Roy Niederhoffer (pictured) graduated magna cum laude from Harvard with a degree in Computational Neuroscience in 1987 and has seen a lot of quantitative hedge funds come and go since 1993, when he established RG Niederhoffer Capital Management Inc, a quantitative trading advisor that employs a short-term contrarian investment strategy taking its inspiration from the field in which Niederhoffer studied.
Niederhoffer will be sitting on the panel Artificial Intelligence – should we unplug man from the machine? at next month’s Amsterdam Investor Forum. In his view, whilst it is clear that in some domains machine learning and artificial intelligence is starting to make a big difference, the key is understanding which domains are appropriate and which domains are potentially problematic. Some domains, like object and speech recognition, linguistic analysis, and credit analysis are perfect for machine learning and particularly, deep learning algorithms. But in Niederhoffer’s experience, making short term market predictions using machine learning is perilous, though possible.
“We began exploring neural networks in the early 1990s, and we’ve continued to make forays into these areas over the years. We are a heavy user of one particular machine learning technique since the mid-2000s and another, which we are very excited about, entered our research program about two years ago. By now, it is becoming a significant piece of our investment strategy. We are using various technologies but there are always caveats in terms of where you can use them and how you can use them. You have to be extremely careful. Machine learning is not the Holy Grail in terms of trying to forecast the direction of price in the markets,” explains Niederhoffer.
Given its automated, quantitative investment strategy, the firm has a natural affinity toward advanced computing techniques. Broadly speaking, the RG Niederhoffer flagship Diversified Program, is designed to do its best during periods of stress, volatility and emotional arousal such as equity market declines, rising interest rate periods, and moments of illiquidity. These are the conditions, says Niederhoffer, in which market participant (both discretionary and systematic) are most susceptible to behavioral biases, markets become more predictable, and the RG Niederhoffer Diversified Program has succeeded.
“Since we started back in 1993, we have tried to avoid the common path of being long equities, long fixed income, short volatility, all three of which have been fantastic for the last eight years. Our strategy has always been designed to work in harmony with portfolios that already contain those exposures,” says Niederhoffer. Because of this, the firm tends to take non-consensus, contrarian views more often than not.
When thinking about the sophistication of quantitative funds over the last decade, it is fanciful to think that we could set the computers free and let them trade the markets without any human intervention.
The thing is, computers are very good at figuring out what works most of the time.
For example, it is easy to find strategies that appear to predict bull markets effectively based on bullish sentiment. This can, however, easily lead to machine learning models producing spurious predictors of future market direction. One only has to point to the market crashes of October 2000 and October 1987 to underscore this point.
What this has led to at RG Niederhoffer is a conviction that one needs to know in advance what the key independent variables are, to put into the computer model, rather than letting the computers identify the variables.
“Our experience is that computers aren’t best at identifying predictive variables. Our research has shown that it is more effective to screen variables, using prior knowledge, before you begin a machine learning process,” comments Niederhoffer.
A large price movement can be caused by one event, representing one observation of many. A computer algorithm that attempts, for example, to find the “reason” in the data that the Brexit vote won through by 52 per cent to 48 per cent may end up barking up the wrong tree, so to speak, and find a spurious variable that “explains” a completely exogenous event.
Niederhoffer’s cautious utilization of machine learning techniques does not mean that the firm uses human discretion in running its investment strategy. Theirs is a fully systematic quantitative approach that uses human intelligence “to guide our machine learning strategy, rather than setting the computers free on the data with no constraints.”
Recently, Elon Musk spoke of his concerns over the machines taking over in a film by Werner Herzog called Lo and Behold, Reveries of the Connected World (2016). He gives an example of what could happen were a hedge fund to leave it up to AI to maximize the returns of a portfolio. The AI system might determine that the best way to do that would be to short consumer stocks, go long defensive stocks and start a war.
This is just one example of where AI could create inadvertent crises, whether planned or otherwise, if unplugged entirely from man.
“It is difficult to predict what the power and influence of computer algorithms could be over the next 30 to 50 years,” states Niederhoffer. “We are only just starting to scratch the surface. Take the current issue of fake news and how algorithms dictate search results and the stories one is shown. We are only beginning to look at the impact of AI on our access to information.”
“One technology that we are particularly excited about is virtual reality. VR, when combined with AI and realistic simulations of humans, is going to create a new way of experiencing the world.”
In terms of navigating global markets, 2016 was unique in that it was both tumultuous and at the same time surprisingly tranquil. The three-month period leading up to the US Presidential Election, for example, was one of the least volatile periods in the S&P 500’s history.
“We do not take the view that markets’ tranquil state last summer or early this year is predictive of what will happen for the remainder of 2017 and beyond. As a short-term systematic quantitative strategy, we refrain from predicting catalysts. We believe that there are many ways for volatility to rapidly return to markets creating challenges for traditional and alternative portfolios. We intend to be there for our clients when volatility returns,” concludes Niederhoffer.
For now, as long as human intelligence remains vital to identifying key indicators of future market moves, computers will be kept on a leash.
