Chapter 54
Polymarket Prediction Markets: High-Liquidity Crypto Event Contracts
Notebook43 cells
Polymarket Prediction Markets: High-Liquidity Crypto Event Contracts
Chapter 4: Fundamental and Alternative Data
Docker image: ml4t
Purpose
Polymarket is the world's largest prediction market by trading volume, operating on the Polygon blockchain with USDC settlement. This notebook loads pre-downloaded Polymarket OHLCV data from the centralized data pipeline and compares it with the Kalshi data from the previous notebook to illustrate cross-platform differences in liquidity, pricing, and market structure. An optional live API demo is available for interactive use but disabled by default for automated runs.
Learning Objectives
After completing this notebook, you will be able to:
- Understand Polymarket contract structure and crypto mechanics
- Load and explore pre-downloaded Polymarket OHLCV data
- Compare Polymarket vs Kalshi on overlapping markets (Fed rate decisions)
- Build implied probability indicators from prediction market OHLCV
Cross-References
- Upstream:
data/prediction_markets/download.py(batch downloads) - Downstream: Chapter 8 event features, macro regime indicators
- Related:
12_kalshi_prediction_markets(CFTC-regulated alternative)
In [1]python · cell 2
python
"""Polymarket Prediction Markets - compare crypto-based event contracts with Kalshi for ML feature engineering."""
import re
import warnings
warnings.filterwarnings("ignore")
import plotly.graph_objects as go
import polars as pl
from plotly.subplots import make_subplots
from data.prediction_markets.loader import load_kalshi, load_polymarket
from utils.paths import get_output_dir
from utils.style import COLORSIn [2]python · cell 3
python
# Production defaults - Papermill injects overrides for CI
LIVE = False1. Polymarket Contract Structure
Polymarket uses conditional tokens on the Polygon blockchain. Each market has YES and NO outcome tokens that trade against USDC.
| Feature | Description |
|---|---|
| Settlement | USDC (stablecoin) |
| Position Limit | None |
| Trading | 24/7 |
| Fees | Maker rebates, taker fees (~1-2%) |
| Min Order | ~$1 |
Polymarket vs Kalshi
| Aspect | Polymarket | Kalshi |
|---|---|---|
| Regulation | Unregulated | CFTC-regulated |
| Liquidity | Higher | Lower |
| US Access | Restricted | Yes |
| Settlement | USDC (crypto) | USD (fiat) |
| Position Limits | None | $25,000 |
2. Load Polymarket Data
We load pre-downloaded OHLCV data produced by
data/prediction_markets/download.py. This keeps Chapter 4 notebooks aligned
with the centralized data workflow used elsewhere in the book.
In [3]python · cell 6
python
df = load_polymarket()
print(f"Loaded {len(df):,} observations across {df['symbol'].n_unique()} markets")
print(f"Date range: {df['timestamp'].min()} to {df['timestamp'].max()}")
df.group_by("symbol").len().rename({"len": "days"}).sort("symbol")Output
Loaded 22 observations across 15 markets Date range: 2026-05-01 to 2026-05-02
shape: (15, 2) ┌─────────────────────────────────┬──────┐ │ symbol ┆ days │ │ --- ┆ --- │ │ str ┆ u32 │ ╞═════════════════════════════════╪══════╡ │ BITCOIN-ABOVE-70K-ON-MAY-2:YES ┆ 1 │ │ BITCOIN-ABOVE-72K-ON-MAY-2:YES ┆ 1 │ │ BITCOIN-ABOVE-80K-ON-MAY-2:YES ┆ 2 │ │ BITCOIN-ABOVE-84K-ON-MAY-2:YES ┆ 1 │ │ JEROME-POWELL-OUT-AS-FED-CHAIR… ┆ 2 │ │ … ┆ … │ │ WILL-TESLA-BE-THE-LARGEST-COMP… ┆ 1 │ │ WILL-TESLA-BE-THE-LARGEST-COMP… ┆ 1 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 2 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 2 │ │ WILL-THE-FED-INCREASE-INTEREST… ┆ 2 │ └─────────────────────────────────┴──────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (15, 2)
| symbol | days |
|---|---|
| str | u32 |
| "BITCOIN-ABOVE-70K-ON-MAY-2:YES" | 1 |
| "BITCOIN-ABOVE-72K-ON-MAY-2:YES" | 1 |
| "BITCOIN-ABOVE-80K-ON-MAY-2:YES" | 2 |
| "BITCOIN-ABOVE-84K-ON-MAY-2:YES" | 1 |
| "JEROME-POWELL-OUT-AS-FED-CHAIR… | 2 |
| … | … |
| "WILL-TESLA-BE-THE-LARGEST-COMP… | 1 |
| "WILL-TESLA-BE-THE-LARGEST-COMP… | 1 |
| "WILL-THE-FED-DECREASE-INTEREST… | 2 |
| "WILL-THE-FED-DECREASE-INTEREST… | 2 |
| "WILL-THE-FED-INCREASE-INTEREST… | 2 |
3. Contract Universe
The downloaded dataset contains daily OHLCV bars for curated, non-political
Polymarket contracts. The close price is the implied probability of the YES
outcome. We summarize the downloaded universe before comparing it with Kalshi.
