Katya Malinova PRO
I am an Associate Professor, Mackenzie Investments Chair in Evidence-Based Investment Management at the DeGroote School of Business, McMaster University, Canada.
Trading Across Fragmented Markets
Workshop on Frontier Areas in Financial Analytics
May 2019
The Fields Institute
Katya Malinova and Andreas Park
Developments in Equity Trading over the last Decade
-
computerized (equity) trading
- no more phones and human specialists
- electronic access through brokers
- electronic traders: autonomous and fast (high-frequency traders)
- multiple marketplaces
-
regulation that mandates electronic linkage of multiple markets
- U.S. and Canada: protected quotes (or no-trade through rule)
- = market-wide price priority
- U.S. and Canada: protected quotes (or no-trade through rule)
Popular Line of Reasoning
-
traders report that after they submit orders, all hell breaks loose:
-
quotes "fade"/"slide" on other venues
-
"others" get to trade on other venues before them
-
-
=> HFTs and fragmented markets are at fault
What do HFTs do after trades?
- What explains HFTs' reactions (if present)?
- Is there an impact?
- What is the role of fragmentation?
Trading with multiple markets
1,000
Shares at Canadian Offer
300
400
Regular Joe sends buy order to broker
buy 1,500 shares
no trade through => broker must split among three venues
Shares
1-tick off
400
100
2,000
Through the microscope
1,000:
Shares at Canadian Offer
300:
400:
100
100
100
100
100
100
100
- They may not want to trade all that they post.
- will try to cancel quickly
- once their order gets hit or
- once they see trades
- => flurry of cancellations
- will try to cancel quickly
Flurry of HFT Activities after Trades
50% of trades -- by a non-HFT -- are quickly followed by a cancellation -- by an HFT -- on a different venue within 5ms of the trade
Within 50ms: about 70% of trades followed by cancels
And: the reaction is more extreme
after multi-market trades!
Flurry of HFT Activities after Trades
13% of trades -- by a non-HFT -- are followed by an aggressive trade -- by an HFT -- on another venue in the direction of the original trade within 5ms
Within 50ms, 29% of trades are followed
Again: the reaction is more extreme for multi-market trades!
This project's goal:
Characterize this behavior and its impact in fragmented markets
-
Step 1: Characterize/describe fast (HFT) traders’ reaction to trades:
- Do they cancel their orders? (yes)
- Do they submit own aggressive orders? (yes)
- Different reaction to single vs multi-market orders? (yes)
-
Step 2: What explains the different reaction?
-
size?
-
type of trader?
-
information?
-
- Step 3: Does the HFT behavior have an impact on the market?
Not the first to look at fragmented markets
- Long literature, including
- Joel Hasbrouck (e.g., "One Security, Many Markets: Determining the Contributions to Price Discovery", JF 1995)
- O'Hara & Ye (JFE 2011): good for mkt quality
- Degryse, de Jong, van Kervel (2014): visible vs. dark fragmentation
- Bernales, Riarte, Sagade, Valenzuela, Westhei (2017): fragmentation without competition
- most closely related:
- van Kervel (RFS 2015): over-posting exists
- Baldauf & Mollner (WP 2015) (theory): splitting of liquidity across markets
- Brogaard, Riordan, Hendershott (JF 2018): HFT generate price discovery with limit orders, without trading.
Critical Ingredients
-
Our focus:
-
HFTs in regulation-mandated integrated mkts
-
- What do we need?
- examine HFT? => need trader level info
- impact of mandated multi-mkt? => need trader level data to identify multi-mkt vs single mkt
- examine the impact => need identification
-
proprietary masked trader-level data for all Canadian equity markets (provided by IIROC, the Canadian regulator)
-
use 30 most frequently traded non-crosslisted stocks, March - May, 2013.
-
-
a critical market-organization change that eliminated latency between two of the three main markets (markets A and B) on April 29, 2013 => identification
Data on Investors vs Traders
Identifying Traders
-
Similar to Comerton-Forde, Malinova, Park (2018)
-
Fast traders: Use three criteria (across many securities on many days: 307 securities, Jan& Feb, 2013)
-
regularly submit and cancel orders very quickly
-
median submit-to-cancel times < 250ms.
-
-
submit/cancel most orders very quickly subsequent to someone else’s activity
-
85% of activity within 1ms of someone else.
