.
• Position reconstruction at any timestamp via GET /traders/{address}/positions with time range filters.
• PnL time series via GET /traders/{address}/pnl with configurable bucketing.
• Universe-level data via GET /trades for global fill data across wallets.
• All PnL computed under PMWAS, so backtests reconcile with the live Analytics endpoints.

APIs used

API	What a research workflow uses it for
Data API	Trade logs, PnL series, and positions for any public wallet, paginated for bulk extraction
UMA API	Resolution outcomes and timing, needed to label historical events correctly

Worked example

Pull all fills for a leaderboard cohort and load them into a dataframe:

import os, requests, pandas as pd

API = "https://bravado-api-k7kaq.ondigitalocean.app"
H = {"Authorization": f"Bearer {os.environ['BRAVADO_API_KEY']}"}

def all_trades(addr):
    rows = []
    cursor_ts, cursor_log = None, None
    while True:
        params = {"limit": 500}
        if cursor_ts:
            params["cursor_ts"] = cursor_ts
            params["next_cursor_log_idx"] = cursor_log

        r = requests.get(f"{API}/traders/{addr}/trades", params=params, headers=H).json()
        rows.extend(r["trades"])
        if not r["next_cursor"]:
            break
        cursor_ts = r["next_cursor"]
        cursor_log = r.get("next_cursor_log_idx")
    return rows

# Top 50 traders by 30-day PnL.
leaderboard = requests.get(f"{API}/leaderboard?window=30d&limit=50", headers=H).json()

frames = []
for t in leaderboard["traders"]:
    trades = all_trades(t["address"])
    df = pd.DataFrame(trades)
    df["wallet"] = t["address"]
    frames.append(df)

all_df = pd.concat(frames, ignore_index=True)
all_df.to_parquet("polymarket_top50_fills.parquet")

The resulting parquet has every fill for every wallet, with columns for symbol, side, size, price, fee, block_timestamp, and market metadata. Feed it directly to a Jupyter notebook or a research feature store.

Common research patterns

• Alpha decay. Track leader wallets’ PnL post-signal and correlate with mirroring latency.
• Fee sensitivity. Simulate strategies at different fee levels using PMWAS cost-basis math.
• Liquidity modeling. Cross-reference fill sizes against depth snapshots to model slippage.
• Neg-risk arbitrage. Detect stale prices across a neg-risk basket where the YES-share sum diverges from $1.

Notes on data

• Numeric fields are strings. Cast to Decimal before arithmetic, not float. See Numeric conventions.
• Cursor pagination is stable against new inserts. Backfill jobs can resume from a persisted cursor.
• Fees are itemized per fill.

Related

• Data API: PMWAS accounting
• Reference: Pagination
• Data API reference