By Christian Zimmer, Hellinton Hatsuo Takada, Ji Kong Chan
Christian Zimmer, Head of Quantitative Trading and Research, Itaú Asset Management and co-author of ‘High Frequency Trading in Brazil: Mirage or Miracle?’, originally printed in the June issue of FIXGlobalTrading, responds to reader questions about co-location and exchange pricing models.
Ji Kong Chan, Sell-side Quant/HFT DMA, Tokyo:
I shall approach my follow-up comments, focusing on two key words: 1) “increment in HFT trading volume” and 2) “cost reduction program does not exist”
1) “increment in HFT trading volume”
Tokyo Stock Exchange (TSE) implemented Arrowhead (for equities) in January 2010. Over the past 18 months, the percentage of co-located trades (compared to all trades through the TSE) have almost tripled. For ease of categorization, I treat high frequency trades as mostly constituted of trades coming from co-location, which I argue is close to the truth. I hope the majority will agree that there is not a clear-cut definition of HFT, let alone a viable methodology for segregating HFT and non-HFT trades. This is probably why I have not encountered any reliable statistics comparing how much HFT was being done before and after Arrowhead.
2) “cost reduction program does not exist”
All else being equal, the fixed costs borne by brokers and payable to the TSE are arguably higher post-Arrowhead. That said, the variable costs take on more significance.
Christian Zimmer:
I totally agree with Mr. Chan that the use of co-location trades as a proxy for high frequency trades is the best available approach. BM&FBOVESPA uses form of intraday trade frequency, but this measurement is open to many criticisms. On one hand, the identifying the type of trade relies on identifying the final investor. Using this method of identification has two major problems. First, no one else has this information, and thus, the statistics cannot be replicated. Second, many big firms identify their trades under an omnibus account and this must be properly treated.
Having acknowledged these issues, it is still not clear what threshold is used for high frequency trades and why exactly this level is used for identifying the trading as high frequency. Bearing in mind that in most cases the nature of the HFT strategies requires low latency, it seems to be the best available alternative to use the co-location numbers. But what about proximity hosting? Especially in the Brazilian market where equity and future markets’ co-location facilities are separated and not allowed to communicate, you need proximity for strategies like index arbitrage. Further, some players, like Itaú Asset have fast direct lines (DMA3) that come close to proximity solutions and can be fast enough for most of the strategies that use short-term directional exposures.
As HFT is normally VERY cost sensitive, the additional costs for co-location are important. It is mainly a fixed cost, but when it is paid via soft dollars by the broker, the variable cost per trade increases. If the HFT pays the co-location costs, it is just a simple break-even analysis. Can you make a sufficient amount of trades that will allow you to pay the costs?
Finally, we do not have the number of international players split up by HFT and non-HFT (using the exchange’s nomenclature).
Click here to read the original article, High Frequency Trading in Brazil: Mirage or Miracle?.