In [4]python · cell 8
python
# Map every contract into the four trading-relevant buckets the chapter
# discusses (monetary_policy, crypto, commodities, geopolitics) using a
# native polars regex chain - no Python UDF. The inference takes
# precedence over the loader's coarser provider category (e.g. an
# "economics" Fed-rate contract maps to monetary_policy here) so that
# the cross-platform Fed comparison below picks it up. Anything that
# doesn't match falls back to the provider category, then to "other".
slug = pl.col("symbol").str.to_lowercase().str.replace_all("-", " ")
inferred_category = (
pl.when(slug.str.contains(r"\b(fed|interest rate|fomc|inflation|cpi|gdp)\b"))
.then(pl.lit("monetary_policy"))
.when(slug.str.contains(r"\b(bitcoin|btc|ethereum|eth|crypto|solana)\b"))
.then(pl.lit("crypto"))
.when(slug.str.contains(r"\b(crude oil|gold|s&p|nasdaq|spy|stock)\b"))
.then(pl.lit("commodities"))
.when(slug.str.contains(r"\b(iran|china|russia|war|ceasefire|ukraine)\b"))
.then(pl.lit("geopolitics"))
.otherwise(None)
)
df = df.with_columns(
pl.coalesce(inferred_category, pl.col("category"), pl.lit("other")).alias("market_category"),
(pl.col("high") - pl.col("low")).alias("intraday_range"),
(pl.col("close") - 0.5).abs().alias("conviction"),
((pl.col("close") > 0.8) | (pl.col("close") < 0.2)).cast(pl.Int8).alias("high_confidence"),
)
contracts = (
df.group_by(["symbol", "market_category"])
.agg(
pl.col("close").first().alias("initial_prob"),
pl.col("close").last().alias("latest_prob"),
pl.col("volume").sum().alias("total_volume"),
pl.col("intraday_range").mean().alias("avg_intraday_range"),
pl.col("timestamp").min().alias("first_date"),
pl.col("timestamp").max().alias("last_date"),
pl.len().alias("observations"),
)
.sort("total_volume", descending=True)
)
contractsOutput
shape: (15, 9) ┌───────────┬───────────┬───────────┬───────────┬───┬───────────┬───────────┬───────────┬──────────┐ │ symbol ┆ market_ca ┆ initial_p ┆ latest_pr ┆ … ┆ avg_intra ┆ first_dat ┆ last_date ┆ observat │ │ --- ┆ tegory ┆ rob ┆ ob ┆ ┆ day_range ┆ e ┆ --- ┆ ions │ │ str ┆ --- ┆ --- ┆ --- ┆ ┆ --- ┆ --- ┆ date ┆ --- │ │ ┆ str ┆ f64 ┆ f64 ┆ ┆ f64 ┆ date ┆ ┆ u32 │ ╞═══════════╪═══════════╪═══════════╪═══════════╪═══╪═══════════╪═══════════╪═══════════╪══════════╡ │ WILL-TESL ┆ technolog ┆ 0.0015 ┆ 0.0015 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ A-BE-THE- ┆ y ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ LARGEST-C ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ OMP… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ BITCOIN-A ┆ crypto ┆ 0.9985 ┆ 0.9985 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ BOVE-72K- ┆ ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ ON-MAY-2: ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ YES ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ WILL-THE- ┆ monetary_ ┆ 0.0075 ┆ 0.0075 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 2 │ │ FED-INCRE ┆ policy ┆ ┆ ┆ ┆ ┆ 1 ┆ 2 ┆ │ │ ASE-INTER ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ EST… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ WILL-SPAC ┆ science ┆ 0.91 ┆ 0.91 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ EX-IPO-BY ┆ ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ -SEPTEMBE ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ R-3… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ BITCOIN-A ┆ crypto ┆ 0.0015 ┆ 0.0015 ┆ … ┆ 0.001 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ BOVE-84K- ┆ ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ ON-MAY-2: ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ YES ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ … ┆ … ┆ … ┆ … ┆ … ┆ … ┆ … ┆ … ┆ … │ │ WILL-OPEN ┆ technolog ┆ 0.034 ┆ 0.034 ┆ … ┆ 0.001 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ AI-HAVE-T ┆ y ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ HE-BEST-A ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ I-M… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ BITCOIN-A ┆ crypto ┆ 0.9995 ┆ 0.9995 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ BOVE-70K- ┆ ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ ON-MAY-2: ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ YES ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ WILL-SPAC ┆ science ┆ 0.0155 ┆ 0.0155 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 1 │ │ EX-IPO-BY ┆ ┆ ┆ ┆ ┆ ┆ 2 ┆ 2 ┆ │ │ -MAY-31-2 ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ 026… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ WILL-BITC ┆ crypto ┆ 0.0205 ┆ 0.0175 ┆ … ┆ 0.0015 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 2 │ │ OIN-HIT-1 ┆ ┆ ┆ ┆ ┆ ┆ 1 ┆ 2 ┆ │ │ 50K-BY-JU ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ NE-… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ WILL-THE- ┆ monetary_ ┆ 0.0405 ┆ 0.0405 ┆ … ┆ 0.0 ┆ 2026-05-0 ┆ 2026-05-0 ┆ 2 │ │ FED-DECRE ┆ policy ┆ ┆ ┆ ┆ ┆ 1 ┆ 2 ┆ │ │ ASE-INTER ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ │ EST… ┆ ┆ ┆ ┆ ┆ ┆ ┆ ┆ │ └───────────┴───────────┴───────────┴───────────┴───┴───────────┴───────────┴───────────┴──────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (15, 9)
| symbol | market_category | initial_prob | latest_prob | total_volume | avg_intraday_range | first_date | last_date | observations |
|---|---|---|---|---|---|---|---|---|
| str | str | f64 | f64 | f64 | f64 | date | date | u32 |
| "WILL-TESLA-BE-THE-LARGEST-COMP… | "technology" | 0.0015 | 0.0015 | 62.0 | 0.0 | 2026-05-02 | 2026-05-02 | 1 |
| "BITCOIN-ABOVE-72K-ON-MAY-2:YES" | "crypto" | 0.9985 | 0.9985 | 62.0 | 0.0 | 2026-05-02 | 2026-05-02 | 1 |
| "WILL-THE-FED-INCREASE-INTEREST… | "monetary_policy" | 0.0075 | 0.0075 | 62.0 | 0.0 | 2026-05-01 | 2026-05-02 | 2 |
| "WILL-SPACEX-IPO-BY-SEPTEMBER-3… | "science" | 0.91 | 0.91 | 62.0 | 0.0 | 2026-05-02 | 2026-05-02 | 1 |
| "BITCOIN-ABOVE-84K-ON-MAY-2:YES" | "crypto" | 0.0015 | 0.0015 | 62.0 | 0.001 | 2026-05-02 | 2026-05-02 | 1 |
| … | … | … | … | … | … | … | … | … |
| "WILL-OPENAI-HAVE-THE-BEST-AI-M… | "technology" | 0.034 | 0.034 | 62.0 | 0.001 | 2026-05-02 | 2026-05-02 | 1 |
| "BITCOIN-ABOVE-70K-ON-MAY-2:YES" | "crypto" | 0.9995 | 0.9995 | 62.0 | 0.0 | 2026-05-02 | 2026-05-02 | 1 |
| "WILL-SPACEX-IPO-BY-MAY-31-2026… | "science" | 0.0155 | 0.0155 | 61.0 | 0.0 | 2026-05-02 | 2026-05-02 | 1 |
| "WILL-BITCOIN-HIT-150K-BY-JUNE-… | "crypto" | 0.0205 | 0.0175 | 61.0 | 0.0015 | 2026-05-01 | 2026-05-02 | 2 |
| "WILL-THE-FED-DECREASE-INTEREST… | "monetary_policy" | 0.0405 | 0.0405 | 61.0 | 0.0 | 2026-05-01 | 2026-05-02 | 2 |
The centralized download focuses on curated, non-political contracts with useful history. This is a better fit for reproducible notebook execution than querying whatever happens to be live on Polymarket at run time.
