How to Measure a CS2 Update's Market Impact

Use POST /v1/prices/history to compare a fixed basket across equal windows before and after the update timestamp. Keep the marketplace, price side, interval, and basket unchanged.

This design measures an association around the event. It does not prove that the update caused every observed price move.

Recent OHLC coverage begins on December 24, 2025. The endpoint is available on Scale and Enterprise plans.

Fix the study design before reading results#

Record:

• the update timestamp in UTC;
• a pre-event and post-event window of equal length;
• one marketplace source;
• one price side and candle field;
• one interval;
• the exact item basket and any weights.

Hourly close_ask is a practical signal for many short event studies. It represents observed listing floors, not completed trades.

Avoid changing the basket after seeing which items moved. That turns a market study into a winner selection exercise.

Build comparable item returns#

For each item, select the last valid observation before the event and the first valid observation at the chosen post-event horizon. Require both points and record their actual close_time values.

Then aggregate item returns with a declared method:

Do not let missing items silently change weights from hour to hour.

Report coverage beside impact#

An event result should include:

• requested and comparable item counts;
• requested and returned candle counts;
• median gap from nominal bucket time;
• first and last actual observations;
• items excluded for missing data;
• result with and without the thinnest items.

Preserve missing buckets. Forward-filling across the update can turn an old quote into a fake post-event observation.

Separate broad and item-specific effects#

Build a fixed market index for the broad move, then compare categories or affected items against it. The CS2 market index guide explains how to keep constituent coverage stable.

You can also examine volatility and visible order counts, but label each measure precisely. Listing count changes are not completed sales, and higher sample_count can reflect observation coverage rather than demand.

Interpret cautiously#

Other announcements, marketplace outages, currency moves, and ordinary market drift can overlap the event window. A control basket can reduce some noise, but it still does not establish causality.

Use the smallest predeclared window that captures the expected reaction, show wider windows as sensitivity checks, and preserve the complete methodology with the result.