These days, high-frequency trading (HFT) is a controversial topic in economic discussions. Critics bemoan the proliferation of high-speed trading. They claim it destabilizes global financial systems, while proponents say that HFT is a natural progression in the development of new financial-trading tools.
As algorithmic trading continues to advance, the line between human trading and machine trading is becoming blurred. It’s difficult to guess the upper limits for speed in trading. Yet, the next wave of evolution in technology for trading is likely to be the expanded use of artificial intelligence (AI) to build trading systems which trade smarter instead of merely faster.
If managed correctly and built with transparent logic, this new generation of intelligent trading machines can help counterbalance the negative effects of black-box, algorithmic HFT tools. Not only would it help restore public trust in the financial-trading sector, it might also help level the playing field for all traders.
In recent years, there has been a common perception that large traders using sophisticated high-speed trading platforms have brought world economies close to disaster on at least several occasions – Witness the uproars about the “London Whale,” LIBOR manipulations, the Barings Bank collapse, and other situations in which a handful of traders used powerful technologies to wreak financial havoc, even if short-lived.
Faster trading is only better when it’s less emotional
Perhaps greater use of “machine intelligence” will give traders and the public more confidence in the rationality of trading decisions. At the very least, machine trading should be expected to outperform human trading during times when human emotions overtake rationality and create chaos in the markets.
AI technologies serve two primary purposes within the trading world: To increase the speed of trading, and to handle data for purposes of logical predictions. Thus far, most traders have been focused on the speediness of HFT instead of looking toward the predictive capability of AI technologies.
Algorithmic trading arose to meet traders’ need to aggregate and process large amounts of market data. The earliest generation of algorithmic trading tools merely used mathematical “if-then” rules to trigger machine actions.
Even though most current HFT systems use machine logic for rapid data assessment and trade execution, these systems are still bound by human rules — They require humans to input the rules about expected price movements upon which the machine relies.
Yet, the newer generations of AI-based trading technology promise to help traders find and execute trades without being hindered by emotional baggage.
What’s the alternative to HFT?
Logically, the next step in the development of trading tools is for machines to assume even more of the overall decision-making process, including independent learning and intelligent growth in trading ability. Given the rapid pace of development of AI, and its convergence with other financial technology developments, this seems increasingly likely to occur soon.
Without true machine intelligence, HFT alone is simply an “arms race” toward faster executions based on the same data-crunching capabilities. Up to this point, trading technologists’ strategy has been to make trading faster. Now, the time has come to employ more artificial intelligence to make trading smarter as well as faster.
If our trading systems were both smart and fast, perhaps we could soothe the critics of HFT, who point to “flash crashes” as a warning that faster trading may lead to faster losses during crises. Faster trading alone cannot solve such problems because HFT still based on human emotions.
A smart trading system that builds itself, and learns by experience
The deployment of artificial intelligence and machine learning in trading will make a difference because these new types of trading systems can be built autonomously and tested using their own purely-rational standards – Human emotion has no place in the equation.
Regarding the deployment of AI and machine-learning technologies for trading, it’s important to make a distinction between the older concept of “neural networks” and the more-current use of decision trees for such technologies. Neural networks are complex, and they aren’t transparent.
In contrast, AI based on transparent decision trees has already proven effective in healthcare, weather prediction, financial modeling, and certain other applications. This transparency may help ease concerns by the lay public as well as by government authorities who are wary of the powerful effects of “black box” trading systems.
Artificial intelligence and machine learning are the next frontier in trading
Artificial intelligence may be defined as a system which is perceptive about its environment and takes action, including learning, in order to maximize the chance of success in that environment.
The earliest uses of AI were focused on solving specific problems. Trading is an appropriate application, since AI features include statistical and computational abilities, logic, and economic principles.
AI includes machine learning, which seems the most fertile ground for improving trading beyond the current evolutionary stage of HFT. Machine learning lets the computer grow as a trader, yet without suffering from the emotions of a human trader.
Stated simply, machine learning involves the development of algorithms which improve themselves autonomously and automatically through experience with market events. Machine learning may be either supervised, or unsupervised.
Supervised machine learning means the machine can classify data according to its type, and learn to recognize types after seeing several examples. Regression analysis is then applied in order to create mathematical functions which predict how the trading response should change when the input data changes. Current trading technology is already well established at this supervised level.
Faster and smarter
Unsupervised learning means the trading machine can find patterns in a stream of data, and adjust its actions autonomously to deal with those patterns. Of course, even though emotion isn’t present in AI itself, the successful trading machine must still account for emotion in order to predict the actions of human traders in the marketplace.
Going forward, the deployment of fully-autonomous, ever-learning trading systems may lead to a marketplace in which trading is truly “mechanical” as well as speedy. Ultimately, creativity in trading will likely become a hallmark of good machine trading systems.
Absent human emotion and delay, this improved future marketplace based on faster and smarter trading should be more stable, and therefore ease the current worries about HFT.