4. Categorizing Markets
We compare the downloaded Polymarket markets by broad event category.
In [5]python · cell 11
python
category_stats = (
contracts.group_by("market_category")
.agg(
pl.len().alias("n_markets"),
pl.col("total_volume").sum().alias("total_volume"),
pl.col("avg_intraday_range").mean().alias("avg_intraday_range"),
pl.col("latest_prob").mean().alias("avg_latest_prob"),
)
.sort("total_volume", descending=True)
)
category_statsOutput
shape: (4, 5) ┌─────────────────┬───────────┬──────────────┬────────────────────┬─────────────────┐ │ market_category ┆ n_markets ┆ total_volume ┆ avg_intraday_range ┆ avg_latest_prob │ │ --- ┆ --- ┆ --- ┆ --- ┆ --- │ │ str ┆ u32 ┆ f64 ┆ f64 ┆ f64 │ ╞═════════════════╪═══════════╪══════════════╪════════════════════╪═════════════════╡ │ monetary_policy ┆ 5 ┆ 309.0 ┆ 0.0006 ┆ 0.0167 │ │ crypto ┆ 5 ┆ 309.0 ┆ 0.0038 ┆ 0.4123 │ │ technology ┆ 3 ┆ 186.0 ┆ 0.000333 ┆ 0.013 │ │ science ┆ 2 ┆ 123.0 ┆ 0.0 ┆ 0.46275 │ └─────────────────┴───────────┴──────────────┴────────────────────┴─────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (4, 5)
| market_category | n_markets | total_volume | avg_intraday_range | avg_latest_prob |
|---|---|---|---|---|
| str | u32 | f64 | f64 | f64 |
| "monetary_policy" | 5 | 309.0 | 0.0006 | 0.0167 |
| "crypto" | 5 | 309.0 | 0.0038 | 0.4123 |
| "technology" | 3 | 186.0 | 0.000333 | 0.013 |
| "science" | 2 | 123.0 | 0.0 | 0.46275 |
5. Implied Probability Distribution
The closing price is the market's implied probability of the event occurring. We visualize the latest contract probabilities across the downloaded universe.
In [6]python · cell 13
python
valid_prices = contracts.filter(pl.col("latest_prob").is_not_null())
fig = go.Figure()
fig.add_trace(
go.Histogram(
x=valid_prices["latest_prob"].to_list(),
nbinsx=20,
marker_color=COLORS["blue"],
opacity=0.8,
)
)
fig.update_layout(
title=dict(
text="Downloaded Polymarket contracts cluster at near-consensus 0 or 1",
y=0.95,
),
xaxis_title="Latest Implied Probability",
yaxis_title="Number of Contracts",
xaxis=dict(tickformat=".0%", range=[0, 1.05]),
height=450,
margin=dict(t=90, b=60, l=60, r=40),
)
fig.show()Output
Contracts near 0 or 1 reflect near-consensus outcomes. The contracts between 0.2 and 0.8 are where genuine uncertainty exists and where prediction market signals are typically most useful for feature engineering.
6. Volume vs Intraday Range
In the current snapshot, the downloaded universe has 1-2 daily bars per market, so per-contract total volume is essentially constant across rows (the sum-of-bars proxy clusters in a narrow band) and the corresponding scatter degenerates into a vertical strip. Rather than ship a misleading visual, we report the same information as a table: contracts ranked by average intraday range within each category. When the downloader window is widened (more bars per market), the scatter regains its usual volume-vs-volatility spread.