-
-
react quickly (500ms) to a particular, regular, market-wide news announcement (the market-on-close imbalance).
-
classified: ~82 (out of ~4,900)
-
- Retail: special order type that can only be used by retail
-
Institutions: trade-strings:
- at least 10 distinct orders
- single direction on a day
What is a multi-market trade?
-
same trader ID
- caveat: "trader ID" is not the same as "trader"
- submits marketable order on separate markets within 5 milliseconds
Post-trade Cancellation
- cancellation by fast trader
- within 1,2,...,5 milliseconds
Post-trade Aggressive Order
- in the same direction as the trade
- aggressive by fast trader
- within 1,2,...,5 milliseconds
What should we expect?
- HFT are often voluntary market makers (MMs)
- MMs don’t want to absorb large inventories because of
- capital commitments;
- risk of adverse price movements.
- MM should respond to trades.
-
post on multiple venues => cancel to avoid overtrading
-
learn new info => cancel/reprice existing quotes + "backrunning"
-
accumulate inventory => revert (=trade aggressive with trade)
-
- There is still a question if the reaction
- warranted or
- an over-reaction
- why different for multi-market
-
Trades = information.
- Baldauf and Mollner (2015): only smart trade everywhere
- van Kervel (2015): only sophisticated have access to smart order routers/multiple venues
-
Market makers post everywhere but only want to trade once.
- Cancel existing orders (van Kervel (RFS 2015))
Are multi-market trades different?
- Market making should lead to flurry of activities
- cancellations are to be expected
- aggressive orders can be part of market-making
-
Are the HFT actions merely seamlessly integrating markets?
-
First steps: are there any "in plain sight" differences between single- and multi-market orders?
- Are they submitted by more informed traders?
- Are they larger?
-
First steps: are there any "in plain sight" differences between single- and multi-market orders?
Measure of Interest: Price Impact
- signed difference of the prevailing mid-point of the NBBO bid-ask spread τ time units in the future and the mid-point that prevailed at the time of the trade
- qit = 1 for buyer-initiated and -1 for sells
- mit = midpoint at time t
Are multi-market trades more informed?
Simple summary stat: price impact
Naive conclusion: multi-market trades have higher price impact, therefore they are more informed
Consistent with Baldauf & Mollner and van Kervel:
multi-market = smarter
HFTs must react stronger!
End of story?
Who trades in multiple markets?
All trades
Multi-market trades
Multi-market trades:
≈ 32% of total $-value
Price impact for retail trades
Conclusion: If we believe that retail orders are not informed, then the price impacts for single vs. multi-market orders shouldn't look this different.
Reminder:
- Broker SORs may have to split larger orders to obey OPR
- Brokers may want to split larger orders to avoid high costs
Multi-Market
Single Market
vs.
Is it Size?
| trade size | % of all value | % of multimarket value |
|---|---|---|
| 100-200 | 23 | 3 |
| 201-500 | 17 | 11 |
| 501-1,000 | 15 | 15 |
| 1,001 -- 5,000 | 25 | 36 |
| >5,000 | 19 | 36 |
$ value of trades per bucket
$ value of all trades
$ value of multi-mkt trades per bucket
$ value of all multi-mkt trades
Size distribution
Conclusion: multi-mkt orders are larger
Is it size?
Single Market
Multi-Market
Larger size => larger price impact?
Plotting: price impact multi-mkt minus single-mkt
Conclusion: even for similar size, price impact of multi-market orders is larger.
Is it size?
Could price impact be larger because of the HFT reaction?
Plotting: price impact with HFT reaction minus price impact without HFT reaction
reaction = cancellations
Reaction= trades
Observation: HFT makes your trade look fat
fast aggressive orders(same direction minus opposite direction)
total number of transactions
Conclusion: HFT reaction looks like there is much more activity than warranted by the original trade
Multi-Market
Single Market
Bottom Line
- multi-mkt trades are larger but need not be smarter
- using multi-mkt is/can be regulation requirement
- possibly exchange fee considerations
- retail (brokers) use them regularly
- multi-mkt have larger price impact
- even for retail
- for same size
- multi-mkt with HFT cancellations/aggressive submissions
- have larger price impacts.
