Trading Show Chicago 2012

1.
어떤 분이 Trading Show Chicago에 다녀왔나 봅니다. 관련된 이야기를 짧게 메일로 주셨네요. 국제적으로 보면 한국자본시장은 중심이 아닙니다. 따라서 자본시장IT도 중심일 수 없습니다. 나라별 특수성을 고려한다고 하더라도 월스트리트의 흐름을 뒷따라가는 수준입니다.

월스트리트의 IT는 여러 행사를 통해 생각과 정보를 교환하는 듯 합니다. 워낙 행사도 많지만 서로 소통을 통하여 표준을 만들어나가는 문화가 정착된 듯 합니다. Trading Show Chigaco을 주최하는 Terrapinn이 2012년 행사중 컨퍼런스에서 나온 이야기를 소개하고 있습니다. 각 주제중 눈에 들어온 부분만 소개합니다. 나머지는 연결한 자료를 읽어보시길 바랍니다.

천천히 읽어보니까 다루는 주제가 무척 방대합니다. 감독정책, 전략, 기술등 트레이딩에 영향을 주는 모든 요소를 다루고 있습니다. 읽으면서 새겨읽어할 부분은 빨간색으로 표시하였습니다.

Specifically, which languages are used for development? The answer to this question is heavy on C / C++ using gcc compilers v.4.6-4.8, “Java, C# may work now but not later.” A colleague warned against elaborate software engineering, which may incur more costs and open more opportunities for misuse. FPGAs are also referenced for their parallel opportunities and speed, but programming them is a little more difficult than other technologies. A colleague added, “Your trading platform is the truest expression of your algorithm model at the micro-structure level. Measure, measure, measure!”

Rounding out the session were a few questions from the audience, “What components add the most latency in your system?” They respond with networks, switches, and geo-arbitrage execution. Specifically, what part of the internal stack adds the most latency? “The core accessing data, which uses 1.5 ~ 2.0 microseconds to make decisions.”
Insights on low latency execution practices중에서

Ram Ahluwalia of Winged Foot Capital moderated a panel session on the next generation of trading routes. Mr. Ahluwalia asked the panel, “What is the return on investment for firms with execution time close to zero?” Peter Nabicht of Allston Trading responded, “People tend to concentrate on speed but possibly neglect bandwidth. … Writing a check to get the fastest speed is not a competitive strategy due to other firms having the same ability. There are false insecurities in being the fastest.” Scott Nichols of Rotella Capital stated, “Reliability is crucial in addition to speed. Being able to quantify speeds throughout is a necessity. There is a constant need to refresh the technology to keep up pace but there should still be concentration on the strategies involved.” Marty Snyder of Communication Infrastructure Corp. presented a comment with respect to technology, “Microwave technology presents 33% speedup when the distances are longer. Also, cost to build microwave is significantly less than fiber optic networks.”
Staying inline with the topic, Mr. Ahluwalia asked, “Where are we on the race to zero?” Charles Dilts of Dell responded, “Is it worthwhile to go after that? Industry should try to generalize a way to quantize latency to understand what the speed needs are beforehand.” Mr. Nabicht took a concrete stance on the matter, “The race to zero is done … Constraints are going to be the variability of the exchanges if they can’t respond fast enough.”

What else can firms do to be competitive? Mr. Nichols stated, “Execution algorithms still need to be used efficiently and correctly. Forecasting the technology that will be the next trading technology in the race to zero may introduce skepticism.” Mr. Nabicht believes quantitative strategies are not easily bought while speed strategies can be developed as long as there are resources to purchase the technology, “There needs to be a threshold for speed and an infinite push for smarts.”
Discovering the next generation of fastest and most efficient trading routes중에서

Peter van Kleef of Lakeview Capital Market Services Gmbh has spent approximately 20 years doing electronic trading. To understand where the edge is coming from, Mr. van Kleef addressed seven major points; market data, market access, development, testing, deployment, setup, and being on top of the latest developments.(중략)
Testing has its advantages, but also its constraints. For instance, back testing is good for finding errors but not good for predicting performance. Also, testing requires much effort to rule out potential errors. Live simulations are good for developing intuition, but it is not a reliable source for having confidence in your strategies.
HFT Arms Race ? Developing technology to match your strategy중에서

What is FPGA? The acronym stands for “field programmable gate array”. This technology is considered to be field programmable, which means that the circuit network is modifiable by developers post-manufacturing. A hardware description language (HDL) is necessary for the modifications and popular choices are Verilog and VHDL. How does FPGA work? Using wires in combination with basic arithmetic logic operations to create more complex logic. FPGA innovation can be enhanced using IP cores, intellectual property cores which is a specific design of a set of instructions on the hardware that may be accessible, similar to accessing tools from header files in a program.