In [7]python · cell 16
python
category_colors = {
"monetary_policy": COLORS["amber"],
"crypto": COLORS["slate"],
"technology": COLORS["copper"],
"science": COLORS["positive"],
"commodities": COLORS["blue"],
"geopolitics": COLORS["negative"],
"other": COLORS["neutral"],
}
contracts.select(
"symbol",
"market_category",
"total_volume",
"avg_intraday_range",
).sort("avg_intraday_range", descending=True).head(10)Output
shape: (10, 4) ┌─────────────────────────────────┬─────────────────┬──────────────┬────────────────────┐ │ symbol ┆ market_category ┆ total_volume ┆ avg_intraday_range │ │ --- ┆ --- ┆ --- ┆ --- │ │ str ┆ str ┆ f64 ┆ f64 │ ╞═════════════════════════════════╪═════════════════╪══════════════╪════════════════════╡ │ BITCOIN-ABOVE-80K-ON-MAY-2:YES ┆ crypto ┆ 62.0 ┆ 0.0165 │ │ JEROME-POWELL-OUT-AS-FED-CHAIR… ┆ monetary_policy ┆ 62.0 ┆ 0.003 │ │ WILL-BITCOIN-HIT-150K-BY-JUNE-… ┆ crypto ┆ 61.0 ┆ 0.0015 │ │ WILL-OPENAI-HAVE-THE-BEST-AI-M… ┆ technology ┆ 62.0 ┆ 0.001 │ │ BITCOIN-ABOVE-84K-ON-MAY-2:YES ┆ crypto ┆ 62.0 ┆ 0.001 │ │ WILL-TESLA-BE-THE-LARGEST-COMP… ┆ technology ┆ 62.0 ┆ 0.0 │ │ BITCOIN-ABOVE-72K-ON-MAY-2:YES ┆ crypto ┆ 62.0 ┆ 0.0 │ │ WILL-THE-FED-INCREASE-INTEREST… ┆ monetary_policy ┆ 62.0 ┆ 0.0 │ │ WILL-SPACEX-IPO-BY-SEPTEMBER-3… ┆ science ┆ 62.0 ┆ 0.0 │ │ WILL-JUDY-SHELTON-BE-CONFIRMED… ┆ monetary_policy ┆ 62.0 ┆ 0.0 │ └─────────────────────────────────┴─────────────────┴──────────────┴────────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (10, 4)
| symbol | market_category | total_volume | avg_intraday_range |
|---|---|---|---|
| str | str | f64 | f64 |
| "BITCOIN-ABOVE-80K-ON-MAY-2:YES" | "crypto" | 62.0 | 0.0165 |
| "JEROME-POWELL-OUT-AS-FED-CHAIR… | "monetary_policy" | 62.0 | 0.003 |
| "WILL-BITCOIN-HIT-150K-BY-JUNE-… | "crypto" | 61.0 | 0.0015 |
| "WILL-OPENAI-HAVE-THE-BEST-AI-M… | "technology" | 62.0 | 0.001 |
| "BITCOIN-ABOVE-84K-ON-MAY-2:YES" | "crypto" | 62.0 | 0.001 |
| "WILL-TESLA-BE-THE-LARGEST-COMP… | "technology" | 62.0 | 0.0 |
| "BITCOIN-ABOVE-72K-ON-MAY-2:YES" | "crypto" | 62.0 | 0.0 |
| "WILL-THE-FED-INCREASE-INTEREST… | "monetary_policy" | 62.0 | 0.0 |
| "WILL-SPACEX-IPO-BY-SEPTEMBER-3… | "science" | 62.0 | 0.0 |
| "WILL-JUDY-SHELTON-BE-CONFIRMED… | "monetary_policy" | 62.0 | 0.0 |
High-volume, high-range contracts are the ones where new information moved probabilities meaningfully - the natural candidates for event features or cross-platform comparison.
7. Probability Evolution
The downloaded OHLCV data lets us inspect how probability evolved through time for the most active contracts.
In [8]python · cell 19
python
# Pick the three contracts whose probability moved most over the snapshot
# (largest |latest_prob - initial_prob|). Picking by total_volume can land
# on contracts that flat-lined near 0 or 1 across both available bars,
# producing a near-blank chart.
movers = (
contracts.with_columns(
(pl.col("latest_prob") - pl.col("initial_prob")).abs().alias("abs_move"),
)
.filter(pl.col("latest_prob").is_not_null() & pl.col("initial_prob").is_not_null())
.sort("abs_move", descending=True)
.head(3)
)
top_contracts = movers["symbol"].to_list()
fig = go.Figure()
palette = [COLORS["blue"], COLORS["amber"], COLORS["copper"]]
for sym, color in zip(top_contracts, palette, strict=False):
data = df.filter(pl.col("symbol") == sym).sort("timestamp").to_pandas()
label = sym[:50] + "..." if len(sym) > 50 else sym
fig.add_trace(
go.Scatter(
x=data["timestamp"],
y=data["close"],
mode="lines+markers",
name=label,
line=dict(color=color, width=2),
)
)
fig.update_layout(
title=dict(
text="Top-mover Polymarket contracts stayed near 0 over the 24-hour snapshot",
y=0.97,
),
xaxis_title="Date",
yaxis_title="Implied Probability",
yaxis=dict(tickformat=".0%", range=[0, 1.05]),
height=500,
margin=dict(t=160, b=60, l=60, r=40),
legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="center", x=0.5),
)
fig.show()Output
The snapshot bundled with the book covers a 24-hour window with one or two
daily bars per contract, so the trajectories appear close to flat: the
downloader is configured for a reproducible smoke test, not a long history.
Widen the window via the downloader's --max-markets-per-category and
--search-results-per-query flags to surface multi-week histories. The
most informative contracts are those that both moved materially and stayed
away from the trivial 0 or 1 endpoints for part of the sample.
8. Optional Live Market Discovery Demo
The main notebook path should use the centralized downloads. For interactive
exploration, set LIVE=True via papermill or in the notebook to inspect the
current Polymarket market universe without affecting routine test runs.
In [9]python · cell 22
python
live_snapshot = None
if LIVE:
from ml4t.data.providers.polymarket import PolymarketProvider
provider = PolymarketProvider()
live_markets = provider.list_markets(active=True, closed=False, limit=200)
provider.close()
live_snapshot = pl.DataFrame(
{
"slug": [m.get("slug", "") for m in live_markets],
"question": [m.get("question", "")[:80] for m in live_markets],
"volume": [float(m.get("volume", 0)) for m in live_markets],
"liquidity": [float(m.get("liquidity", 0)) for m in live_markets],
}
).sort("volume", descending=True)
print(f"LIVE=True: fetched {len(live_snapshot)} active Polymarket markets")
else:
print("LIVE=False: skipping live Polymarket API demo")
live_snapshot.head(10) if live_snapshot is not None else NoneOutput
LIVE=False: skipping live Polymarket API demo
9. Fed Rate Markets - Cross-Platform Comparison
Both Polymarket and Kalshi offer binary contracts on Federal Reserve rate decisions. This creates a natural cross-platform comparison. We load the Kalshi OHLCV data from disk (see notebook 12) and compare it with downloaded Polymarket monetary-policy contracts.
