Methodology

Every number on this site is computed by our code from public price history or quoted from a cited source. This page explains how.

Data

• Company filings. Earnings release dates from SEC EDGAR.
• Prices. Daily closes from Yahoo Finance for indices, stocks (adjusted for splits and dividends), crypto, futures and currencies; FRED (Federal Reserve Bank of St. Louis) for Treasury yields, policy rates and IMF monthly commodity prices. 74 markets in total.
• Completed sessions only. A bar dated today may be intraday, so all figures are as of the previous close (currently October 6, 2026).
• Futures. Commodity prices use continuous front-month futures, which jump when the contract rolls. Monthly returns therefore include some roll effects, which matters most for markets with steep curves such as natural gas.
• Monthly averages. European gas, aluminium, nickel and the German 10-year yield are monthly averages from the IMF and OECD, which smooths their moves compared with month-end prices.

Seasonality

A monthly return runs from the last close of one month to the last close of the next. For yields we use the change in percentage points and show it in basis points. For each calendar month we report the average, median, share of up years, best and worst year, the average of the last ten years and the number of years. Months with a gap in the data are skipped, never merged.

Pattern rating

We score each month from 0 to 5: one point if the average points the same way in both halves of the sample, one if the direction held in at least 60% of years and another at 67%, one if the t-statistic of the average is at least 1.5 and another at 2.2. With fewer than ten years, the score is capped at 2. Four or more is strong, three is moderate, two is weak, and below that there is no clear pattern. With 74 markets and twelve months, some patterns will look strong by chance, which is why we also require consistency across both halves of the history and explain a plausible cause before writing about one.

Event studies

Reactions start at the last close before the event day. Day 1 is the close on the event day, day 5 and day 20 are five and twenty sessions later. Fed decisions come from the target-rate series on FRED from February 1994, when the Fed started announcing decisions. ECB decisions come from the deposit facility rate and are dated to the Thursday before each change takes effect. Tariff headlines are a curated list of dated posts, speeches and orders; when a headline came after the US close, the reaction starts from that close. Oil supply shocks are a curated list of OPEC+ decisions and disruptions measured the same way. Earnings reactions use the dates of earnings releases filed with the SEC (Form 8-K, Item 2.02) for US companies and measure the move from the close before the filing day to the close of the next session, so both before-the-open and after-the-close releases are captured. Event averages are descriptive and small samples are labelled as such.

Calendar studies

The presidential cycle uses S&P 500 calendar-year price returns since 1928, with year one being the year after a presidential election. Sell in May compares November to April with May to October. The halving study measures bitcoin from each halving date.

The daily brief

1. Collect. Each weekday at 05:00 UTC an AI model with web search gathers news from the last 36 hours on our markets and topics, restricted to an allowlist of 39 official and 49 established sources.
2. Verify. Each fact must come from an allowlisted domain, and every number in it must appear on the page the model fetched. Facts that fail are dropped.
3. Compute. Our code refreshes prices and statistics. The model never calculates a statistic.
4. Draft. The model writes insights around the verified facts. Historical figures are inserted as references that our code fills in, so a model cannot change them.
5. Validate. A validator rejects any insight that contains a number not found in its cited facts, any unresolved reference, advice language or banned punctuation. Rejected drafts are retried once and otherwise dropped. When a brief is published, every computed figure in it is frozen at that day's values, so the text you read later is the text we published.
6. Publish and grade. The site is rebuilt and published. Scenarios that name a market and direction are graded on the scorecard when their horizon ends.

Corrections

If you find an error, email hello@marketseasonality.com. We correct the page, note the correction at the bottom of the insight and do not silently change published scenarios.

Not investment advice. Market Seasonality publishes historical statistics, news summaries and scenario analysis for information and education. Past patterns do not guarantee future results. Nothing here is a recommendation to buy, sell or hold any asset, and scenarios describe possible paths without predicting them. See the disclaimer and methodology.