Compare Bitcoin and oil without fooling yourself

A pair of rising charts is a starting question, not evidence that Bitcoin tracks oil. To investigate a relationship, specify the data, align observations, choose a calculation, and show what changes when you alter the assumptions. This article describes a method; it does not publish a current correlation coefficient or a profitable strategy.

Write the question before downloading data

A testable question is: did daily BTC/USD percentage returns and daily returns of a specified WTI series move together over a stated interval? That is narrower than asking whether Bitcoin is digital oil. Save the series identifiers, source links, date range, timezone, and download date so another person can reproduce the work.

Choose an oil series deliberately. The EIA historical WTI spot series is one possible research input. A futures series is a different input, and joining consecutive contracts can introduce roll effects. Do not silently splice a spot series and a futures series because their labels both say WTI.

Compare returns, not just price levels

For a simple percentage return, divide the price change by the earlier price: return = (new price / old price) − 1. If an illustrative price moves from 80 to 82, the return is 2.5%. From 82 back to 80, it is about −2.44%, because the denominator changed. This asymmetry matters when checking a spreadsheet.

Shared trends in price levels can create a misleading impression of connection. Returns answer a different question: whether changes tend to move together. If your data includes zero or negative oil prices, ordinary percentage or logarithmic returns require special care. Document exclusions or use a justified alternative; do not let a spreadsheet hide invalid values.

Align time honestly

Bitcoin trades on weekends while an oil series may have observations only on business days. Repeating Friday's oil value across the weekend creates artificial zero returns. Instead, select matched observation dates and define consistent endpoints for both return intervals. Even same-date daily values can refer to different observation times.

• Report the sample size, missing observations, and time-alignment rule.
• Show the chosen window, such as 30 or 90 matched observations, rather than calling it simply recent.
• Try more than one sensible window and retain results that disagree with your original idea.

Interpret a coefficient carefully

Pearson correlation ranges from −1 to +1 and summarizes linear co-movement in the chosen sample. A value near zero does not rule out every relationship. A high value does not explain the cause, show which asset moves first, or establish that a strategy would survive transaction costs. Rolling windows overlap, so adjacent estimates are not independent new discoveries.

Use the dashboard to observe both markets, and the weekly market view to record questions. Neither substitutes for the aligned dataset described here.