What sort of trading technology can we build with FPGAs? Mr. Flannery broke the question into two example applications. The first application is HFT logic for network communication. Minimizing network connection time is achieved by removing the switch so that the FPGA directly connects to your exchange, “Parameters are passed to the FPGA through the PCIe bus and a register file which can be challenging for large codes.” Mr. Flannery added, when building your FPGA system, avoid PCIe interrupt latencies and try to achieve single digit microsecond response time. For the second application, market data applications using FPGAs are examined. Relying on software to filter through data causes issues in two ways; pure software implementations have variable latency while latency will grow under loads and software will frequently drop packets. FPGA technology allows for more deterministic code to avoid some of the pitfalls of pure software implementations. Some examples of FPGA in financial applications are ETF computations, performing historical tick analysis, feed consolidations can be faster, and GUI applications can be developed on this technology as well.

“If it’s all about bypassing the PCIe bus latency, FPGA is not as promising but if you want to maximize wire speed, this is a good methodology.”
Boosting network performance with FPGA and potential alternatives중에서

Mr. Jessop’s presentation was broken up into two components. In the first part, he stresses the importance of knowing what data and models you occupy. With regards to data, Mr. Jessop referenced his analysts that provide recommendations and price targets. Mr. Jessop also alluded to his analysts using a technique entitled “Internal Alpha Preferences” whereby the traders pick the best and worst stocks and perform analysis on this list as well. Along with the analyst data and recommendations, there is also trading data which is fairly familiar to gather details concerning survey, volume, flow, price, volatility, etc.

The second part of the presentation was entitled “Understand your Forecasts.” Certainly it is easy for your machine to accept data and produce a value but what does that value represent in its entirety. More importantly, what are the forecasts capturing i.e., stock or factor information for instance? Mr. Jessop recommended taking the time to go through your model. For example, this approach allows you to think about how it works in the framework of breaking forecasts into stock specific and factor specific forecasts. A technique also recommended by Mr. Jessop is to process the signals before running your optimization i.e., regressing the signal into stock specific and factor specific models. Take out the factors and build a portfolio on the stock paths while getting rid of factor views.
Adding proprietary data to quant strategies중에서

Given that some members of the HFT community are on the “race to zero”, the next question was right in line with cost versus return when speeds are extremely fast. Mr. Dehghanpisheh asked, “Is the incremental cost worth it?” Mr. Goode answered, “There is a natural limit to about 50 ~ 100 ms, everyone will be there. Execution quality will be the divisor. Alpha driving will have speed as the base cost.” Mr. Kamarei’s response agreed with Mr.Goode’s and added some perspective, “As people’s speeds will converge, is the latency advantage an edge or a cost (like a tax) to participate?”

Speaking of advantages, the next inquiry posed was where to find balance in terms of what the shops are looking for in the smart vs. speed debate? Mr. Page responded “We are a small shop and have some budget challenges, we don’t need any more traders. … What we look for is a basic technical background, smart, flexible, and hardworking person.” Mr. Goode focused the response further by stating, “There is not an oversupply in technical skills … Typically, the best ones are well versed with Linux and Open Source.” To support this statement, there was some consensus throughout the conference that it is hard to find talent that has experience working with large data (graduating students should take note). Mr. Kamarei’s response provided a fresh perspective on talent acquisition. “In terms of finding talent, the market is better now than two years ago … finding new people is the challenge and we all have to contend against consulting, other financial firms, and now Internet companies like Google and Facebook.”
Examining the Latency Race, Data Management, and Connectivity중에서

The Flash Crash of 2010 has many theories which develop the reasons why the dynamics of the market acted out of expectation. Speed in this arena is vital and the race to zero has certainly begun. Mr. Hull’s experience certainly puts him at the forefront of understanding the relationship between speed, volume, strategy, and execution. From his perspective, “Strategy depends not exclusively on speed but rather heavily on queue positions. Getting to the front of the queue is a strategy.”

