PI benchmarking

Retail PI statistics published by wholesalers are misleading. Price improvement vs NBBO does not take into account odd lot liquidity and for small retail orders it could be very substantial. To illustrate the scale of it, we computed virtual price improvement of US market to itself, i.e. we take all trades on lit markets and see how much better is the price on those trades from taker’s perspective vs NBBO. For 2020 we end up with 1.55 bil of totally fake price improvement against what was displayed as the NBBO. Notably the stock with the biggest “PI to itself” is TSLA ( high price high vol, small top of book size), which also tends to have heavy retail presence. 

So, a more accurate way of benchmarking PI would be to take into account all liquidity on lit markets, including odd lots. Is this enough, would this be true price improvement? No, not really. A lot of midpoint liquidity is available in dark pools, and unfiltered retail flow would be welcome in each one of these pools. True benchmark for wholesale price improvement would have been a good quality institutional agency SOR which would ping all relevant dark pools of liquidity, and then hit the lit market.

It is highly unlikely that after this proper benchmarking is introduced the remaining “true” PI is even positive.

And contrary to the claim of wholesalers, building such an SOR would not require “billions of dollars of investment into technology”. This may be true if the goal is to monetize retail flow to pay for PI, PFOF and leave billions of dollars for the wholesaler, however the bar for the retail broker to hit current execution quality is much lower.