Every test is only as reliable as the data it runs on. Errors in data rarely announce themselves; they quietly produce results that look better than reality.

Three biases to remember

  • Look-ahead bias — using information before it was available.
  • Survivorship bias — testing only on assets that still exist.
  • Selection bias — choosing a period or universe because it makes results look good.

Adjusted and unadjusted prices

Splits and dividends change price series. Adjusted prices make returns continuous and are right for return calculations; unadjusted prices show what actually traded. Keep both and know which you are using.

Timestamps

Ask when each data point became known, not just what date it refers to. A company's annual results describe the year to December but may be published in March. Using them from January creates look-ahead bias. Data stored this way is called point-in-time.

Read more in our explainer Market Data 101.

Check your understanding

What is survivorship bias?

Testing only on assets that exist today, which ignores failed or delisted ones and flatters results.

Why does publication date matter for fundamental data?

Using figures before they were published introduces look-ahead bias.