Shipping Signals: How to Make Money on Prediction Markets with AIS Data

How to replicate this yourself

This is a step-by-step walkthrough. By the end you will have shipping metrics and market history flowing into your own code. No special access required - just an API key.

Step 1: Get a Veblyn API key

Create a free account at veblyn.com/auth. Your API key starts with vb_ and lives in the dashboard under Settings.

Step 2: Pull shipping markets

One call returns every active shipping market with its current price, volume, and metadata. Filter by category to get only shipping.

import httpx

client = httpx.Client(headers={"Authorization": "Bearer vb_YOUR_KEY"})

# Get all shipping markets
resp = client.get("https://api.veblyn.com/api/v1/market",
    params={"category": "shipping", "status": "active"})
markets = resp.json()["market"]

for m in markets:
    print(f"{m['title']}: {m['price']:.1%}")

Step 3: Pull the shipping metric you care about

Shipping metrics live under the PortWatch stream. List the available metrics, then fetch history for the one you want to study.

# List PortWatch metrics and inspect Suez throughput history
metrics = client.get(
    "https://api.veblyn.com/api/v1/stream/portwatch/metric"
).json()

suez_throughput = next(
    m for m in metrics
    if m["entity_id"] == "suez" and m["metric_name"] == "throughput_count"
)

history = client.get(
    f"https://api.veblyn.com/api/v1/metric/{suez_throughput['id']}/history",
    params={"limit": 30},
).json()

for point in history[-5:]:
    print(point["ts"], point["value"])

Step 4: Compare metric trend to market price

Pull market quote history separately, then compare it against the metric series. When throughput drops but price has not moved, you are looking at a mispricing.

market_id = markets[0]["id"]
quote_history = client.get(
    f"https://api.veblyn.com/api/v1/market/{market_id}/history",
    params={"limit": 30},
).json()

# Compare quote_history to the metric history above.
# If throughput_count is dropping while price hasn't moved,
# the market may be underpricing the disruption.

You do not need a model to act on this. A daily check of metric direction vs. market price is enough to catch most mispricings. The model helps you scale it - but the raw data is the real product.

The model we built

We trained a HistGradientBoostingClassifier from scikit-learn on historical shipping events to see how far signal data alone can go.

Calibration uses temperature scaling (Platt scaling) with temp=3.0. This smooths the model's raw probabilities so they better represent true likelihoods rather than overconfident point estimates.

Understanding the Brier score

The Brier score measures calibration - how well predicted probabilities match actual outcomes. Lower is better. A perfect forecaster scores 0.0. A coin flip scores 0.25.

The 6 chokepoints

Global shipping funnels through a handful of narrow passages. When any one of these gets disrupted, prediction markets react - but usually late. Here is what each one controls.

Veblyn tracks all six. Each chokepoint gets its own set of five transport signals, updated continuously from AIS data and maritime advisories.

What to watch for

You do not need to build a model to make money on shipping markets. Here are the practical rules we use internally.