CRSP error-detection relied on the independence of monthly price changes.
Evidence
Rather than coding and punching all prices twice and then resolving discrepancies manually, we found a better procedure. We know that the change in the price of a stock during one month is very nearly independent of its change during the next month. Therefore, if a price changes a large amount from one date to a second date, and by a similar amount in the opposite direction from the second date to a third, there is a reason to believe that at the second date the price was misrecorded. A “large change” was rather arbitrarily taken to mean a change in magnitude of more than 10 per cent of the previous price plus a dollar. — Donald MacKenzie, An Engine, Not a Camera, 113
Categorization
- QUESTION: How does the CRSP cleaning algorithm, which assumes monthly price-change independence, interact with serial-correlation strategies that violated that assumption? This is a data-shaping gatekeeping fact worth chasing.
- DEFINE: Shows concretely how even “data cleaning” encodes a theoretical assumption about price behavior.
- XREF: connects to: The randomness of price changes was a foundational open problem in finance — Donald MacKenzie, An Engine, Not a Camera: How Financial…, 463
Broader Connections
- The randomness of price changes was a foundational open problem in finance
#qualifies(0.80) — Donald MacKenzie, An Engine, Not a Camera, 463