The Anatomy of Trading Algorithms

Discussion by Katya Malinova 

Tyler Beason & Sunil Wahal

\(\Rightarrow\) Many cool, new stylized facts 

  • Sizes, duration, number of child orders
  • Usage of market vs limit (& dark vs not) 

Amazing data!

  • Parent and child orders for 2012-2016
  • Large algo provider, 4 algos
  • 5,000+ securities, 961 institutions, $675B  

" open up the black box of trading algorithms"

Key Contributions

Representation? Distribution of algo volume

Algos are very, very different!

Parent order cancels:

32%

59%

Algos are very, very different!

Aggressive LO:

71%

38%

71%

65%

Any insights re: strategies by different trader types?

  • Proxies for traditional informed/uninformed?
  • Market vs. limit: who uses aggressive limit? 
  • The basic insight from theory: more informed \(\to\) more aggressive.
    • Any evidence?

Comment 1. Algo usage: who and/or when?

Different algos = different users/info vs. different timing?

  • Parent order = order to an algo (???).
    • Relatively short: 20-90 min.
  • Does the same institution use different algos over the day?
    • E.g., more aggressive early (while the info is new)?
    • Or later? Day-end-deadline?

Not all parent orders are created equal

  • Do characteristics of a parent order drive # of child orders, aggressiveness, visibility etc. ?
  • Any endogeneity concerns when analyzing the impact of these?
  • The algo-level analysis (really nice!) addresses a chunk of these? 
    • But only four algos ... rather coarse?

Comment 2. Parent Order Costs

As an aside:

  • How do you proxy for the costs of non-execution?
  • Why the impact of "visibility" (+ve vs -ve) is so algo-dependent?

Any insights to those of us using public data?

  • Better tools to compute institutional trading costs?
  • Insights on how to identify "parent orders" in public data?
  • Proxies for the costs using public data?

Comment 3. Prop to public data?

Further Comments: (No) Price Impact 

Filled

Unfilled

Many (even aggressive!) orders have no price impact!

  • at 100ms: 80%!
  • Malinova and Park (2020) document similar facts for trades.
  • What do the per-stock-per-time-interval averages tell us?
  • When do the orders do have the price impact?

Perspective that "cannot infer future order flow".

Further Comments/Questions: Order Anticipation 

Source: Doug Clark (ITG at the time) 2012 presentation at the Bank of Canada

Can these be anticipated? 2012 vs 2016?

@katyamalinova

malinovk@mcmaster.ca

slides.com/kmalinova

sites.google.com/site/katyamalinova/

FMA Discussion: "The anatomy of trading algorithms"

By Katya Malinova

FMA Discussion: "The anatomy of trading algorithms"

This deck is arranged as a 2x2 matrix, to be viewed column by column.

  • 743