The Big Problems with Machine Learning Algorithms in Finance

The Big Problems with Machine Learning Algorithms in Finance
Share this:

Machine learning is allowing investors to leverage massive amounts of data sets including social media postings in a manner that no human being can. Despite this technology’s huge potential, its record is still mixed.

To put this into perspective, the Eurekahedge AI Hedge Fund Index, which is known for tracking the returns of about 13 hedge funds that utilize machine learning, has attained only 7% per year for the last five years.

Also, the S&P 500 managed returns of 13% per year. This year alone, the Eurekahedge benchmark experienced a 5% decline through September.

According to Marcos Lopez de Prado, the main challenge affecting machine learning strategies nowadays is the considerable low signal-to-noise ratio, particularly in financial markets.

He became part of AQR Capital Management in September by assuming the role of head of machine learning and is known for being the author behind the 2018 book dubbed Advances in Financial Machine Learning. “Machine learning algorithms will always identify a pattern, even if there is none. In other words, the algorithms can view flukes as patterns and hence are likely to identify false strategies. It takes a deep knowledge of the markets to apply machine learning successfully to financial series,” said López de Prado.

Nigol Koulajian also shares the same sentiments as Lopez de Prado. The CIO and founder of New York-based Quest Partners, a systematic macro hedge fund that is involved in the managing of $1.7 billion said that quants being drawn from finance programs and high-technology companies mostly anticipate to institute optimizations at a higher level of accuracy compared to what is allowed in the finance world.

“They’re coming with a mindset that we’re going to conquer the world with big data. In finance, though, the market regime is not static, and markets aren’t closed systems like a chess game. You can have one little pin drop that can basically make you lose over 20 years of returns,” said Koulajian.

Bottom-fishing equity indexes serve as one risk-on technique that has worked perfectly in the past decade, particularly since the financial meltdown.

Koulajian said: “Everyone’s buying the dips. There are all these people who have learned to basically suppress the vol or volatility.” If you utilize machine learning, you can apply dozens of versions of this approach.

According to Koulajain, the risk is that the persistent bull market that facilitated such strategies to work effectively was powered by central bank liquidity, and is currently being dragged away.

He also said that skew, a tail risk measure, is showing signs of the S&P 500 dropping by 30 percent.

Back in August, the CBOE Skew Index, which aids in tracking out-of-the-money index options, hit a record high. Koulajian said that in case you are purchasing the dips with machine learning, it is easy to applaud yourself on utilizing a more advanced model optimization as well as lose focus of the bigger risks.

According to Robert Frey, machine learning is not a new concept. In the late 1980s, he began a hedge fund, which was later absorbed into Renaissance Technologies a few years later.

As part of Renaissance, it emerged as the driving force behind the statistical arbitrage approach in the massively successful Medallion Fund. “You hear all this stuff about machine learning and AI.

Most of those techniques, however, have been around for decades—and we, in fact, used a lot of them at Renaissance.

The fundamental processes that we’re talking about here are a combination of advanced statistics—computationally intensive statistical analysis—and then the neural-network-type branch where you’re looking at these models, which are basically classifiers,” said Frey.

Upon retiring from Renaissance back in 2014, Frey began the quantitative finance program, specifically at Stony Brook University before opening a family office that later became FQS Capital Partners LP.

At the firm, Frey utilizes machine learning methods to assess hedge funds, particularly those concentrating on modeling their underlying behavior as well as tying it back to economic trends and systematic economic and market trends.

Stephen Cucchiaro represents another well-known user of machine learning approaches. The CIO and president of 3 Edge Asset Management LP, which is an exchange-traded fund strategist situated in Boston.

He said: “No scientist or engineer would ever use correlation analysis for analyzing a nonlinear complex system of interrelated variables, because those correlations are often nothing more than statistical coincidences, not true cause-and-effect relationships.

