The Artificial Intelligence Revolution
In a period of a massive use of algorithmic trading systems, ways to stand out from the crowd are rare. While some of the current methods fail at supporting the balance of the financial system, transparent artificial intelligence models could contribute to regain the trust in the financial industry.
During the global debt crisis, the headlines are all about bank scandals [JPMorgan sold most of "London Whale" position, How Barclays Made Money On LIBOR Manipulation , HSBC to apologize at U.S. Senate money laundering hearing], price drops, baffled analysts, and bankrupt countries [Spain to Accept Rescue From Europe for Its Ailing Banks].
As a result, only 25% of the US citizens currently trust the financial industry, marking a 46% drop since 2008 and making banking the most unpopular sector among all [Edelman Trust Barometer 2011]. The financial sector has a trust problem, and this does not seem to be an accident. Surprisingly, the recent headlines about spectacular gambling losses in major banks are just the tip of the iceberg. In fact, 84% of all US equity fund managers were not able to beat their benchmark index in 2011, which is another 20% more than in 2008 [S&P SPIVA Report 2011]. The question becomes: why would people give their hard-earned money to a fund manager that performs worse than the average market and even gets paid to do so?
Partly to blame for the collective underperformance would be the inadequacy of human analysts to manage their decision process during critical and emotionally driven times. As mentioned above, most mutual funds underperform their benchmark; but during emotional times, the losses are even more significant. A look at the historical performance of most large equity funds compared to their benchmark index, reveals that it significantly decreases during 2008 and the end of 2011. After all, fund managers are human, and all humans are subject to their emotions to some extent.
In an effort to limit human influences, algorithmic trading emerged in order to handle the enormous amounts of data to process without any emotional distraction. However, a closer look reveals that this is not entirely true in most cases. Algorithmic trading systems use mathematical models – for example a set of IF-THEN rules – that contain knowledge about certain factors that likely lead to price changes of equities. While the systems are indeed not subject to emotions when applying their models on the market – e.g. continuously checking the rules and acting accordingly –, the model itself contains human influences, as it was previously defined by a human analyst.
In the light of higher profits, trading became faster and faster over the last years. We are currently in an arms race in high frequency trading, resulting in response times in the dimension of milliseconds. So instead of making the models smarter, they were made faster. However, faster models do not only lead to potentially higher profits, but can also cause higher losses. The widely discussed flash crash of 2010, when the Dow Jones Index lost about 10% in only 8 minutes, is only one example of how these models put our markets at risk. While trading frequencies are about to reach their physical limits, we will eventually discover that high frequency trading is not the solution. The emotional influences are still there.
To sum up: human analysts, current algorithmic models and high frequency trading systems are all driven by emotions that make them vulnerable. The only way to wipe out emotional influences completely from the decision process is to make the machine learn its model by itself. And here is where artificial intelligence (AI) and machine learning comes to play. Trading systems using AI techniques are able to build their own models autonomously using only rational and objective measures.
AI covers several approaches, from complex ones such as neural networks to rather simple ones such as rule sets or decision trees. However, for a long while the term ‘artificial intelligence’ has been misguidedly used as a synonym for neural networks. This type of model is designed after the structure of the human brain, and the trading knowledge is kept inside a network of interconnected artificial neurons. Just like our brain, these networks tend to be very complex, containing probably ten thousands of neurons and millions of connections! Because of their complexity, neural networks are able to capture high-dimensional knowledge, but are everything but transparent.
The model works like a black box, a believe-it-or-not system; its decisions can neither be interpreted nor verified. Intervention becomes impossible, especially whenever a problem appears, be it a wrong trading decision or a simple software bug – you will never find out. And complex software systems are never free of bugs. In financial applications, missing transparency is dangerous. This is why neural networks have never been successful in the markets since they appeared in the 90’s. Most of today’s algorithmic trading systems stick to more classical approaches.
The answer to successful AI trading lies in the use of simpler and more transparent models. Due to their understandability, decision trees, for instance, have been proven successful in many other areas of application: diagnosis of myocardial infarctions, military planning, and detection of money launderers.