In [10]python · cell 24
python
kalshi_df = load_kalshi()
print(f"Kalshi: {len(kalshi_df):,} observations, {kalshi_df['symbol'].n_unique()} contracts")
print(f"Date range: {kalshi_df['timestamp'].min()} to {kalshi_df['timestamp'].max()}")Output
Kalshi: 687 observations, 9 contracts Date range: 2025-10-07 to 2025-12-31
In [11]python · cell 25
python
fed_markets = contracts.filter(pl.col("market_category") == "monetary_policy")
print(f"Polymarket Fed/Macro markets: {len(fed_markets)}")
fed_markets.select("symbol", "latest_prob", "total_volume", "avg_intraday_range")Output
Polymarket Fed/Macro markets: 5
shape: (5, 4) ┌─────────────────────────────────┬─────────────┬──────────────┬────────────────────┐ │ symbol ┆ latest_prob ┆ total_volume ┆ avg_intraday_range │ │ --- ┆ --- ┆ --- ┆ --- │ │ str ┆ f64 ┆ f64 ┆ f64 │ ╞═════════════════════════════════╪═════════════╪══════════════╪════════════════════╡ │ WILL-THE-FED-INCREASE-INTEREST… ┆ 0.0075 ┆ 62.0 ┆ 0.0 │ │ JEROME-POWELL-OUT-AS-FED-CHAIR… ┆ 0.0275 ┆ 62.0 ┆ 0.003 │ │ WILL-JUDY-SHELTON-BE-CONFIRMED… ┆ 0.0035 ┆ 62.0 ┆ 0.0 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 0.0045 ┆ 62.0 ┆ 0.0 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 0.0405 ┆ 61.0 ┆ 0.0 │ └─────────────────────────────────┴─────────────┴──────────────┴────────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (5, 4)
| symbol | latest_prob | total_volume | avg_intraday_range |
|---|---|---|---|
| str | f64 | f64 | f64 |
| "WILL-THE-FED-INCREASE-INTEREST… | 0.0075 | 62.0 | 0.0 |
| "JEROME-POWELL-OUT-AS-FED-CHAIR… | 0.0275 | 62.0 | 0.003 |
| "WILL-JUDY-SHELTON-BE-CONFIRMED… | 0.0035 | 62.0 | 0.0 |
| "WILL-THE-FED-DECREASE-INTEREST… | 0.0045 | 62.0 | 0.0 |
| "WILL-THE-FED-DECREASE-INTEREST… | 0.0405 | 61.0 | 0.0 |
10. Two Ways to Trade the Fed: Kalshi Ladder vs Polymarket Events
The two platforms cover Fed policy very differently, and the contrast is the point. Kalshi offers many threshold contracts per FOMC meeting (rate above 0.25%, 2.25%, 3.75%, ...), which together form a full implied-probability distribution over rate outcomes - so its prices span roughly 1% to 95%. Polymarket's Fed markets are instead a handful of discrete event bets (a 25bp cut after the June meeting, Powell out as chair, Shelton confirmed), and in this snapshot every one is a low-probability tail event under 5%. That is why the panels below use separate x-scales: forcing them onto one axis would render the Polymarket bets invisible and imply a false like-for-like comparison.
In [12]python · cell 27
python
kalshi_latest = (
kalshi_df.sort("timestamp")
.group_by("symbol")
.agg(
pl.col("close").last().alias("kalshi_prob"),
pl.col("volume").sum().alias("kalshi_volume"),
pl.col("timestamp").max().alias("last_date"),
)
.sort("symbol")
)
kalshi_latestOutput
shape: (9, 4) ┌───────────────────┬─────────────┬───────────────┬────────────┐ │ symbol ┆ kalshi_prob ┆ kalshi_volume ┆ last_date │ │ --- ┆ --- ┆ --- ┆ --- │ │ str ┆ f64 ┆ f64 ┆ date │ ╞═══════════════════╪═════════════╪═══════════════╪════════════╡ │ KXFED-27APR-T0.25 ┆ 0.92 ┆ 0.0 ┆ 2025-12-31 │ │ KXFED-27APR-T2.25 ┆ 0.74 ┆ 0.0 ┆ 2025-12-31 │ │ KXFED-27APR-T3.75 ┆ 0.14 ┆ 2.0 ┆ 2025-12-31 │ │ KXFED-27APR-T4.00 ┆ 0.06 ┆ 0.0 ┆ 2025-12-31 │ │ KXFED-27APR-T4.25 ┆ 0.01 ┆ 0.0 ┆ 2025-12-31 │ │ KXFED-27JAN-T1.00 ┆ 0.95 ┆ 0.0 ┆ 2025-12-31 │ │ KXFED-27JAN-T3.00 ┆ 0.39 ┆ 312.0 ┆ 2025-12-31 │ │ KXFED-27MAR-T0.50 ┆ 0.95 ┆ 1.0 ┆ 2025-12-31 │ │ KXFED-27MAR-T2.75 ┆ 0.51 ┆ 0.0 ┆ 2025-12-31 │ └───────────────────┴─────────────┴───────────────┴────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (9, 4)
| symbol | kalshi_prob | kalshi_volume | last_date |
|---|---|---|---|
| str | f64 | f64 | date |
| "KXFED-27APR-T0.25" | 0.92 | 0.0 | 2025-12-31 |
| "KXFED-27APR-T2.25" | 0.74 | 0.0 | 2025-12-31 |
| "KXFED-27APR-T3.75" | 0.14 | 2.0 | 2025-12-31 |
| "KXFED-27APR-T4.00" | 0.06 | 0.0 | 2025-12-31 |
| "KXFED-27APR-T4.25" | 0.01 | 0.0 | 2025-12-31 |
| "KXFED-27JAN-T1.00" | 0.95 | 0.0 | 2025-12-31 |
| "KXFED-27JAN-T3.00" | 0.39 | 312.0 | 2025-12-31 |
| "KXFED-27MAR-T0.50" | 0.95 | 1.0 | 2025-12-31 |
| "KXFED-27MAR-T2.75" | 0.51 | 0.0 | 2025-12-31 |
In [13]python · cell 28
python
def _short_kalshi_label(sym: str) -> str:
# KXFED-27APR-T0.25 -> "Apr'27 >=0.25%"
parts = sym.split("-")
meeting = parts[1] if len(parts) > 1 else sym
thr = parts[-1][1:] if parts[-1].startswith("T") else ""
m = re.match(r"(\d{2})([A-Z]{3})", meeting)
stem = f"{m.group(2).title()}'{m.group(1)}" if m else meeting
return f"{stem} ≥{thr}%" if thr else stem
def _short_poly_label(sym: str) -> str:
