Insights into the World of Algorithmic Trading

Last Updated on December 4, 2019 by Mark Ursell

Trading into the future

Algorithmic trading is more popular than ever. No longer the preserve of Wall Street and hedge funds, it is moving into the mainstream. Driven by increasing processing power and worldwide connectivity, algorithmic trading systems are more widely available than ever.

I am fascinated by the idea of using technology and quantitative approaches to improve my back-tests models and my trading performance.

I discussed algorithmic trading with Matt Goss. Matt has been involved in the financial markets for many years working as a member and floor trader on both the Comex and Nymex Exchanges and also for Flextrade. Matt’s latest project is StrategyDB, a platform providing technical strategy performance evaluation. The platform delivers reports and alerts in real-time.

This interview was a great opportunity for me to question someone who has first-hand experience of how the financial industry is developing and using algos.

In this wide-ranging interview we discuss execution and alpha-generating algos. High Frequency Trading (HFT) and how the ongoing rise of algos will affect the financial markets in the future. Matt also provides some great advice for people developing their own trading signals.

Can you tell me what interests you particularly about algos in the financial markets?

Yes, the ability for the machine computer code, algo intelligence, to “cover” and process a great deal of information more than I would be able to cover myself.

Great. Could you describe to me the main types of algos being used today in the financial markets.

Yes, the main types of algorithms these days are mainly execution algorithms.

Can you tell me why are people in the financial markets using execution algos [to get an advantage]?

Yes, that is the main reason. Execution has many pitfalls, and execution algorithms are more intelligent, some very complex, rather than just buy at market or sell at market etc., the basic execution types.

Execution risk can be measured in different ways including Volume Weighted Average Price (VWAP) and implementation shortfall. VWAP was introduced by JP Bialkowski to define a way of tracking the actual price of a transaction compared to the ideal price. Execution algos using VWAP and implementation shortfall processes are designed to ensure that the best possible price is achieved.

In your opinion are they using them to get ahead of the retail traders or is it an arms race with the other brokers?

Initially, the large players never care about retail, it is not a game to beat retail. It is only to get the best fills for their purpose, and to beat other peers by generating higher yields, via lower execution costs.

Your previous work was on a project called FlexEdge. Could you give me a bit more information about this project and your involvement.

Yes, FlexTrade’s FlexEdge is a product that is still offered by FlexTrade, my former employer. I originally pitched the owner Vijay Kedia, by showing him my intraday trading signals, and how his clients, already using execution algos, could benefit from a trading signal that would inform or guide the end user either manually or programmatically.

Once he accepted that I actually had valid signal, and decided to bring me in, I was inserted onto the quant team, and had to perform approximately six more months of R&D surrounding my signals. After proof, I combined my intraday signals with those of my teammates, and helped to design the back end of the product, the GUI, and supporting materials

Trading signals in this sense means that we instruct, based on our own proprietary conditions, when to buy and or sell a name. The research and development surrounding these signals must ensure that these signals are real, not noise, and have value. That value is determined by a signal that produces a profit that ranges anywhere from 15 bps or more per trade, along with other metrics

OK. Changing the subject slightly. HFT hit the headlines and continues to generate a lot of comment. There are ongoing investigations by the Securities and Exchange Commission into HFT firms accusing them of market manipulation. For many people, HFT is synonymous with algos. What is your take on this?

Yes, HFT at its finest, meaning true HFT, low latency, high speed, millisecond or microsecond co-located trading, done by algos, may have either good intent or bad intent.
There are certainly malware HFT algos out there, but I assume that these are not the only players. For example, there are market-making HFTs that are supportive of the market, and not just front running other trades. The list of HFT algos is long and varied.

There is another type of algos, which is the alpha-generating type. Could you explain a bit more about them.

Yes, my algos, the ones I provide for the FlexEdge product, are alpha-generating algorithms.
Let us not overcomplicate, the term algorithm sounds more complex than condition. An alpha generating “algorithm” could be as simple as buying above a moving average or it could be as complex as someone providing an array of conditions via Mathlab using stochastic jump equations, and physics turned on finance

So alpha generation simply means that you are trying to provide a signal or guidance towards a purchase or sale that will generate yield above and beyond what is provided using other measures. These other measures might be buy and hold or a benchmark indicator. The simplicity or complexity of the guidance is usually unknown or black box.

Your present company is called StrategyDB. Could you give me a bit of background and explain what services StrategyDB provides.

Sure happy to. StrategyDB has the ability to be a retail product or an institutional level product, it depends on the view. Essentially, the idea here is that many end users do not have the proper skill set required when it comes to back-testing,

And the other idea is that if we back-test en masse. we can data mine the results, instead of manually coming up with a hypothesis, coding it, and then moving to R&D, a very lengthy process

So, we add functionality that saves back-tests (we do the work) and then we have tools that enable the user to qualify the results.

Our large universe of symbols and bar intervals and strategies ensures that some of our results will lead to further endeavors. One may pursue potentially profitable strategies much more quickly using our tools.

Ok thanks very much for that. Do you have any advice for people develop their own alpha generating algos/signal generators?

Yes, happy to share some ideas, and these dovetail with the StrategyDB concept.
Alpha generating alogos may be based on fundamental, technical, statistical or other – it does not matter. In my experience, the simple strategies that contain very few parameters and are NOT optimized at all, function the best in real time.

Also, don’t throw away any work, save both the good and the bad results. StrategyDB does this, we do not filter out bad results. Negative returns and poor backtest results offer information that may be reused, repurposed, monitored, or actually used in logic towards alpha generation.

That is very helpful. I was reading your article “How to Succeed at Everything” I really like this idea. How do you think this relates to signal generating/trading/creating alpha?

Being able to leverage the power of computers and today’s big data database power enables us to iterate through a massive number of possibilities and combinations for our algorithm results. My goal for StrategyDB is to become a search engine / big data hub for strategy results that can be built upon. Accumulate, discard very little, and locate successful results, then repeat.

I think machine learning code is becoming popular and heading in this direction.

OK, that’s interesting. Could you tell me some of the unexpected benefits or surprises that you have experienced through your involvement in the world of trading algorithms?

I am surprised at how little real-world benefit actually arises from lengthy periods of algorithm R&D and implementation.

In reality, StrategyDB speeds to market the ability for one to locate a good candidate and pursue it. In the real world, most researchers below academia or institutional, spend too many resources, including money and time, to arrive at results that are often worthless, that surprises me.

Finally, do you have any thoughts or recommendations for those traders or investors trying to understand algos not necessarily to use them but to understand how the financial markets are being changed by them [assuming they are being changed by them].

In order to understand algorithms and how they are morphing present-day financial markets, one needs to consider inflation. Over time these algorithms are not only increasing but they are also all starting to look alike.

in the future the market will be not represented well, the actual underlying fundamentals of the companies that people are trying to trade, but rather will represent only the best predictive algorithms, and in the end, the algorithms will become the market, and fundamentals may be discarded.