Preventing Information Leakage

Discussion by Katya Malinova

Paper by Jonathan Brogaard, Dan Li,  Matthew Ma, and Ryan Riordan

Main Findings

Trades split across brokers:

  • larger & more profitable

Compared to similar trades:

  • smaller price impacts
  • larger profitability

\(\Rightarrow\) Split trades prevent info leakage(?)

How it fits into the literature?

information leakage by brokers 

  • McNally, Shkilko, Smith (MS 2016): brokers leak presence of insider trades
  • DiMaggio, Franzoni, Kermani, Sommavilla (WP 2018): broker networks

Delegated venue routing decisions

Spliting orders across time

  • Menkfeld & van Kervel (JF 2018): interaction of HFTs and large orders
  • Korajczyk & Murphy (RFS 2019): HFT liquidity provision to large orders
  • Malinova & Park (WP 2017): multi- vs single-market routing
  • Battalio, Hatch Saglam (WP 2019): routing of orders to HFTs
  • Anand, Samadi, Sokobin, Venkateraman (WP 2019): brokers routing to their own platforms

This paper: 

  • no delegation (=DMA)
  • splitting across brokers
  • leakage by brokers

Q1: why DMA clients and who are they?

Pro: User ID info \(\Rightarrow\) can identify all trades

But: who are these firms?

  • What info do they trade on?
  • What are the trading/info horizons?
  • E.g., HFT vs. buy-side institutions?
  • What about non-DMA buy-side institutions?
  1. How prominent are DMA arrangements among buy-side?
  2. Can you condition on the DMA firm type?

Q2: Mechanics of leakage?

  • Authors: "exploiting the information passed by the broker".
  • Background idea: contact the broker, they know who you are & tip off other clients 
    • McNally, Shkilko, Smith (2015): trades of insiders
  • Here: DMA clients.
  • How do the brokers detect & broadcast these?
    • Run an algo? (Commit fraud and document it???)

Q2: Leakage, detection, or spurious correlation?

Alternate explanations:

  1. Broker/HFT fixed effects
  2. No "leakage" but split trades to avoid detection
  • Many HFTs come through CIBC and BofA
  • Extreme scenario: if split trades exclude CIBC
    • \(\Rightarrow\) different firm types use split vs. non-split
    • \(\Rightarrow\) more likely to have "followers" with non-split (CIBC)

Q2: Leakage, detection, or spurious correlation?

  • Canada: public broker attribution 
  • \(\Rightarrow\) Harder to detect large trades if split across brokers 
  • Authors: "trade followers" from the same broker
  • Suggestion: check "followers" from other brokers
    • Caution: broker/HFT fixed effect

Alternate explanations:

  1. Broker/HFT fixed effects
  2. No "leakage" but split trades to avoid detection

Q3: Causality & Decision to Split?

Why don't they always split across brokers?

Split vs. non-split for different strategies?

  • E.g., split long-run "buy and hold", don't for arbitrage
  • Can "negative" profits be one-leg of a multi-leg strategy?
    • E.g., arbitrage for cross-listed stocks or trades across multiple asset classes?

Do firms learn to split over time?

  • Is the effect persistent through the sample?

Q3: Causality & Decision to Split?

Summary stats?

Summary stats?

  • More controls/robustness checks? 
    • Active vs passive
    • Cross-listed
    • Broker fixed effects and/or fraction of HFT DMAs 

Other Comments

  • Followers: trades over one to two weeks
    • With large trades identified daily, who follows who? 
  • Do the followers "follow" or "behave as always"?
  • Duration of split vs. non-split trades?
    • Large trades: daily
    • Do matched split and non-split trades last the same time?

@katyamalinova

malinovk@mcmaster.ca

slides.com/kmalinova

https://sites.google.com/site/katyamalinova/