Research Opportunities in Order Book and Execution Data

Summary

The document describes the data available to an institutional equities execution desk: full-depth order book events, the broker’s own submitted quotes, links from child orders to parent orders, and client order characteristics such as average daily volume, time of day, spread, and volatility. The desk has already examined post-trade price reversion at short horizons, book activity around submissions, passive-order adverse selection, and how child-order impact changes over a parent order’s life.

It is primarily a request for research ideas, rather than a proposed method or a report of findings. Its practical lesson is that these linked records can support studies of execution quality, market response, and order-level effects in context. The document provides no suggested study designs, results, or controls for confounding factors, so it does not establish which analyses would improve an algorithm. Any conclusions would need careful separation of the effects of the algo’s decisions from market conditions and parent-order urgency.

Key ideas

• Linking order book events, submitted quotes, child orders, and parent orders enables analysis across execution levels.
• The desk has measured short-horizon post-trade price reversion as an indicator of passive-order adverse selection.
• Order book changes around submissions can be studied to characterize market response to the algorithm.
• Child-order impact may vary over the duration of a parent order.
• The document poses research questions but provides no proposed study designs or empirical results.

Tags

# What to do with L3 orderbook data, quote engine, and parent order information


# What to do with L3 orderbook data, quote engine, and parent order information












I work on a trading desk as a quant/trader at a broker dealer on the electronic/algo equities execution desk. Our clients are institutional (hedge funds, asset manager) that utilise our algos to trade their orders. They use mainly liquidity seeking, POV, and VWAP. We have devs building our own stack (proprietary algos, SOR, volume predictions), so everything is in-house.

I have extremely granular data:

- Level 3 orderbook across all markets.

- Our own quotes that our algo engines sent to the market. I matched L3 orderbook quotes with our engine's own quotes, so I know exactly what action was taken on the book, and if it ours or not. And it is linked back to a parent order (i.e. 100 child orders can be linked back to a single parent order)

- Parent order information from our clients (and information on the orders, like ADV, time of day, spread, volatility, etc...)

I spent a long time looking at things like post-trade reversion (mid-price change 1ms to 60s after) on a child level to see if we are getting adversely selected when posting passively on the book, understanding if there was more movement in the orderbook before and after sending orders out to the market, if child impact increases overtime in a parent order etc...

I realise this is an incredibly general question, but I'm in need of inspiration. What are some studies that would be interesting to run to improve our algos capabilities?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.