3Edge’s model is based on causal factors that are regime-dependent and nonlinear, and the firm uses algorithms derived from AI to search for optimizations that also take into account key portfolio characteristics.

Our end solution is very different from more traditional AI approaches since we end up with a model that provides us with not only a prediction of market behavior but also an explanation of why.”

The CEO of Periscope Capital Inc. Jamie Wise said that there is a common idea out there that artificial intelligence (AI) is all-knowing and that it would phase out individuals from their jobs.

However, after creating a technique that leverages the use of machine learning and neural networks, he views it in largely more ordinary terms. “It’s really just a tool—and it’s a very task-specific tool, too,” he said.

Wise got into this field nearly six years ago when he looked at how people around the world were posting about hotels and restaurants on social media. He observed that they were likely to begin conversing about investment and stock portfolios the same way.

In turn, Wise asked himself how his firm could take advantage of that. “The exciting concept for us was this idea of measuring sentiment directly at the individual stock level. The sentiment is obviously a key driver of stock prices, yet it’s typically gauged indirectly through proxies such as put-call ratios, inferred from a comment by someone on TV, or tracked with a lag in surveys. Intuitively, sentiment has these predictive measures, but we could never measure it for stocks,” said Wise.

With that, Wise’s company started researching stock chat rooms. First, he said that they discovered a lot of what you may expect to find: stock promoters advertising thinly traded, pump-and-dump candidates.

Nonetheless, the conversation began to grow. “StockTwits was a really big part of that,” Wise says, referring to the social network for investors and traders that started in 2008. “So was Twitter.”

Back in 2012, Twitter added cashtags—$ that was followed by a ticker symbol— for tagging tweets associated with a given company. This situation triggered more people to post about their stocks on the social media platform. “And the bigger the stock, the more likely it was that people were talking about it,” said Wise.

To create an investment approach based around that particular conversation, Periscope recruited three individuals with backgrounds in both machine learning and natural language.

It was vital to start building from scratch, since models that are trained on hotel reviews, for instance, would not know would lack a clue about what to make out of a comment such as” XYZ going to 50.” Alone, such a statement lacks a clear negative or positive meaning. “But you and I know that that can be clearly positive or clearly negative depending on where the stock was when that person said that,” said Wise. “Then it’s really easy.”

The company’s neural network was, in turn, trained to make sure that when a comment, blog posting, or tweet went in, a sentiment score would be generated.

However, after the model was finally running, the question about how to create a portfolio that was based on that emerged.

Contrary to the sentiment-based approaches that aim to jump in based on, let’s say, a rise in social media postings Periscope assumed a longer view.

The initial criterion was that a stock required a certain scope of conversation in a bid to qualify for the portfolio. “The way I think of sentiment is like an ocean,” he said. After attaining a massive volume of posts on a particular stock regularly, there can be a flow and ebb, mainly to the aggregate sentiment they express.

“If you can identify some kind of wave building in the ocean,” said Wise, then it should carry forward for some amount of time. “But it won’t be hours—it will probably be weeks, maybe even months.”

The aim of this strategy for Wise is to apply it in a hedge fund, which launched in May 2018.

However, along the way, he also created an ETF that was based on the particular strategy, as proof of concept. Since inception back in 2016, the $11 million BUZZ US Sentiment Leaders ETF is known for an average of 26% returns. In fact, this October 5, 2018, it attained 25%.

Wise said: “machine learning algorithms can model complex data structures much better than Markowitz-style solutions—and yield superior performance out-of-sample.” This statement was in reference to the mean-variance portfolio construction model that was created by Nobel laureate Harry Markowitz.

According to Lopez de Prado, machine learning can be utilized in detecting false strategies. He added: “Most empirical discoveries in finance are false, particularly when they lack economic intuition. The culprit is backtest overfitting. Machine learning can help determine the probability that an investment strategy is false.”

Source Bloomberg

Share this:

Leave a Reply

avatar
  Subscribe  
Notify of