Why randomize? Execution algo and slippage

Question

Historical data show that orders routed through your new execution algo had lower slippage than orders that were not. But traders decided case by case whether to use the algo. Leadership wants to mandate the algo everywhere based on this gap.

Explain why the historical gap cannot establish that the algo reduces slippage, and design a test that can. Why is randomization essential?

Accepted Answer

Why the gap is not enough. Traders chose when to use the algo, so usage was not random. They likely reached for the algo on easy, liquid orders and traded the hard, illiquid ones by hand. Order difficulty is a confounder: easy orders have low slippage and were also the ones sent to the algo, so the algo looks good partly because it got the easy work. Skilled traders self-selecting into the algo would add another confound. Mandating it on the hard orders too could give very different results. A test that can prove causation. Randomize at the order level: for eligible orders, a coin flip (or a …

Why randomize? Execution algo and slippage

Your answer

More Statistics questions