452 anomalies.
One honest re-test.
Hou, Xue and Zhang rebuilt 452 published factors on clean, value-weighted data, then asked how many actually survive a real statistical bar.
|
How many survive |
| Published anomalies tested | 452 |
| Clear the basic bar (t over 1.96) | 158 |
| Clear the honest bar (t over 2.78) | 81 |
| Have a real economic reason to pay | about 6 |
Test 100 useless signals at the usual 5 pct bar and about 5 clear it by pure luck. Fish hundreds of factors out of the same stretch of history and most of the winners are exactly that. Harvey, Liu and Zhu argued the old t over 2 bar is far too low for a field this data mined, and pushed it to t over 3.