They may not have the sophisticated-sounding complexity of neural networks, but they are absolutely intuitive and hence can be interpreted easily by humans. If the system has a bug or produces a wrong prediction, the source of misconception can be identified, understood, and corrected. Intervention is possible anytime, but not required unless there is a problem. Additionally, since the model’s knowledge can be visualized, we can analyze it and might even discover new “human” knowledge. The black box turns into a white box.
This does not mean that complex models are all bad or simple models are all good. In fact, a perfectly designed neural network leads to the same accuracy as a perfectly designed decision tree. But in the world of predictions, partial knowledge and complex software systems, there is no such a thing as a perfect model. All we can do is trying to get closer to this optimum, and this can be done only with models we fully understand. Why did the digital era revolutionize our world? It uses simple 0s and 1s, which are comprehensible for everyone.
After all, being successful in financial markets means being ahead of others and improving the model continuously. Using transparent prediction models, improvements are just easier. Furthermore, not all the responsibility is blindly given to the system; humans stay in control to the extend they desire. By setting parameters, observing their impact, or simply experimenting with the model, humans can guide the system. In a market, where an estimated 70% of all transactions are triggered by machines, this is the way to stand out from the crowd and to be better than others.
Thrilled by the perspectives of machine learning successes, a handful of firms, among them PASS Consulting Group (PASS), are developing fully autonomous and transparent artificial intelligence trading systems. Even more interesting are the outstanding results of these methods. To prove its performance, PASS took the challenge to assess its AI trading system KAIROS in a one year live-trading test conducted by WifOR, an esteemed German economic research institute. Within March 2011 to March 2012, a period of high volatility and high risk, KAIROS managed a virtual portfolio in real-time. KAIROS achieved an annual performance of about 28%, beating its benchmark by about 22%.
These results show that even in times of high uncertainty and perplexity, innovative proposals and solutions exist. Instead of sticking to paradigms that have served their time, the financial industry should now break the mold, including the implementation of more transparent, dynamic and adaptable trading models. After all, the time is right for an artificial intelligence revolution.
By,Timo Bozsolik is a Senior Innovation Consultant working at the Research & Development Department of the PASS Consulting Group. As an Innovation Consultant at PASS, Timo is in charge of the design and implementation of prototypes related to upcoming trends in the financial industry. He is the head architect of the KAIROS project, an artificial intelligence and machine learning based trading system. for more detail on algorithmic trading system by PASS Consulting Group.
DON’T MISS: Hit List
- How Italy and Spain helped calm investor nerves in Europe- Stock markets suffered their worst?
Last week when stock markets suffered their worst week, Candian press reports, A pledge from the European Central Bank to support the shaky bonds of Italy and Spain helped calm investor nerves in Euro…
- Exclusive Analysis Eurozone: Eurozone crisis is no longer simply a sovereign debt problem, but also a major systemic threat to the global banking system
As in the 2008 banking crisis, interbank lending has started to decline as banks are increasingly concerned about the Eurozone sovereign debt held by their peers. Interbank lending is likely to dec…
- France rating moved closer to perilous levels – slipping into recession in 2012
As the euro zone debt crisis has continued to deepen, France’s AAA rating moved closer to perilous levels, It’s credit outlook was lowered by Fitch Ratings, placed Spain, Italy, Belgium, Slovenia, Ire…
- What is debt ceiling or strategically default, should we raise it?
Debt ceiling or strategically default is one of the hottest topics these days. PEW Charitable Trusts prepared a great document that explains debt ceiling or strategically default, should we raise it?
- Should There Be Some Increase in Taxes to Reduce the Deficit?
Standard and Poor’s downgrading of the U.S. credit rating on Friday. Gallup POLLS RESULTS WHEN, asked about potential approaches the subcommittee may consider to reduce federal debt, a majority of Ame…
This post was written by a guest contributor. Please see their details in the post above. If you'd like to guest post for CEOWORLD Magazine check out our "Become a Contributor" page for details OR send an email to email@example.com