Along with recommendations for transactions, he also has some recommendations for those who want to manage their own quant / HFT fund. The numbers for technology, administration, and human capital average around 200k for a smaller firm. These costs can be mitigated if firms use a method of “co-operation” which implies sharing resources among the entities that do not inhibit either of their execution abilities. In terms of human capital components Mr. Hull suggests three skills that are required; a developer, a financial engineer, and a trader. Anyone person can have all three skills but that would limit their specific strength if they had to rely on themselves for all related tasks.
Pioneer Perspectives중에서

Post discussing the regulations, the next issue for the panel related to the Quant-agion effect. Mr. Kaul’s perspective implies that the hypothesis behind quant models needs to be developed. If someone has model insight, then the foundations as to why the model will work need to be readily explained. Quant-imentality relates to the human vs. machine issue and this is the next discussion. Mr. Tai suggests, “When push comes to shove human judgment will give you the decision making edge.” This position is affirmed by Mr. Koutoulas, “Machines are still programmed by humans.” Dr. Andre responds, “Looking at the variables are still machine learning responsibilities. Humans will be the data scientists and the data repair men. Feature states, negotiations, etc. are still very much a human activity. Humans are still needed on the front end i.e., for garbage in-garbage out.
Future Trends for the Modern Trader중에서

There are some advantages of having market segmentation. For one, competitive pricing becomes necessary. Secondly, the reaches from segmented but still connected markets make way for greater efficiency and innovation. Arbitrage between different trading venues is practically instant and with more opportunities to list shares, arbitrage opportunities increase as well. All things considered, there are disadvantages as well. An obvious issue is the redundancies that accumulate from the dual listings and the increased search cost associated therein.
Market Fragmentation: Managing cross-market trading risks중에서

Incorporating HFT to the underlying options creates additional pressure which then needs to be modeled. With respect to modeling, as is in the case with most markets, the issue of big data seeps through. How much data is necessary to interpret a strike by strike volatility skew?

Obviously, this raises questions and opportunities for HFT in the options market. From the presentation, Mr. Balasanov suggests quantitative traders have an opportunity to discover and trade new features of the microstructure of volatility that become significant only in HFT. Also, simultaneous volume and price discovery may not be found with the current financial models, implying they may have to imbed HF effects into the models to understand current risk and prices. This is a wide open field which seems as through insight from seasoned professionals to participants in relative technology are necessary to revisit a sense of formalism to the financial markets.
Using HFT to take advantage of the options market중에서

Karim Taleb of Robust Methods responded, “I do believe technology has been evolving faster than our ability to establish laws. … It’s time to step back and look at it from a fair distribution of flow and value”. Joe Gawronski of Rosenblatt Securities states, “Market structure is far from perfect but the SEC is working on them. By any metric you look at, the market has improved due to intense competition.” In terms of looking at fairness across the globe, Mehmet Yanilmaz of Myra Trading suggests that key issues in the debate are incentivizing order flows for traders and understanding what the incentives are for order cancelation flows. Mr. Yanilmaz then goes to quote, “in the US there are a few lessons we can learn from EU and Austrialia … but at the same time those markets are not as fragmented as the US’s market”.
Fairness of outcomes vs. opportunities중에서

Ugur Arslan of AienTech provides some background on how AI is utilized at his firm and why it is a good idea. To deal with market fragmentation, strategies need to “auto-correct” themselves to deal with situations that the human element would potentially not respond fast enough to. Since there are differing bank platforms, ECNs, retail brokers, institutional brokers, etc., these all present liquidity effects. The challenge is then how to assess the utility of a strategy for all these variations in the transaction. Should the human trader be responsible for managing the effects of these differences or should the human element respond to innovations in strategies and let the machine keep track of how to respond to these variations? It is believed we are moving towards a time when the separation of tasks between the human user and the algorithm will widen. Algorithmic AI will potentially take on more responsibility as these methods keep generating alpha.
Alpha vs. Arbitrage ? Real world applications of AI in HFT중에서