# Turn a Polymarket slug into a short, distinct human label.
s = sym.replace(":YES", "")
up = s.upper()
m = re.search(
r"(JAN|FEB|MAR|APR|MAY|JUN|JUL|AUG|SEP|OCT|NOV|DEC)[A-Z]*-(?:\d{1,2}-)?(\d{4})", up
)
when = f" ({m.group(1).title()} '{m.group(2)[2:]})" if m else ""
bps = re.search(r"(\d+)-BPS", up)
if "DECREASE" in up:
return f"Cut {bps.group(1)}bps{when}" if bps else f"Rate cut{when}"
if "INCREASE" in up:
return f"Hike {bps.group(1)}bps{when}" if bps else f"Rate hike{when}"
if "POWELL" in up and "OUT" in up:
return f"Powell out{when}"
if "SHELTON" in up:
return "Shelton as chair"
words = s.replace("-", " ").split()
return " ".join(w.title() for w in words[:4]) + ("…" if len(words) > 4 else "")
# Horizontal bars so the long market names read cleanly on the y-axis. The two
# platforms are NOT the same contract: Kalshi is a granular rate-threshold ladder
# (a full distribution, ~1-95%); Polymarket's Fed markets are a handful of discrete
# tail-event bets, all well under 5% here - so each panel gets its own x-scale with
# explicit probability labels rather than a shared axis that would hide them.
fig = make_subplots(
rows=1,
cols=2,
subplot_titles=(
"Kalshi: rate-threshold ladder",
"Polymarket: discrete event bets",
),
horizontal_spacing=0.30,
)
kalshi_data = kalshi_latest.sort("kalshi_prob").to_pandas()
fig.add_trace(
go.Bar(
x=kalshi_data["kalshi_prob"],
y=[_short_kalshi_label(s) for s in kalshi_data["symbol"]],
orientation="h",
marker_color=COLORS["blue"],
text=[f"{p:.0%}" for p in kalshi_data["kalshi_prob"]],
textposition="outside",
cliponaxis=False,
),
row=1,
col=1,
)
fig.update_xaxes(range=[0, 1.12], tickformat=".0%", title_text="Implied probability", row=1, col=1)
if not fed_markets.is_empty():
fed_data = (
fed_markets.filter(pl.col("latest_prob").is_not_null()).sort("latest_prob").to_pandas()
)
poly_max = float(fed_data["latest_prob"].max())
fig.add_trace(
go.Bar(
x=fed_data["latest_prob"],
y=[_short_poly_label(s) for s in fed_data["symbol"]],
orientation="h",
marker_color=COLORS["amber"],
text=[f"{p:.1%}" for p in fed_data["latest_prob"]],
textposition="outside",
cliponaxis=False,
),
row=1,
col=2,
)
fig.update_xaxes(
range=[0, poly_max * 1.4],
tickformat=".1%",
title_text="Implied probability (own scale)",
row=1,
col=2,
)
fig.update_layout(
title=dict(
text="Kalshi prices a full rate ladder; Polymarket lists a few low-probability Fed bets",
y=0.97,
),
height=460,
showlegend=False,
margin=dict(t=90, b=60, l=150, r=80),
)
fig.show()Output
[省略较大 image/png 输出]
Kalshi's threshold structure (rate above X%) provides a richer view of the probability distribution over rate outcomes. Polymarket tends to structure the same underlying event as separate yes/no questions. Both encode similar information, but with different contract designs.
11. Volume Comparison
Both datasets expose provider-specific volume fields. The absolute units are not directly comparable, but relative activity still helps identify which contracts matter most within each platform.
Polymarket reports volume as USDC notional traded on the bar; Kalshi reports daily contract count. The two are not directly comparable and the table below should be read column-by-column to rank contracts within each provider, not as a cross-provider volume ratio.
In [14]python · cell 32
python
pl.DataFrame(
{
"provider": ["Polymarket (Fed)", "Kalshi (all KXFED)"],
"volume": [
fed_markets["total_volume"].sum(),
kalshi_df["volume"].sum(),
],
"unit": ["USDC notional (sum of bars)", "contracts (daily OHLCV sum)"],
}
)Output
shape: (2, 3) ┌────────────────────┬────────┬─────────────────────────────┐ │ provider ┆ volume ┆ unit │ │ --- ┆ --- ┆ --- │ │ str ┆ f64 ┆ str │ ╞════════════════════╪════════╪═════════════════════════════╡ │ Polymarket (Fed) ┆ 309.0 ┆ USDC notional (sum of bars) │ │ Kalshi (all KXFED) ┆ 315.0 ┆ contracts (daily OHLCV sum) │ └────────────────────┴────────┴─────────────────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (2, 3)
| provider | volume | unit |
|---|---|---|
| str | f64 | str |
| "Polymarket (Fed)" | 309.0 | "USDC notional (sum of bars)" |
| "Kalshi (all KXFED)" | 315.0 | "contracts (daily OHLCV sum)" |
In [15]python · cell 33
python
fed_markets.select("symbol", "total_volume", "avg_intraday_range").sort(
"total_volume", descending=True
)Output
shape: (5, 3) ┌─────────────────────────────────┬──────────────┬────────────────────┐ │ symbol ┆ total_volume ┆ avg_intraday_range │ │ --- ┆ --- ┆ --- │ │ str ┆ f64 ┆ f64 │ ╞═════════════════════════════════╪══════════════╪════════════════════╡ │ WILL-THE-FED-INCREASE-INTEREST… ┆ 62.0 ┆ 0.0 │ │ JEROME-POWELL-OUT-AS-FED-CHAIR… ┆ 62.0 ┆ 0.003 │ │ WILL-JUDY-SHELTON-BE-CONFIRMED… ┆ 62.0 ┆ 0.0 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 62.0 ┆ 0.0 │ │ WILL-THE-FED-DECREASE-INTEREST… ┆ 61.0 ┆ 0.0 │ └─────────────────────────────────┴──────────────┴────────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (5, 3)
| symbol | total_volume | avg_intraday_range |
|---|---|---|
| str | f64 | f64 |
| "WILL-THE-FED-INCREASE-INTEREST… | 62.0 | 0.0 |
| "JEROME-POWELL-OUT-AS-FED-CHAIR… | 62.0 | 0.003 |
| "WILL-JUDY-SHELTON-BE-CONFIRMED… | 62.0 | 0.0 |
| "WILL-THE-FED-DECREASE-INTEREST… | 62.0 | 0.0 |
| "WILL-THE-FED-DECREASE-INTEREST… | 61.0 | 0.0 |
Within each platform, the busiest contracts tend to be the most informative because they aggregate more views and move more cleanly in response to new information.