- look "bigger"
"Trade clusters"
- "Merge" all transactions that occur within a short (10ms) time interval into a "trade cluster", instead of restricting these to be by a single trader
- Average cluster lasts for 3.4ms
- Compare trade clusters :
- with single vs. multiple traders (i.e., where the original trade was followed by others)
- that occur on a single vs. multiple markets
Price Impact of Trade clusters
- Assign all trade clusters into one of the four categories:
| Single Trader | Multi Trader | |
| Single Mkt | ||
| Multi Mkt |
- Estimate a regression over all trades to assess whether following by other other traders & multi-market trading matters for the price impact, controlling for the size
Price Impact in Trade Clusters
| Dummy | Coeff controlling for the vol of the first trade | Coeff controlling for the aggregate cluster volume |
|---|---|---|
| MultiMkt-SingleTrader | 0.645*** | 0.605*** |
| SingleMkt-MultiTrader | 1.919*** | 1.132*** |
| MultiMkt-MultiTrader | 2.095*** | 1.137*** |
- Effects on the price impact 5ms after the trade (in basis points), controlling for the size of the first trade in a cluster or for the aggregate volume of the trade cluster (2 separate regressions):
Who follows the trades and why?
- Over 60% of follow-up trades stem from 14 fast traders
- As a group, these:
- Earn $1.25M on in-cluster trades
- Lose $0.28M on out-of-cluster (active) trades
- Compute return per trade, relative to the trade's VWAP and assuming that positions are closed at end-day prices:
- Estimate:
Who follows the trades and why?
Bottom Line
- Price Impacts are higher for
- Trades that execute on multiple markets -- even after controlling for trade size
- Possible explanations: liquidity effects: the choice of whether to go to multiple markets is endogenous
- Trades that are followed by other traders -- again, after controlling for trade size
- This latter effect is even more pronounced for trades that are followed by one of the 14 fast traders
- Trades that execute on multiple markets -- even after controlling for trade size
- Remains unresolved whether followers:
- increase the price impact by making trades look "bigger"
- or follow the trades that are informed ("back running")
The Big Question
-
Is the reaction:
- reacting to information/price discovery
- HFT push prices to the "right" level
- or noise
- HFT reaction obfuscates learning
- reacting to information/price discovery
How to identify?
- Idea: if latency between venues disappears:
- non-HFT order flow should remain similar
- if price discovery => HFT can create same level of it.
- "Following" is harder to implement:
- if noise => lower price impacts
- non-HFT order flow should remain similar
How do you make physical latency disappear?
Market A and B move to the same data centre: April 29, 2013
What would we expect?
- if you post on both A and B, you cannot avoid being "hit" on both, i.e. no more outrunning
- depth should decline
- spreads may increase
- fewer post-trade cancellations
- impact on post-trade aggressive trading is unclear
- Notable differences between A and B:
- Market A: lowest take fee/make rebate
- Market B: "big", highest take fee/make rebate
Empirical Strategy
Examine changes in market quality and trader behavior
- before/after (March 1-May 31) and also by market;
- DV dependent variable of interest
- Dummies for the event and the market
- Controls include VIX for volatility & an ETF for the S&P500 GSCI commodity index for the price level
- Security and mkt fixed effects
- Std errors double-clustered by date and security
Changes in Liquidity and fast trading activities
Changes in Responses to Trades
(no)
Untabulated: no significant changes in size or usage of multi-mkt trades
Changes in Price Impacts
Bottom line: price impacts of multi-market orders decline
Difference in differences of multi- vs. single-mkt orders before vs after
Summary and Conclusion
- Multi-market trades are
- common
- often required by regulation
- also performed by choice (and without need?)
- not the sole purview of sophisticated traders
- How do fast traders react to trades?
- Fast traders cancel quotes rapidly and take out (stale?) quotes after trades.
- Stronger reactions to multi-market trades
-
What does HFT behavior do?
- Some evidence that it increases price impact of orders
- indication that in multiple mkts, HFT may obfuscate learning/price discovery
- Some evidence that it increases price impact of orders
Trading Across Fragmented Markets Workshop on Frontier Areas in Financial Analytics May 2019 The Fields Institute Katya Malinova and Andreas Park
Trading Across Fragmented Markets
By Katya Malinova
Trading Across Fragmented Markets
This is a deck that I used for a presentation at the Fields Workshop in May 2019
- 1,047