As the race to zero continues, there are inquiries from the trading world as to the balance of how smart do the strategies need to be versus how fast is really competitive. Some sides of the coin prefer to have balance between the two worlds. Others depend on quantitative analysis to further their competitive edge while some traders believe incremental speed differentials can really make the difference. To address these questions about Smarter vs. Faster, our third session of the day groups practitioners from big banks to proprietary trading firms to get a broad perspective on this question.(중략)

Lastly, the question of balancing the quants and the speed relationship is addressed by Mr. Young. “Strategies that worked before can work today but the issue is how to implement these strategies robustly.”
Smarter vs. Faster ? How to improve execution, risk, control procedures중에서

2.
발언한 내용중 일부만 요약하게 쓴 기사라 맥락을 이해할 수 없는 한계가 있습니다. 혹 ppt가 있으면 좋겠지만 이런 행사때 발표하는 자료는 한국과 다릅니다. 큰 제목만 몇 개 쓴 자료가 많습니다.(^^) 각자의 경험으로 이해하고 자료를 찾아서 더 연구해보셔야 합니다.

소개한 기사중 가장 인상에 남는 귀절이 있습니다.

“Strategy depends not exclusively on speed but rather heavily on queue positions. Getting to the front of the queue is a strategy.”

한국의 상황에 딱 맞는 말입니다. 대부분 호가주문을 매매체결큐에 1순위로 올리기 위해 속도를 요구합니다. 그런데 이 말은 반대입니다. “1순위로 올릴 수 있는 전략을 찾아라”라고 주장합니다.

3.
여러 행사중 시장의 공정성을 다루는 세션에서 다음과 같은 발언이 있다고 합니다.

“One of the challenges is overall different professional ethics. Traders have exchange rules, computer engineers have IEEE ethics, and quants have ethics superseded by mathematical truth. Can we derive some ethic that crosses traditional boundaries?”

IEEE 윤리강령이 있다는 걸 처음 알았습니다. 전기전자공학이면 컴퓨터공학이나 소프트웨어공학도 포함되리라 생각합니다. 소프트웨어개발자들은 알고 있을지?(^^)

IEEE 윤리강령 (원문)

(미국 전기전자공학회의 윤리강령으로, ‘전 세계를 통하여 삶의 질에 영향을 미치는 우리의 공학 기술의 중요성을 인식하고, 우리의 직업, 조직의 구성원, 또 우리가 봉사하는 공동체에 대하여 개인적인 의무를 짊에 있어서, 최상의 윤리적, 직업적 행위를 수행할 것을 서약하고, 다음 사항에 동의한다는 강령)

1. 공공의 안전, 건강, 복지에 맞는 공학적 결정을 하는 것에 책임을 지고, 사람이나 환경에 위험을 줄 요인들을 즉각적으로 알린다.
2. 실제적이거나 예상되는 利害의 충돌은 가능한 한 언제든지 피하고, 그러한 충돌이 있을 시에는 당사자들에게 이를 알린다.
3. 유효한 자료에 근거하여 청구나 평가를 진술함에 있어서 정직하고 현실적이어야 한다.
4. 어떤 형태이든 뇌물을 거절한다.
5. 공학 기술과 그 적절한 응용, 그리고 잠정적 결과에 대한 이해를 향상시킨다.
6. 우리의 기술적 역량을 유지하고 향상시키며, 오직 수련이나 경험에 의해 자격이 부여될 때, 혹은 관련된 한계점을 완전히 드러낸 후에 타인을 위한 기술적 과제를 맡는다.
7. 기술적인 일에 대한 진솔한 비판을 추구하며, 수용하고 그리고 제공한다. 또 잘못을 범했을 시에는 이를 인정하고 바로 잡으며 다른 사람이 기여한 바를 올바르게 인정한다.
8. 인종, 종교, 성별, 장애, 나이, 출신국가에 관계없이 모든 사람을 공평하게 대한다.
9. 다른 사람의 신체, 재산, 명성이나 직업을 거짓되고 악의적 행동으로 상해(傷害)하는 것을 피한다.
10. 동료나 동업자가 직업적인 발전이 있도록 돕고, 이 윤리 강령을 준수하도록 조력한다.