12. Event Indicators for ML
The downloaded OHLCV data supports simple event features: implied probability, conviction, intraday range, and a high-confidence flag.
In [16]python · cell 36
python
features = df.select(
[
"timestamp",
"symbol",
pl.col("market_category").alias("category"),
"open",
"high",
"low",
"close",
"volume",
"intraday_range",
"conviction",
"high_confidence",
]
)
print(f"Feature matrix: {features.shape}")
features.select(
"timestamp",
"symbol",
"close",
"conviction",
"high_confidence",
"intraday_range",
"category",
).head(15)Output
Feature matrix: (22, 11)
shape: (15, 7) ┌────────────┬───────────────┬────────┬────────────┬───────────────┬───────────────┬───────────────┐ │ timestamp ┆ symbol ┆ close ┆ conviction ┆ high_confiden ┆ intraday_rang ┆ category │ │ --- ┆ --- ┆ --- ┆ --- ┆ ce ┆ e ┆ --- │ │ date ┆ str ┆ f64 ┆ f64 ┆ --- ┆ --- ┆ str │ │ ┆ ┆ ┆ ┆ i8 ┆ f64 ┆ │ ╞════════════╪═══════════════╪════════╪════════════╪═══════════════╪═══════════════╪═══════════════╡ │ 2026-05-02 ┆ BITCOIN-ABOVE ┆ 0.9995 ┆ 0.4995 ┆ 1 ┆ 0.0 ┆ crypto │ │ ┆ -70K-ON-MAY-2 ┆ ┆ ┆ ┆ ┆ │ │ ┆ :YES ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ BITCOIN-ABOVE ┆ 0.9985 ┆ 0.4985 ┆ 1 ┆ 0.0 ┆ crypto │ │ ┆ -72K-ON-MAY-2 ┆ ┆ ┆ ┆ ┆ │ │ ┆ :YES ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-01 ┆ BITCOIN-ABOVE ┆ 0.0275 ┆ 0.4725 ┆ 1 ┆ 0.0 ┆ crypto │ │ ┆ -80K-ON-MAY-2 ┆ ┆ ┆ ┆ ┆ │ │ ┆ :YES ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ BITCOIN-ABOVE ┆ 0.0445 ┆ 0.4555 ┆ 1 ┆ 0.033 ┆ crypto │ │ ┆ -80K-ON-MAY-2 ┆ ┆ ┆ ┆ ┆ │ │ ┆ :YES ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ BITCOIN-ABOVE ┆ 0.0015 ┆ 0.4985 ┆ 1 ┆ 0.001 ┆ crypto │ │ ┆ -84K-ON-MAY-2 ┆ ┆ ┆ ┆ ┆ │ │ ┆ :YES ┆ ┆ ┆ ┆ ┆ │ │ … ┆ … ┆ … ┆ … ┆ … ┆ … ┆ … │ │ 2026-05-02 ┆ WILL-JUDY-SHE ┆ 0.0035 ┆ 0.4965 ┆ 1 ┆ 0.0 ┆ monetary_poli │ │ ┆ LTON-BE-CONFI ┆ ┆ ┆ ┆ ┆ cy │ │ ┆ RMED… ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ WILL-OPENAI-H ┆ 0.034 ┆ 0.466 ┆ 1 ┆ 0.001 ┆ technology │ │ ┆ AVE-THE-BEST- ┆ ┆ ┆ ┆ ┆ │ │ ┆ AI-M… ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ WILL-SPACEX-I ┆ 0.0155 ┆ 0.4845 ┆ 1 ┆ 0.0 ┆ science │ │ ┆ PO-BY-MAY-31- ┆ ┆ ┆ ┆ ┆ │ │ ┆ 2026… ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ WILL-SPACEX-I ┆ 0.91 ┆ 0.41 ┆ 1 ┆ 0.0 ┆ science │ │ ┆ PO-BY-SEPTEMB ┆ ┆ ┆ ┆ ┆ │ │ ┆ ER-3… ┆ ┆ ┆ ┆ ┆ │ │ 2026-05-02 ┆ WILL-TESLA-BE ┆ 0.0035 ┆ 0.4965 ┆ 1 ┆ 0.0 ┆ technology │ │ ┆ -THE-LARGEST- ┆ ┆ ┆ ┆ ┆ │ │ ┆ COMP… ┆ ┆ ┆ ┆ ┆ │ └────────────┴───────────────┴────────┴────────────┴───────────────┴───────────────┴───────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (15, 7)
| timestamp | symbol | close | conviction | high_confidence | intraday_range | category |
|---|---|---|---|---|---|---|
| date | str | f64 | f64 | i8 | f64 | str |
| 2026-05-02 | "BITCOIN-ABOVE-70K-ON-MAY-2:YES" | 0.9995 | 0.4995 | 1 | 0.0 | "crypto" |
| 2026-05-02 | "BITCOIN-ABOVE-72K-ON-MAY-2:YES" | 0.9985 | 0.4985 | 1 | 0.0 | "crypto" |
| 2026-05-01 | "BITCOIN-ABOVE-80K-ON-MAY-2:YES" | 0.0275 | 0.4725 | 1 | 0.0 | "crypto" |
| 2026-05-02 | "BITCOIN-ABOVE-80K-ON-MAY-2:YES" | 0.0445 | 0.4555 | 1 | 0.033 | "crypto" |
| 2026-05-02 | "BITCOIN-ABOVE-84K-ON-MAY-2:YES" | 0.0015 | 0.4985 | 1 | 0.001 | "crypto" |
| … | … | … | … | … | … | … |
| 2026-05-02 | "WILL-JUDY-SHELTON-BE-CONFIRMED… | 0.0035 | 0.4965 | 1 | 0.0 | "monetary_policy" |
| 2026-05-02 | "WILL-OPENAI-HAVE-THE-BEST-AI-M… | 0.034 | 0.466 | 1 | 0.001 | "technology" |
| 2026-05-02 | "WILL-SPACEX-IPO-BY-MAY-31-2026… | 0.0155 | 0.4845 | 1 | 0.0 | "science" |
| 2026-05-02 | "WILL-SPACEX-IPO-BY-SEPTEMBER-3… | 0.91 | 0.41 | 1 | 0.0 | "science" |
| 2026-05-02 | "WILL-TESLA-BE-THE-LARGEST-COMP… | 0.0035 | 0.4965 | 1 | 0.0 | "technology" |
In [17]python · cell 37
python
fig = make_subplots(
rows=1,
cols=2,
subplot_titles=("Conviction by Category", "Intraday Range"),
horizontal_spacing=0.15,
)
