\( \DeclareMathOperator*{\argmin}{\arg\!\min} \)
Everyone Overfits
Consider the work cycle of an equities quant:
1. Posit some mechanism causing market inefficiency (or read about it in a paper,
on the net, or in the NY Times Sunday Business section).
2. Collect the data and write some code to test for the inefficiency.
3. Find inconclusive backtest, and bugs in code.
4. Modify idea while fixing code.
5. Generalize the problem, start fiddling with knobs.
6. Maybe find an effect, but try some other stuff.
...
192. Brute force check all knob settings.
193. Definitely have a tradeable strategy this time, present to investment team.
194. Suffer from amnesia regarding development.
195. Back to 1.
How can this possibly work?