Overview:

1. Liquidity provision over time of HFTs

2. Long-term liquidity provision in a network of trading agents

3. Flurries in trading activity and price changes

• The size of the bid-ask spread

• Transactional costs

• The passive and aggressive sides of a trade

• The size of the bid-ask spread

• Transactional costs

• The passive and aggressive sides of a trade
  • More directly related to the behaviour of the broker
  • Difficult to measure without persistent broker identification or an unfragmented market (which we have)

• All Trades during normal trading hours on the ASX over one year around 2013 (Here we examine a representative subset of 5 high market cap stocks)

##          Date         Time BuyerID SellerID  Price Volume   Qualifier
## 1: 2013-05-17 11:10:02.115       A        B $21.05    100 Passive Buy
## 2:        ...          ...     ...      ...    ...    ...         ...

• Identifiers for the buyer and seller of trades
• Trades are marked with the aggressive/passive agent

• Australian market is relatively unfragmented meaning that this dataset captures an almost complete picture of trading activity

• We follow 8 known proprietary HFTs (De-identified)

Broker A (the smallest of the HFTs) makes liquidity (green) for ~50 days, then takes liquidity (brown) until Day ~160 before exiting the market

Broker A's behaviour is similar throughout all stocks

Broker E exclusively takes liquidity for ~170 days before leaving and reentering the market with a slightly more liquidity-making strategy

Broker H (largest of the HFTs in dollar volume traded) consumes large amounts of liquidity (net 5.9% liquidity reduction in stock 1) across all of the high market cap stocks shown here

B and C are more liquidity-making than liquidity-taking, though the net increase is small.
D shows exclusive liquidity-taking strategies and some interesting periods of liquidity-making.
It is always net negative for liquidity, however.

Brokers E and H show strategies that reduce liquidity overall, while G is slightly net negative liquidity provision in these high market cap stocks.
F is blank- it is included here for completeness due to its activity in low market cap stock.

Examining a selection of low market cap stocks, we can see that Broker F while not active in high market cap stocks, is providing more liquidity than it is taking.
In contrast to the behaviour seen earlier, Broker G is exclusively taking liquidity from these low cap stocks in large amounts.

Considered as a single group and measured in dollar volume, HFTs appear to take liquidity about 70% of the time.
This is skewed heavily by larger HFTs who seem to have one-sided liquidity-taking behaviours.

    Network graphs and graph theory are well-suited to analysis

       concerning flow of information, influence, disease, and so on.

    Can it be used to find market makers who facilitate the flow of

       stock between other brokers?

Generated with data from stock 1 over 180 days
Each broker is rendered as a pie chart showing proportions of Aggressive and Passive Buys and Sells

Centralness of an agent's position in this network is ultimately determined more by the broker's size rather than who they traded with, but there are benefits of viewing a node amongst (and comparing it against) its peers.

Some liquidity makers (green)

Some liquidity makers (green) and takers (brown)

Compared against other brokers, the High Frequency Traders D, E, and H show imbalanced proportions of liquidity provision

• All Trades and Orders on the ASX for 5 stocks in the banking sector during interest rate announcement days (Orders are used to reconstruct millisecond-resolution market depth)
• We examine flurries of activity by attributing the timing of a trade to the agent on the aggressive side.
  Timing is then considered relative to informative events such as:
  - the previous and next trade
  - the previous and next change in price

• HFT provision of liquidity is heterogeneous
• Larger HFTs (by dollar volume) appear to have one-sided liquidity taking strategies
• HFTs, as a group, appear to consume more liquidity than they provide
• HFTs appear to initiate price changes more frequently per trade than non-HFTs
• Larger liquidity-taking HFTs appear to have a tendency to trade aggressively within 0 or 1 millisecond(s) of other trades and less frequently outside of this context

Visualizations were created using the packages:

        bit64
        data.table
        igraph
        ggplot2
        RColorBrewer

                                            Any Questions?