# Iterate over the categories actually present in the snapshot (ordered by
# market count) so every downloaded bucket appears - hard-coding a fixed
# list silently drops categories like technology and science that the
# regex/loader inference surfaces.
present_categories = (
features.group_by("category").len().sort("len", descending=True)["category"].to_list()
)
for cat in present_categories:
subset = features.filter(pl.col("category") == cat)
if subset.is_empty():
continue
color = category_colors.get(cat, COLORS["neutral"])
fig.add_trace(
go.Box(
y=subset["conviction"].to_list(),
name=cat.replace("_", " ").title(),
marker_color=color,
),
row=1,
col=1,
)
fig.add_trace(
go.Box(
y=subset["intraday_range"].to_list(),
name=cat.replace("_", " ").title(),
marker_color=color,
showlegend=False,
),
row=1,
col=2,
)
fig.update_yaxes(title_text="Conviction (|p - 0.5|)", row=1, col=1)
fig.update_yaxes(title_text="Intraday Range", tickformat=".1%", row=1, col=2)
fig.update_layout(
title=dict(text="ML Feature Distributions by Market Category", y=0.98),
height=550,
legend=dict(orientation="h", yanchor="top", y=-0.2, xanchor="center", x=0.5),
margin=dict(t=100, b=130, l=60, r=40),
)
for ann in fig.layout.annotations:
ann.update(y=1.05)
fig.show()Output
Across this snapshot every category sits close to full conviction: prices
cluster near 0 or 1, so |p - 0.5| stays near 0.5 for essentially all
contracts and the high-confidence flag is set on every row. The categories
separate on intraday range instead - crypto contracts move the most within
a bar, while monetary-policy, technology, and science contracts are nearly
static. That intraday dispersion is where ML-driven signals are most likely
to live.
13. Data Quality Assessment
In [18]python · cell 40
python
quality = (
features.group_by("category")
.agg(
pl.len().alias("n_markets"),
pl.col("volume").mean().alias("avg_volume"),
pl.col("intraday_range").mean().alias("avg_intraday_range"),
pl.col("conviction").mean().alias("avg_conviction"),
pl.col("high_confidence").mean().alias("pct_high_confidence"),
)
.sort("avg_volume", descending=True)
)
qualityOutput
shape: (4, 6) ┌─────────────────┬───────────┬────────────┬───────────────────┬────────────────┬──────────────────┐ │ category ┆ n_markets ┆ avg_volume ┆ avg_intraday_rang ┆ avg_conviction ┆ pct_high_confide │ │ --- ┆ --- ┆ --- ┆ e ┆ --- ┆ nce │ │ str ┆ u32 ┆ f64 ┆ --- ┆ f64 ┆ --- │ │ ┆ ┆ ┆ f64 ┆ ┆ f64 │ ╞═════════════════╪═══════════╪════════════╪═══════════════════╪════════════════╪══════════════════╡ │ technology ┆ 3 ┆ 62.0 ┆ 0.000333 ┆ 0.487 ┆ 1.0 │ │ science ┆ 2 ┆ 61.5 ┆ 0.0 ┆ 0.44725 ┆ 1.0 │ │ crypto ┆ 7 ┆ 44.142857 ┆ 0.005286 ┆ 0.483786 ┆ 1.0 │ │ monetary_policy ┆ 10 ┆ 30.9 ┆ 0.0006 ┆ 0.4837 ┆ 1.0 │ └─────────────────┴───────────┴────────────┴───────────────────┴────────────────┴──────────────────┘
.dataframe > thead > tr,
.dataframe > tbody > tr {
text-align: right;
white-space: pre-wrap;
}
shape: (4, 6)
| category | n_markets | avg_volume | avg_intraday_range | avg_conviction | pct_high_confidence |
|---|---|---|---|---|---|
| str | u32 | f64 | f64 | f64 | f64 |
| "technology" | 3 | 62.0 | 0.000333 | 0.487 | 1.0 |
| "science" | 2 | 61.5 | 0.0 | 0.44725 | 1.0 |
| "crypto" | 7 | 44.142857 | 0.005286 | 0.483786 | 1.0 |
| "monetary_policy" | 10 | 30.9 | 0.0006 | 0.4837 | 1.0 |
14. Save Enriched Data
In [19]python · cell 42
python
output_dir = get_output_dir(4, "polymarket")
output_dir.mkdir(parents=True, exist_ok=True)
output_file = output_dir / "polymarket_features.parquet"
features.write_parquet(output_file)
print(f"Saved {len(features)} Polymarket feature rows to {output_file}")Output
Saved 22 Polymarket feature rows to /tmp/ml4t-test-output/ch04_polymarket/polymarket_features.parquet
Key Takeaways
-
Centralized data path: This notebook now uses the shared
data/prediction_markets/download.pypipeline by default, matching the broader book workflow -
Downloaded OHLCV is enough for modeling: Probability level, conviction, and intraday range are available without hitting the live API during notebook runs
-
Optional live demo remains available: Set
LIVE=Truefor an interactive market-discovery demo when current Polymarket snapshots are the point -
Complementary platforms: Kalshi provides richer threshold structure, while Polymarket adds a broader crypto-native event universe. Using both gives a more complete view
Previous: See 12_kalshi_prediction_markets for OHLCV time series
analysis using pre-downloaded Kalshi data.