Flavors of Index Arb

Flavors of Index Arb

Consider a few scenarios:

  1. Your model tells you that the DAX will go up (down). You hence buy (sell) futures and the constituents.
  2. You detect that the DAX future went up (down) and anticipate that the cash equities fill follow. You hence buy (sell) a basket of the constituents.
  3. You were quoting the DAX future, the price went up (down) so that one of your sell (buy) orders was filled. To hedge this exposure you buy (sell) a basket of the cash equities.
  4. You think that the future is rich (cheap) relative to the cash. You hence sell (buy) futures and buy (sell) a basket of equities.

These four scenarios is what the four quadrants in Fig. 1 below correspond to. The plot shows the number of trades for DAX40 constituents on Xetra in the same (upper half) and in the opposite (lower half) direction as a preceding trade for the DAX future (FDAX) on Eurex.

The horizontal axis represents time since an order for the DAX future (FDAX) was received by the exchange ("FDAX t3a"). The time labeled "FDAX t9d" is when the exchange sends out the resulting trade. The length of the interval between these varies from event to events. This is why times are shown in relative terms in the left half of the plot.

Any events in the left half cannot have been reactions to the FDAX trade. Either they had a common external trigger or the same participant targeted both markets simultaneously.

Figure 1: Time between Eurex receiving a request that leads to a trade in the DAX future to Xetra receiving a request that leads to a trade in a DAX40 constituent.

Starting from a trade in the FDAX, we start counting trades in the DAX40 constituents and note their timestamp and side relative to the trigger. The dark blue are is the temporal distribution of the first DAX40 trade. The lighter shades of blue represent successive trades.

The upper left quadrant represents scenario #1 (trades in the same direction, DAX40 cannot be a reaction), the upper right quadrant is scenario #2 (DAX40 trade is a reaction to the FDAX trade), the lower right quadrant is scenario #3 (hedging trade as a reaction), and the lower left quadrant, finally, represents scenario #4 (the classical basis/index arb trade).

There are a few subtleties to the plot:

(a) The thickness of the vertical grey line is 2720 ns. If participants reacted instantaneously to seeing a trade in the future then we would see a reaction on the right edge of the thin grey line. And for trigger/reactions within Eurex one does observe reactions very close to this theoretical limit. But here reactions come with a significant delay. This has to do with the physical setup of the venues in the datacenter.

(b) Related to this: the reactions for trades in the opposite directions are delayed more than trades in the same direction. One hypothesis could be that participants only send out hedging orders upon receiving a fill. The private fill, however, comes after the trade in the public data (by roughly 10 μs). In theory, participants do not have to wait on the private fill. The public trade (ExecutionSummary message) includes enough information for the participant to unambiguously infer his fill.

#eurex #xetra #dax #dax40 #equitytrading #futurestrading #quantitativeresearch #lowlatency #arbitrage

a) From t3a to t9a, the count of trades in both sides gradually increase with timeflow, is it also consistent with your previous article, saying that there're traders who anticipate the future price's move and send orders just before the market data came out? b) The private fill comes after the trade in the public data by roughly 10 μs, is it intended or...? And actually for what? To my knowledge, in some crypto markets, the private fill comes before the public data, so that makers could get the latest price faster than those who only subscribe market data. This makes sense, since the makers provide liquidity and take the risk of adverse selection.

回复

要查看或添加评论,请登录

Stefan Schlamp的更多文章

  • Where'd Your Money Go?

    Where'd Your Money Go?

    Under ideal conditions (ideal from the market maker's point of view), the market maker earns the half-spread on every…

    8 条评论
  • Share Buyback Regs vs. Physics

    Share Buyback Regs vs. Physics

    The regulations for executing share buybacks are oddly specific. For example 2016/1052: Buy orders shall be entered or…

    3 条评论
  • Retail, Dark Midpoint, and CLOB Intraday Volume Profile

    Retail, Dark Midpoint, and CLOB Intraday Volume Profile

    Trades facilitated by retail liquidity providers and trades on the dark midpoint book can be distinguished from…

    2 条评论
  • Variability of Auction Liquidity

    Variability of Auction Liquidity

    Suppose you estimate your price impact on the auction uncrossing price based on some historical average. How far off…

  • Manz AG (M5Z)

    Manz AG (M5Z)

    On 20241218, trading in Manz AG (M5Z) was suspended on Xetra for an upcoming announcement. This turned out to be that…

  • Pie In The Sky

    Pie In The Sky

    Imagine two markets: In the first market, all trades mark out flat. In the second one, half the trades mark out +1, the…

    6 条评论
  • Go Big or Go Home - 2024Q4

    Go Big or Go Home - 2024Q4

    Table 1 is the quarterly round-up of the most noteworthy Xetra and Eurex trades of 2024Q4. Mostly on par with the…

    9 条评论
  • Latency State of Play on Xetra

    Latency State of Play on Xetra

    Since things have reached the physical limit on Eurex, at least to within the measurement uncertainty, it's time to…

  • This is Not Your Grandparents' Central Limit Order Book

    This is Not Your Grandparents' Central Limit Order Book

    The classical central limit order book (CLOB) is pretty straight forward: just a collection of bids and offers. Any…

    5 条评论
  • Price Formation In The Auction Call Phase

    Price Formation In The Auction Call Phase

    During continuous trading, Xetra disseminates full order book data (level 3). In the auction call phase, however, no…

    1 条评论

社区洞察

其他会员也浏览了